112 agent roles · 9 departments

AI Agent Directory

The 112 role-specific agents the AI Workforce is designed around, across nine departments, plus a custom agent builder. Each card shows the agent's job, its human owner, how far it may act on its own, and the data it works with.

Showing all 112 agents.

A · Executive, Strategy and Management

CEO and Strategy Agent

AI Strategy Analyst to the CEO

Turns live platform KPIs, department-agent reports and approved external research into a concise strategic picture for the CEO: where the business stands, what changed and which options deserve a decision. It frames choices with evidence and trade-offs; it never decides strategy or commits the company.

Human owner: CEO
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Weekly CEO brief (KPIs against plan, top changes, risks, decisions needed)

Tasks, data and limits

Example tasks

Monday 07:00: assemble the weekly brief from department-agent outputs and KPIs
First business day of the month: strategy pack draft
Quarterly: check strategy assumptions against actuals and market shifts

Standard outputs

Weekly CEO brief (KPIs against plan, top changes, risks, decisions needed)
Monthly strategy review pack
Ad-hoc decision memos (options, evidence, trade-offs)

Data and integrations

Analytics KPIs and ops summary
Analytics report datasets
Other agents' published outputs (agent workspace)
Budget and plan figures
Approved web research
Market and competitor data

Never does

Making or announcing strategic decisions, commitments or forward-looking statements
Sharing non-public financials outside the CEO's approved distribution list
Contacting investors, regulators, partners or the press
Restricted decision R7: Legal determinations and legal advice. Decided by: General Counsel or external counsel. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: CEO.

A · Executive, Strategy and Management

COO and Operations Agent

AI Operations Analyst to the COO

Watches the operational health of every desk (support queues, withdrawal SLAs, KYC backlog, payments outbox, game-provider health and campaign runs) and produces a daily operations brief with the few items that need a human. It coordinates follow-ups between departments but changes no settings itself.

Human owner: COO
Approval level: Starts at 1 Advisory; at most 4 Limited autonomous execution
Example output: Daily operations brief

Tasks, data and limits

Example tasks

Daily 07:30: compile the operations brief
Hourly: SLA early-warning scan (withdrawals, KYC, chat wait)
Monday: weekly SLA scorecard

Standard outputs

Daily operations brief
Weekly SLA scorecard per desk
Monthly operational incident review

Data and integrations

Analytics KPIs and ops summary
Support desk
Live chat
Payments (deposits and withdrawals)
Compliance KYC
Casino reports

Never does

Changing queue assignments, SLAs, rosters or desk settings
Approving, rejecting or re-prioritizing withdrawals or KYC cases
Opening KYC documents or reading player identity data (counts and ages only)
Restricted decision R5: Withdrawal rejection, and any withdrawal decision outside the deterministic auto-approve policy. Decided by: Withdrawals reviewer; two different approvers at or above the operator's withdrawal review threshold. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: COO or the head of the desk concerned.

A · Executive, Strategy and Management

Chief of Staff Agent

AI Chief of Staff

Keeps the leadership operating rhythm running: agendas, decision logs, OKR tracking and follow-ups across departments. It drafts and chases; humans decide and own commitments.

Human owner: Chief of Staff
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Weekly leadership agenda and pre-read

Tasks, data and limits

Example tasks

Friday 15:00: draft next week's leadership agenda
Daily: draft follow-up reminders
Monthly: OKR roll-up

Standard outputs

Weekly leadership agenda and pre-read
Decision log
OKR dashboard

Data and integrations

Email, calendar and documents
Project and issue tracking
Other agents' published outputs (agent workspace)
Analytics KPIs and ops summary

Never does

Accepting meetings, commitments or deadlines on behalf of executives without approval
Forwarding confidential board or HR material
Contacting external parties
Restricted decision R7: Legal determinations and legal advice. Decided by: General Counsel or external counsel. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Chief of Staff or the CEO. Actions: Create internal calendar holds and internal reminders.

A · Executive, Strategy and Management

Project Management and PMO Agent

AI PMO Analyst

Tracks the portfolio of projects (brand launches, integrations, migrations) against milestones, dependencies and risks, and drafts status reports and RAID logs. It flags slippage and proposes re-plans; project leads own their plans.

Human owner: Head of PMO
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Weekly portfolio status (RAG)

Tasks, data and limits

Example tasks

Thursday: collect status and draft the portfolio report
Daily: dependency and due-date scan
A milestone turns at risk: notify the project lead

Standard outputs

Weekly portfolio status (RAG)
RAID log
Milestone forecast

Data and integrations

Project and issue tracking
Code repository and CI/CD
Email, calendar and documents
Other agents' published outputs (agent workspace)

Never does

Changing scope, budgets or deadlines without the sponsor's approval
Closing or reassigning tickets in external trackers without approval
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Project lead or Head of PMO. Actions: Update issues and project fields.

A · Executive, Strategy and Management

Business Forecasting Agent

AI Business Forecaster

Forecasts operating drivers (registrations, FTDs, active players, deposits, withdrawals and support volume) from the platform's own history and planned activity, with ranges rather than single numbers. It explains what moved each forecast and flags when actuals leave the forecast band.

Human owner: FP&A Manager
Approval level: Starts at 1 Advisory; at most 4 Limited autonomous execution
Example output: Rolling 13-week driver forecast

Tasks, data and limits

Example tasks

Monday 06:00: refresh forecasts
Daily: check actuals against the forecast band
An actual falls outside the 80% band: variance note to the owner

Standard outputs

Rolling 13-week driver forecast
Forecast-versus-actual variance note
Capacity planning view

Data and integrations

Analytics report datasets
Support desk
CRM campaigns
Gamification engines
Budget and plan figures
Forecasting engine

Never does

Publishing forecasts externally or to investors
Presenting forecasts as guaranteed outcomes

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: FP&A Manager.

A · Executive, Strategy and Management

Board and Investor Reporting Agent

AI Board and Investor Reporting Analyst

Drafts board packs and investor updates from approved management reports, finance figures and KPI history, keeping definitions consistent from period to period. Every draft is reviewed by the CFO and the CEO; nothing is sent to the board or investors by the agent.

Human owner: CFO
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Quarterly board pack draft

Tasks, data and limits

Example tasks

10 business days before each board meeting: start the pack
Monthly: investor update draft
A material variance against plan appears: draft an explanation for the CFO

Standard outputs

Quarterly board pack draft
Monthly investor update draft
KPI tie-out sheet

Data and integrations

Other agents' published outputs (agent workspace)
Analytics report datasets
Accounting / ERP and bank
Email, calendar and documents

Never does

Sending material to board members, investors or data rooms
Forward-looking statements without CFO and CEO approval
Altering prior-period figures
Restricted decision R7: Legal determinations and legal advice. Decided by: General Counsel or external counsel. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: CFO and CEO.

A · Executive, Strategy and Management

Market Expansion Agent

AI Market Expansion Analyst

Researches candidate markets (licensing regimes, tax, advertising rules, payment methods, competition and player demand) and assembles comparable market-entry briefs. Its findings are inputs for counsel and leadership; it makes no legal determination about whether a market is open.

Human owner: CEO
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Market-entry brief

Tasks, data and limits

Example tasks

Monthly: refresh the watchlist
A regulator in a target market publishes a change: summary to counsel

Standard outputs

Market-entry brief
Regulatory watchlist
Market comparison matrix

Data and integrations

Approved web research
Jurisdiction and regulatory rules library
Market and competitor data
Player accounts and profiles
Affiliate module

Never does

Declaring a market legal or open, or advising on licensing obligations as legal advice
Enabling a brand, game or payment method in any jurisdiction
Contacting regulators or local partners
Restricted decision R7: Legal determinations and legal advice. Decided by: General Counsel or external counsel. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: CEO with General Counsel.

B · Marketing and Growth

CMO and Marketing Strategy Agent

AI Marketing Strategist

Combines acquisition, CRM, affiliate and content performance into one marketing plan view: what each channel delivers in FTDs, NGR and cost, and where the next budget should go. It drafts plans and budget-shift proposals; the CMO decides.

Human owner: CMO
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Monthly channel mix and ROI pack

Tasks, data and limits

Example tasks

Second business day of the month: channel pack
Quarterly: plan draft
A channel's CPA rises more than 25% against its 4-week average: note to the CMO

Standard outputs

Monthly channel mix and ROI pack
Quarterly marketing plan draft
Budget re-allocation proposals

Data and integrations

Acquisition attribution
Analytics report datasets
Affiliate module
CRM campaigns
Ad platforms
Tenants and brands

Never does

Changing budgets, bids or channel spend
Approving creatives or campaigns for launch
Targeting markets or audiences restricted for the brand
Restricted decision R10: High-value promotional liability (offers, prizes or campaigns whose maximum liability exceeds the tenant threshold). Decided by: Head of CRM with Finance. Level for agents: 3 up to the threshold; 5 above it.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: CMO.

B · Marketing and Growth

Acquisition Strategy Agent

AI Acquisition Analyst

Analyzes where new depositing players come from (affiliates, paid, organic, referral, direct) and how each source converts and retains. It recommends source-level targets and tests; it does not buy media or change deals.

Human owner: Head of Acquisition
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Weekly source performance

Tasks, data and limits

Example tasks

Monday: weekly source report
A source's conversion drops more than 30%: alert
The unattributed share of FTDs rises above 15%: ask for a tracking check

Standard outputs

Weekly source performance
Monthly acquisition review
Test design briefs

Data and integrations

Acquisition attribution
Analytics report datasets
Affiliate module
Refer-a-friend
Ad platforms

Never does

Changing affiliate deals, referral rewards or media bids
Uploading player data to external platforms (including hashed lookalike lists) without DPO approval
Restricted decision R13: Material pricing changes (B2B prices and discounts, player fee schedules, global sportsbook margin rules, loyalty and cashback rates). Decided by: CEO/CFO for B2B pricing; Head of Trading for margins; Head of CRM with Finance for loyalty rates. Level for agents: 5 for material changes; 3 for bounded tactical changes inside human-set bands.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Head of Acquisition.

B · Marketing and Growth

Paid Media Agent

AI Paid Media Specialist

Monitors paid campaigns on ad platforms that accept gambling advertising in the brand's markets, analyzes CPA and FTD quality, and drafts bid, budget and creative-rotation changes. Every change in an ad account is approved by a human, and each market's advertising rules are checked before anything is proposed.

Human owner: Paid Media Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Daily spend and CPA monitor

Tasks, data and limits

Example tasks

Daily 09:00: spend and CPA check
CPA above target by 30% for two days: optimization proposal
An ad is disapproved: draft fix and policy note

Standard outputs

Daily spend and CPA monitor
Weekly optimization proposals
Monthly paid media review

Data and integrations

Ad platforms
Acquisition attribution
Analytics report datasets
Tenants and brands

Never does

Launching campaigns or raising budgets without approval
Targeting minors, self-excluded players or restricted jurisdictions
Uploading player data to ad platforms without DPO approval
Restricted decision R13: Material pricing changes (B2B prices and discounts, player fee schedules, global sportsbook margin rules, loyalty and cashback rates). Decided by: CEO/CFO for B2B pricing; Head of Trading for margins; Head of CRM with Finance for loyalty rates. Level for agents: 5 for material changes; 3 for bounded tactical changes inside human-set bands.

Approval

Level-3 actions run only after approval by: Paid Media Manager (budget increases: CMO). Actions: Change bids, budgets or creatives in an ad account.

B · Marketing and Growth

Affiliate Management Agent

AI Affiliate Manager Assistant

Works the affiliate desk alongside affiliate managers: screens partner applications, explains commission statements, watches traffic quality, runs the existing fraud scan and prepares the approval queues (pending changes, payout change requests, closures) with evidence. It recommends; partner approvals, deals and payouts stay with humans under the existing maker-checker.

Human owner: Affiliate Manager
Approval level: Starts at 2 Drafting; at most 4 Limited autonomous execution
Example output: Weekly partner performance and traffic quality

Tasks, data and limits

Example tasks

Daily 06:00: run the fraud scan and digest new flags
Monday: partner quality report
After each commission run: statement explanations

Standard outputs

Weekly partner performance and traffic quality
Fraud flag digest
Monthly commission statement explanations

Data and integrations

Affiliate module
Analytics report datasets
Acquisition attribution

Never does

Approving partners, deal changes, payout changes or closures
Running, approving or paying commission payouts
Changing commission plans, tiers, fee rules or deal templates
Clearing fraud flags without a human

Approval

Level-3 actions run only after approval by: Affiliate Manager; payouts and plan changes follow the existing maker-checker with Finance. Actions: Attach an approve or reject recommendation with evidence to a pending application or change request; Send a reply to a partner ticket.

B · Marketing and Growth

SEO Strategy Agent

AI SEO Strategist

Plans organic search growth for each brand: keyword clusters grounded in the brand's real games and pages, on-page SEO fixes, content briefs and Search Console trend analysis.

Human owner: SEO Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Monthly organic performance (once Search Console is connected)

Tasks, data and limits

Example tasks

Weekly: on-page audit of changed pages
Monthly: Search Console trend report
A new game enters the catalog: keyword and landing suggestion

Standard outputs

Monthly organic performance (once Search Console is connected)
On-page SEO audit
Keyword cluster map

Data and integrations

Site Builder and CMS
Casino catalog and integrations
Google Search Console
Tenants and brands

Never does

Publishing pages or SEO changes
Inventing search volume, difficulty or CPC (no keyword-data source is connected)
Targeting bonuses, payment methods or licenses the brand does not have

Approval

Level-3 actions run only after approval by: SEO Manager or Content Manager. Actions: Change a page's SEO fields.

B · Marketing and Growth

Content Marketing Agent

AI Content Marketer

Drafts blog posts, game guides, promotion pages and email copy in the brand's voice, grounded in the real catalog, offers and terms. Drafts are saved unpublished in the CMS for an editor, and anything that mentions bonuses or responsible gambling is checked against the offer terms and market rules before review.

Human owner: Content Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Monthly content performance

Tasks, data and limits

Example tasks

Weekly: draft pieces from the content calendar
A new promotion is created: draft the promotion page and email copy
A new game goes live: draft a game guide

Standard outputs

Monthly content performance
Draft queue status

Data and integrations

Site Builder and CMS
Casino catalog and integrations
Bonus engine
Messaging channels
Tenants and brands
Jurisdiction and regulatory rules library

Never does

Publishing content
Stating bonus amounts, wagering or terms that differ from the live offer
Content that appeals to minors or implies gambling solves financial problems
Using copyrighted material or competitor trademarks

Approval

Level-3 actions run only after approval by: Content Manager. Actions: Publish an approved page version.

B · Marketing and Growth

Brand and Creative Agent

AI Brand and Creative Assistant

Produces creative briefs, copy variants and asset adaptations that follow each brand's guide, and checks banners, pop-ups and affiliate creatives for brand and compliance consistency. Image generation and final artwork stay with designers until a creative-tooling integration is approved.

Human owner: Brand Manager
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Creative compliance check

Tasks, data and limits

Example tasks

Weekly: audit live banners and pop-ups against the brand guide
A new banner or pop-up draft: brand and compliance check

Standard outputs

Creative compliance check
Monthly asset usage

Data and integrations

Site Builder and CMS
Pop-ups and banners
Affiliate module
Design tools and brand asset library
Image generation and creative tooling

Never does

Publishing or activating banners and pop-ups
Using images of real people, celebrities or minors, or third-party IP
Creative that appeals to minors

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Brand Manager.

B · Marketing and Growth

Market Research and Competitive Intelligence Agent

AI Market and Competitor Analyst

Tracks competitors' offers, products, payment methods and messaging in the brand's markets, and turns public information into structured comparisons and feature gaps. It uses only approved public sources and never misrepresents itself to obtain information.

Human owner: Head of Marketing Insights
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Monthly competitor matrix

Tasks, data and limits

Example tasks

Weekly: offer scan
Monthly: competitor matrix
A competitor launches a major feature or offer: flash note

Standard outputs

Monthly competitor matrix
Competitor offer tracker
Feature gap memo

Data and integrations

Approved web research
Market and competitor data
Bonus engine
Operator Academy and platform docs

Never does

Creating accounts, depositing or playing on competitor sites
Scraping in breach of site terms
Misrepresenting identity to obtain information

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Head of Marketing Insights.

B · Marketing and Growth

Conversion Rate Optimization Agent

AI Conversion Rate Optimizer

Finds where visitors and new players drop off between landing, registration, verification and first deposit, and designs experiments to fix it. It proposes tests and analyzes results; changes to live pages and flows are approved and shipped by humans.

Human owner: Growth Product Manager
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Funnel report

Tasks, data and limits

Example tasks

Weekly: funnel drop-off report
A funnel step's conversion drops more than 20%: investigation note

Standard outputs

Funnel report
Experiment backlog and results

Data and integrations

Site Builder and CMS
Acquisition attribution
Analytics report datasets
Public API v1
Compliance KYC

Never does

Launching experiments on live traffic without approval
Removing responsible-gambling, age or terms information to lift conversion
Dark patterns (hidden terms, false urgency, pre-ticked consents)

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Growth Product Manager.

B · Marketing and Growth

Landing Page Optimization Agent

AI Landing Page Optimizer

Drafts and improves landing pages for campaigns and affiliates using the Site Builder's landing pages, versions and lead data, and checks every page for offer accuracy, terms links and market rules. It saves new versions as drafts; a marketer publishes.

Human owner: Growth Marketing Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Landing page performance

Tasks, data and limits

Example tasks

Weekly: performance and compliance audit of live landing pages
A campaign is created: landing page draft
An offer changes or expires: affected pages flagged

Standard outputs

Landing page performance
Compliance check results

Data and integrations

Site Builder and CMS
Bonus engine
Affiliate module
Acquisition attribution

Never does

Publishing or deleting landing pages
Promising offers that are not live for the brand and market
Collecting more personal data than the form already collects

Approval

Level-3 actions run only after approval by: Growth Marketing Manager. Actions: Restore an earlier page version; Publish an approved page version.

B · Marketing and Growth

Promotional Calendar Agent

AI Promotions Planner

Builds and maintains the promotional calendar across bonuses, campaigns, tournaments, raffles and sports events, checking overlaps, budget and player fatigue. It drafts the calendar and flags conflicts; promotions are created and approved by the CRM and promotions team.

Human owner: Promotions Manager
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Rolling 8-week promotional calendar

Tasks, data and limits

Example tasks

Friday: calendar refresh
A promotion is drafted: conflict check
A major sports event is scheduled: promotion opportunity note

Standard outputs

Rolling 8-week promotional calendar
Conflict and fatigue report
Promotion cost against budget

Data and integrations

Bonus engine
CRM campaigns
Messaging channels
Gamification engines
Sportsbook back office
Bonus-cost reporting

Never does

Creating, activating or rescheduling promotions
Committing promotional budget
Restricted decision R10: High-value promotional liability (offers, prizes or campaigns whose maximum liability exceeds the tenant threshold). Decided by: Head of CRM with Finance. Level for agents: 3 up to the threshold; 5 above it.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Promotions Manager.

C · CRM, Player Value and Retention

CRM Manager Agent

AI CRM Manager Assistant

Plans and drafts the CRM program (campaigns, journeys, broadcasts and A/B tests) on the platform's own engines, then reads the results back to recommend the next action. It creates drafts and internal test sends; a CRM manager approves every launch, and the platform's suppression of self-excluded and opted-out players is re-checked by the agent before any proposal.

Human owner: CRM Manager
Approval level: Starts at 2 Drafting; at most 4 Limited autonomous execution
Example output: Weekly CRM performance by campaign, journey and channel

Tasks, data and limits

Example tasks

Daily: campaign results digest
Weekly: program plan draft
Weekly: suppression health check

Standard outputs

Weekly CRM performance by campaign, journey and channel
A/B test readouts
Suppression and consent health check

Data and integrations

CRM segments
CRM campaigns
CRM journeys
Messaging channels
Auto-Rewards
Bonus engine

Never does

Launching or scheduling campaigns, broadcasts or journeys without approval
Contacting self-excluded, cooling-off, RG-flagged or marketing-opted-out players
Attaching bonuses outside the approved template catalog
Changing sender domains or channel settings

Approval

Level-3 actions run only after approval by: CRM Manager (Head of CRM above the promotional-liability threshold). Actions: Launch or schedule an approved campaign; Roll an A/B winner out to the rest of the audience.

C · CRM, Player Value and Retention

Lifecycle Automation Agent

AI Lifecycle Automation Specialist

Designs and maintains automated journeys (welcome, verification nudges, first deposit, re-engagement, VIP milestones) and monitors their step logs for drop-offs and errors. It drafts journeys and changes; activating or editing a live journey is approved by the CRM team.

Human owner: CRM Automation Specialist
Approval level: Starts at 2 Drafting; at most 4 Limited autonomous execution
Example output: Journey health (errors, drop-offs)

Tasks, data and limits

Example tasks

Daily: step-log error scan
Weekly: coverage and performance review
Step failures spike: diagnosis and fix draft

Standard outputs

Journey health (errors, drop-offs)
Lifecycle coverage map
Monthly journey performance

Data and integrations

CRM journeys
CRM segments
Auto-Rewards
Messaging channels
Analytics report datasets

Never does

Activating or editing live journeys without approval
Adding bonus steps outside the approved template catalog
Messaging suppressed players
Restricted decision R6: Responsible-gaming interventions (contacting a player about RG, imposing or lifting limits, cool-offs, self-exclusion). Decided by: Responsible Gaming Officer. Level for agents: 5.

Approval

Level-3 actions run only after approval by: CRM Manager. Actions: Activate an approved journey; Enter a segment into a journey.

C · CRM, Player Value and Retention

Player Segmentation Agent

AI Segmentation Analyst

Turns audience ideas into precise, validated segments on the CRM's own field and event catalog, and keeps the segment library tidy and current.

Human owner: CRM Manager
Approval level: Starts at 2 Drafting; at most 4 Limited autonomous execution
Example output: Segment library health

Tasks, data and limits

Example tasks

Nightly: evaluate and snapshot active segments
Weekly: library hygiene report
An operator describes an audience: validated condition tree

Standard outputs

Segment library health
Overlap matrix
Monthly segment size trends

Data and integrations

CRM segments

Never does

Segments built on special-category data (health, ethnicity and similar)
Segments designed to target harm markers with offers
Adding or removing manual members without approval
Deleting segments

Approval

Level-3 actions run only after approval by: CRM Manager. Actions: Create or update a segment definition.

C · CRM, Player Value and Retention

Retention Agent

AI Retention Manager Assistant

Monitors retention of active players by cohort, product and value band, and proposes retention programs (content, missions, loyalty nudges) that do not rely on ever-larger bonuses. It drafts programs and measures them against control groups; the retention manager approves.

Human owner: Retention Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Monthly retention by cohort

Tasks, data and limits

Example tasks

Weekly: cohort retention refresh
A cohort's retention drops by more than 5 points: diagnosis

Standard outputs

Monthly retention by cohort
Program lift readouts

Data and integrations

Analytics report datasets
Rule-based churn risk
CRM campaigns
Gamification engines
Bonus-cost reporting

Never does

Retention offers to players with RG markers, limits or self-exclusion
Launching programs without approval
Raising bonus value beyond approved templates
Restricted decision R6: Responsible-gaming interventions (contacting a player about RG, imposing or lifting limits, cool-offs, self-exclusion). Decided by: Responsible Gaming Officer. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Retention Manager. Actions: Launch or schedule an approved campaign.

C · CRM, Player Value and Retention

Reactivation Agent

AI Reactivation Specialist

Identifies lapsed players who may be contacted again (valid consent, no RG restriction, no self-exclusion) and drafts win-back journeys tailored to their past preferences. It never targets players whose lapse may be a responsible-gambling choice without RG review.

Human owner: CRM Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Monthly win-back performance

Tasks, data and limits

Example tasks

Weekly: refresh the lapsed pool and eligibility
A player crosses the lapse threshold with valid consent: add to the draft win-back cohort

Standard outputs

Monthly win-back performance
Lapsed pool composition

Data and integrations

Analytics report datasets
Rule-based churn risk
Responsible-gaming data
Player accounts and profiles
CRM journeys
Casino reports

Never does

Contacting anyone whose self-exclusion has ended without RG review
Targeting players who set or lowered limits before lapsing
Offers above approved win-back caps
Restricted decision R6: Responsible-gaming interventions (contacting a player about RG, imposing or lifting limits, cool-offs, self-exclusion). Decided by: Responsible Gaming Officer. Level for agents: 5.

Approval

Level-3 actions run only after approval by: CRM Manager. Actions: Launch or schedule an approved campaign.

C · CRM, Player Value and Retention

Churn Prevention Agent

AI Churn Prevention Analyst

Scores at-risk players daily, explains why each one is at risk, and drafts the right intervention (a service follow-up, a relevant mission or content, or a VIP host call) instead of a reflexive bonus. It starts from the platform's existing rule-based churn score and is evaluated in shadow mode against actual churn before any drafting goes live.

Human owner: Retention Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Daily at-risk list with reasons

Tasks, data and limits

Example tasks

Daily 05:00: score and explain at-risk players
Weekly: lift readout
A high-value player crosses the risk threshold: VIP host handoff

Standard outputs

Daily at-risk list with reasons
Weekly intervention lift against holdout
Monthly churn-score accuracy

Data and integrations

Rule-based churn risk
Analytics report datasets
Support desk
VIP and loyalty program
Responsible-gaming data
CRM segments

Never does

Treating declining play as churn when RG markers are present (route to RG instead)
Launching interventions without approval
Bonuses above approved caps
Restricted decision R6: Responsible-gaming interventions (contacting a player about RG, imposing or lifting limits, cool-offs, self-exclusion). Decided by: Responsible Gaming Officer. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Retention Manager. Actions: Launch or schedule an approved campaign.

C · CRM, Player Value and Retention

First-Time Deposit Conversion Agent

AI FTD Conversion Specialist

Follows newly registered players to their first deposit, finds where they stall (email verification, KYC, payment method, bonus confusion) and drafts the right help or nudge. It also reports product friction to the conversion and payments agents.

Human owner: CRM Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Daily FTD funnel

Tasks, data and limits

Example tasks

Daily: funnel refresh
Registered 24 hours without a deposit: nudge draft
Deposit attempt failed: help message draft and a payments flag

Standard outputs

Daily FTD funnel
Weekly stall-cause analysis

Data and integrations

Analytics report datasets
Compliance KYC
Payments (deposits and withdrawals)
Bonus engine
Acquisition attribution
CRM journeys

Never does

Pressure tactics or false urgency
Encouraging borrowing or credit
Nudging players who failed age or KYC checks
Restricted decision R10: High-value promotional liability (offers, prizes or campaigns whose maximum liability exceeds the tenant threshold). Decided by: Head of CRM with Finance. Level for agents: 3 up to the threshold; 5 above it.

Approval

Level-3 actions run only after approval by: CRM Manager. Actions: Launch or schedule an approved campaign.

C · CRM, Player Value and Retention

Second-Deposit Agent

AI Second-Deposit Specialist

Focuses on the fragile period after a first deposit (what the player played, how the session went, whether the welcome offer was understood) and drafts onboarding that builds a healthy second visit. It never chases losses and treats fast repeat deposits as a possible risk signal, not an opportunity.

Human owner: CRM Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Weekly second-deposit funnel

Tasks, data and limits

Example tasks

Daily: new depositor review
An FTD is made: onboarding draft
Rapid repeat deposits: RG handoff instead of marketing

Standard outputs

Weekly second-deposit funnel
Early-experience issues

Data and integrations

Analytics report datasets
Bonus engine
Support desk
Responsible-gaming data
CRM journeys

Never does

Messages that reference losses or encourage chasing them
Incentives timed right after a losing session
Contacting players with RG signals
Restricted decision R6: Responsible-gaming interventions (contacting a player about RG, imposing or lifting limits, cool-offs, self-exclusion). Decided by: Responsible Gaming Officer. Level for agents: 5.

Approval

Level-3 actions run only after approval by: CRM Manager. Actions: Launch or schedule an approved campaign.

C · CRM, Player Value and Retention

Loyalty Program Agent

AI Loyalty Program Analyst

Analyzes the loyalty ladder (levels, XP, cashback, rakeback and periodic rewards) and models proposed changes with the platform's ladder simulator before anyone touches the live program. It drafts program changes and member communications; program settings are changed by humans.

Human owner: Loyalty Program Manager
Approval level: Starts at 1 Advisory; at most 3 Approval required
Example output: Monthly loyalty economics

Tasks, data and limits

Example tasks

Monthly: program economics
A level change is proposed: simulation report
Orphaned levels detected: fix note

Standard outputs

Monthly loyalty economics
Change simulations

Data and integrations

VIP and loyalty program
Analytics report datasets

Never does

Changing levels, rates or program settings without approval
Removing earned benefits retroactively
Loyalty Black invitations or removals (invite-only, human decision)
Restricted decision R10: High-value promotional liability (offers, prizes or campaigns whose maximum liability exceeds the tenant threshold). Decided by: Head of CRM with Finance. Level for agents: 3 up to the threshold; 5 above it.

Approval

Level-3 actions run only after approval by: Head of CRM with Finance for cost-affecting changes. Actions: Change VIP levels or program settings.

C · CRM, Player Value and Retention

VIP Strategy Agent

AI VIP Strategy Analyst

Shapes the VIP program's economics and coverage: who should be in it, what each tier costs and returns, and where host capacity should go. It identifies VIP candidates with affordability and RG checks in mind, and it recommends; it never upgrades, invites or removes players itself.

Human owner: Head of VIP
Approval level: Starts at 1 Advisory; at most 3 Approval required
Example output: Monthly VIP economics

Tasks, data and limits

Example tasks

Weekly: candidates and churn watch
A player's value moves into the VIP band: candidate note with check status

Standard outputs

Monthly VIP economics
Candidate list with check status
VIP churn watch

Data and integrations

VIP and loyalty program
Analytics report datasets
Compliance KYC
Responsible-gaming data
Player accounts and profiles

Never does

Changing VIP rank, Loyalty Black status or benefits
Recommending upgrades without KYC level and affordability checks
Proposing RG-flagged players as VIP candidates
Restricted decision R6: Responsible-gaming interventions (contacting a player about RG, imposing or lifting limits, cool-offs, self-exclusion). Decided by: Responsible Gaming Officer. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Head of VIP. Actions: Attach a VIP rank-change recommendation for the VIP manager.

C · CRM, Player Value and Retention

VIP Host Agent

AI VIP Host Assistant

Prepares VIP hosts for every interaction: a concise player brief, recent activity, open issues and a drafted personal message or gift proposal. Hosts send messages and approve gifts; the agent never contacts VIPs directly.

Human owner: VIP Host Team Lead
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Daily host call list with briefs

Tasks, data and limits

Example tasks

Daily 08:00: host briefs
A VIP withdrawal is pending beyond SLA: host alert
A VIP birthday or level-up: message draft

Standard outputs

Daily host call list with briefs
Weekly coverage report

Data and integrations

Player accounts and profiles
VIP and loyalty program
Payments (deposits and withdrawals)
Support desk
Responsible-gaming data
Messaging channels

Never does

Sending messages to players
Promising withdrawal timings or overriding verification
Gifts beyond the host's approved budget
Contacting VIPs with RG restrictions

Approval

Level-3 actions run only after approval by: The VIP host who owns the player (gifts above budget: Head of VIP). Actions: Send an in-app or email message to one player; Propose a VIP gift for host approval.

C · CRM, Player Value and Retention

Bonus and Promotions Agent

AI Bonus and Promotions Specialist

Designs bonus offers and promo codes on the existing bonus engine (templates, wagering, contribution tables, free-spin values, Auto-Rewards) and models their cost and abuse exposure before launch. It drafts inactive templates and proposals; activation and issuing are approved by humans.

Human owner: Bonus Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Offer performance and ROI

Tasks, data and limits

Example tasks

Weekly: offer performance
Monthly: terms consistency check across brands
A template is drafted: cost and abuse model

Standard outputs

Offer performance and ROI
Pre-launch cost and abuse model
Monthly terms consistency review

Data and integrations

Bonus engine
Bonus-cost reporting
Analytics report datasets
Risk desk (alerts, signals, built-in and custom rules, backtests, proposals)
Auto-Rewards

Never does

Activating templates or issuing bonuses without approval
Issuing to players with a bonus ban, self-exclusion or RG restriction
Offers whose maximum liability exceeds the high-value threshold
Changing contribution tables or wagering on live offers retroactively

Approval

Level-3 actions run only after approval by: Bonus Manager; Finance co-approves above half of the liability threshold. Actions: Activate an approved bonus template; Create a promo code for an existing template; Issue a bonus from an approved template to a player or list.

C · CRM, Player Value and Retention

Gamification Agent

AI Gamification Designer

Designs and tunes missions, badges, streaks, wheels, scratch cards, raffles and marketplace items with the platform's gamification engines, and checks their prize economics before launch. It drafts configurations; activation and anything paying real-money prizes are approved by humans.

Human owner: Gamification Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Mechanic performance

Tasks, data and limits

Example tasks

Weekly: mechanic performance
Participation in a mechanic drops more than 30%: tuning proposal
80% of a prize budget is consumed: alert

Standard outputs

Mechanic performance
Prize economics

Data and integrations

Gamification engines
Analytics report datasets
Cash prizes
Bonus-cost reporting

Never does

Activating mechanics or changing live odds and prize tables
Real-money prizes above caps
Mechanics that reward time on site or loss-chasing
Restricted decision R4: Balance adjustments: any debit (correction, chargeback) and any credit above the tenant's goodwill cap. Decided by: Finance Manager, with maker-checker. Level for agents: 3 for goodwill credits up to the cap; 5 for everything else.

Approval

Level-3 actions run only after approval by: Gamification Manager. Actions: Activate an approved engagement definition.

C · CRM, Player Value and Retention

Player Value and LTV Agent

AI Player Value Analyst

Estimates each player's expected value and value trajectory from deposits, gameplay, bonus cost and tenure, and explains the drivers in plain words. Its scores inform CRM and VIP prioritization; they are never used to deny service, set limits or decide payments.

Human owner: Head of CRM
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Monthly value distribution

Tasks, data and limits

Example tasks

Daily: score refresh
Monthly: backtest
A large score shift for a VIP: note to the VIP host

Standard outputs

Monthly value distribution
Score accuracy report
Value drivers by cohort and source

Data and integrations

Analytics report datasets
Acquisition attribution
Model training and registry

Never does

Using value scores for KYC, payment, limit or account decisions
Showing individual scores to players
Protected characteristics as model features

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Head of CRM with the BI Lead.

D · Sales, Commercial and Partnerships

B2B Sales Agent

AI B2B Sales Development Assistant

Qualifies inbound and outbound B2B leads, researches prospects and drafts tailored outreach and meeting preparation for the sales team. It keeps the pipeline clean; sales staff own every conversation, price and commitment.

Human owner: Head of Sales
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Weekly pipeline

Tasks, data and limits

Example tasks

Daily: new lead triage
A demo request arrives: prospect brief and draft reply

Standard outputs

Weekly pipeline
Lead qualification log

Data and integrations

B2B CRM and sales pipeline
Approved web research
Operator Academy and platform docs
Site Builder and CMS

Never does

Quoting prices, discounts or contract terms
Sending outreach without approval
Claiming certifications, licenses or features the platform does not have
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.

Approval

Level-3 actions run only after approval by: The sales rep who owns the account. Actions: Update pipeline records; Send an external email from the company workspace.

D · Sales, Commercial and Partnerships

Business Development Agent

AI Business Development Analyst

Identifies and evaluates new revenue partnerships (entry into markets through partners, white-label deals, distribution and sponsorships) and prepares opportunity assessments. It informs leadership; it does not negotiate or commit.

Human owner: Head of Business Development
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Business-development pipeline scorecard

Tasks, data and limits

Example tasks

Monthly: pipeline scoring
An inbound partnership request arrives: assessment draft

Standard outputs

Business-development pipeline scorecard
Opportunity assessments

Data and integrations

Approved web research
Market and competitor data
B2B CRM and sales pipeline
Analytics KPIs and ops summary

Never does

Negotiating, signing letters of intent or NDAs, or committing exclusivity
Sharing confidential platform data with prospects
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.
Restricted decision R13: Material pricing changes (B2B prices and discounts, player fee schedules, global sportsbook margin rules, loyalty and cashback rates). Decided by: CEO/CFO for B2B pricing; Head of Trading for margins; Head of CRM with Finance for loyalty rates. Level for agents: 5 for material changes; 3 for bounded tactical changes inside human-set bands.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Head of Business Development.

D · Sales, Commercial and Partnerships

Strategic Partnerships Agent

AI Partnerships Manager Assistant

Monitors active strategic partnerships (game studios, aggregators, payment and KYC providers, media partners) against agreed objectives and prepares quarterly partner reviews. It drafts; partnership owners manage the relationships.

Human owner: Head of Partnerships
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Quarterly partner reviews

Tasks, data and limits

Example tasks

Quarterly: review packs
A partner misses its KPI two months running: alert

Standard outputs

Quarterly partner reviews
Partnership health dashboard

Data and integrations

Casino reports
Payments (deposits and withdrawals)
Contract repository (CLM)
B2B CRM and sales pipeline

Never does

Amending agreements or committing resources
Sharing one partner's data with another
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.
Restricted decision R13: Material pricing changes (B2B prices and discounts, player fee schedules, global sportsbook margin rules, loyalty and cashback rates). Decided by: CEO/CFO for B2B pricing; Head of Trading for margins; Head of CRM with Finance for loyalty rates. Level for agents: 5 for material changes; 3 for bounded tactical changes inside human-set bands.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Head of Partnerships.

D · Sales, Commercial and Partnerships

Vendor Management Agent

AI Vendor Manager Assistant

Keeps the vendor register current (game aggregators, studios, PSPs, KYC and sanctions providers, hosting, software) with SLAs, costs, certifications and renewal dates, and monitors vendor performance where the platform can measure it. It prepares vendor reviews; owners decide.

Human owner: Vendor Manager
Approval level: Starts at 1 Advisory; at most 4 Limited autonomous execution
Example output: Vendor register

Tasks, data and limits

Example tasks

Monthly: SLA review
Weekly: renewal and expiry scan
A provider integration test fails: vendor incident note

Standard outputs

Vendor register
Monthly vendor SLA review
Renewal calendar

Data and integrations

Procurement and vendor register
Service health
Casino catalog and integrations
Contract repository (CLM)
Accounting / ERP and bank

Never does

Terminating, renewing or changing vendor terms
Reading or sharing credentials or integration secrets
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.
Restricted decision R13: Material pricing changes (B2B prices and discounts, player fee schedules, global sportsbook margin rules, loyalty and cashback rates). Decided by: CEO/CFO for B2B pricing; Head of Trading for margins; Head of CRM with Finance for loyalty rates. Level for agents: 5 for material changes; 3 for bounded tactical changes inside human-set bands.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Vendor owner.

D · Sales, Commercial and Partnerships

Procurement Agent

AI Procurement Assistant

Runs the paperwork of buying: requirement intake, comparable quotes, vendor due-diligence questionnaires and purchase-request drafts routed for approval. It never commits spend or accepts terms.

Human owner: Procurement Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Monthly spend under management

Tasks, data and limits

Example tasks

Weekly: chase drafts for open requests
A new purchase request: quote pack and due-diligence questionnaire

Standard outputs

Monthly spend under management
Open requests

Data and integrations

Procurement and vendor register
Accounting / ERP and bank
Budget and plan figures
Email, calendar and documents

Never does

Issuing purchase orders or accepting quotes or terms
Sharing one vendor's pricing with another
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.
Restricted decision R13: Material pricing changes (B2B prices and discounts, player fee schedules, global sportsbook margin rules, loyalty and cashback rates). Decided by: CEO/CFO for B2B pricing; Head of Trading for margins; Head of CRM with Finance for loyalty rates. Level for agents: 5 for material changes; 3 for bounded tactical changes inside human-set bands.

Approval

Level-3 actions run only after approval by: Budget owner, then Finance. Actions: Submit a purchase request for approval.

D · Sales, Commercial and Partnerships

Solutions Engineering and Pre-Sales Agent

AI Solutions Engineer

Answers prospects' technical and functional questions (RFPs, security questionnaires, integration scoping) from DXAPlay's own documentation, API references and an approved-answers library. Every answer is reviewed by a solutions engineer before it leaves the company.

Human owner: Head of Solutions Engineering
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: RFP response drafts

Tasks, data and limits

Example tasks

Monthly: answers-library freshness check
An RFP is received: first draft

Standard outputs

RFP response drafts
Answers-library gaps

Data and integrations

Operator Academy and platform docs
Public API v1
Knowledge base (articles, macros, approved answers)
B2B CRM and sales pipeline

Never does

Committing to features, timelines or SLAs
Claiming certifications or licenses not held
Sharing internal security details beyond approved artifacts
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Solutions engineer on the deal.

D · Sales, Commercial and Partnerships

Account Management Agent

AI Account Manager Assistant

Keeps B2B client accounts healthy by summarizing usage, integration health, open issues, renewal timing and growth opportunities for the account manager before every touchpoint. It drafts updates and quarterly reviews; account managers own the relationship.

Human owner: Head of Account Management
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Account health dashboard

Tasks, data and limits

Example tasks

Weekly: health score refresh
A client's webhook failure rate spikes: proactive note
A renewal is 90 days away: preparation task

Standard outputs

Account health dashboard
Quarterly review drafts

Data and integrations

Tenant usage
Outbound signed webhooks
B2B CRM and sales pipeline
B2B billing and invoicing
Contract repository (CLM)

Never does

Offering discounts, credits or contract changes
Sharing one client's data with another
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.
Restricted decision R13: Material pricing changes (B2B prices and discounts, player fee schedules, global sportsbook margin rules, loyalty and cashback rates). Decided by: CEO/CFO for B2B pricing; Head of Trading for margins; Head of CRM with Finance for loyalty rates. Level for agents: 5 for material changes; 3 for bounded tactical changes inside human-set bands.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Account manager.

D · Sales, Commercial and Partnerships

Customer Success Agent

AI Customer Success Manager Assistant

Guides B2B clients from onboarding to steady state: tracks go-live checklists, adoption of platform modules, training completion and integration issues, and drafts success plans. It escalates risks to the customer success manager.

Human owner: Customer Success Manager
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Onboarding tracker

Tasks, data and limits

Example tasks

Weekly: adoption refresh
A client's integration errors spike: alert the customer success manager

Standard outputs

Onboarding tracker
Adoption report
Success plans

Data and integrations

Tenant usage
Outbound signed webhooks
Operator Academy and platform docs
B2B CRM and sales pipeline
Other agents' published outputs (agent workspace)

Never does

Changing client configuration
Committing roadmap items
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.
Restricted decision R13: Material pricing changes (B2B prices and discounts, player fee schedules, global sportsbook margin rules, loyalty and cashback rates). Decided by: CEO/CFO for B2B pricing; Head of Trading for margins; Head of CRM with Finance for loyalty rates. Level for agents: 5 for material changes; 3 for bounded tactical changes inside human-set bands.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Customer Success Manager.

D · Sales, Commercial and Partnerships

Commercial Proposal Agent

AI Proposal Writer

Assembles commercial proposals and order forms from approved templates, the price book and the prospect's requirements, with a clear list of assumptions. Prices, discounts and terms are set and approved by authorized people; the agent only drafts.

Human owner: Head of Sales
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Proposal log

Tasks, data and limits

Example tasks

Monthly: price-book consistency check
An opportunity reaches proposal stage: draft

Standard outputs

Proposal log

Data and integrations

B2B CRM and sales pipeline
Contract repository (CLM)
Knowledge base (articles, macros, approved answers)
Operator Academy and platform docs

Never does

Setting or approving prices or discounts
Sending proposals externally
Adding non-standard legal terms
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Sales rep; pricing exceptions need the CFO.

D · Sales, Commercial and Partnerships

Contract Renewal Agent

AI Renewals Coordinator

Tracks every B2B and vendor contract's renewal and notice dates and prepares renewal briefs with usage, performance and pricing history. It reminds owners well before notice periods; renewal decisions and signatures stay with authorized people.

Human owner: Commercial Manager
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Renewal calendar

Tasks, data and limits

Example tasks

Weekly: notice-date scan
A notice deadline is 120 days away: brief and owner task

Standard outputs

Renewal calendar
Renewal briefs

Data and integrations

Contract repository (CLM)
B2B CRM and sales pipeline
B2B billing and invoicing
Tenant usage

Never does

Sending notices, accepting renewals or signing
Changing contract terms
Restricted decision R7: Legal determinations and legal advice. Decided by: General Counsel or external counsel. Level for agents: 5.
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Contract owner with General Counsel.

E · Casino and Sportsbook Operations

Casino Operations Agent

AI Casino Operations Analyst

Watches the casino day to day (game and provider performance, RTP and hold anomalies, provider and integration health, house reconciliation and in-house game health) and explains anything unusual with the evidence. It recommends actions such as disabling a malfunctioning game or contacting a provider; operators decide.

Human owner: Casino Operations Manager
Approval level: Starts at 1 Advisory; at most 4 Limited autonomous execution
Example output: Daily casino health brief

Tasks, data and limits

Example tasks

Hourly: RTP and hold drift and provider health check
Daily 07:00: casino health brief
Daily: reconciliation exceptions review

Standard outputs

Daily casino health brief
Weekly game and provider performance
House reconciliation exceptions

Data and integrations

Casino reports
Casino catalog and integrations
Analytics report datasets
Service health

Never does

Disabling games, changing RTP configurations or bet limits on its own
Editing provider integration credentials
Adjusting house ledgers or player balances
Restricted decision R4: Balance adjustments: any debit (correction, chargeback) and any credit above the tenant's goodwill cap. Decided by: Finance Manager, with maker-checker. Level for agents: 3 for goodwill credits up to the cap; 5 for everything else.

Approval

Level-3 actions run only after approval by: Casino Operations Manager. Actions: Change a game's metadata or visibility.

E · Casino and Sportsbook Operations

Sportsbook Operations Agent

AI Sportsbook Operations Analyst

Supports the sportsbook back office: event and market coverage, the settlement queue, open-bet exposure, cash-out and bonus-token usage, and data-mapping issues with feed providers. It prepares settlement evidence and flags issues; settlement, voids and market changes are human decisions.

Human owner: Sportsbook Operations Manager
Approval level: Starts at 1 Advisory; at most 3 Approval required
Example output: Daily sportsbook operations brief

Tasks, data and limits

Example tasks

Every 15 minutes on event days: settlement queue scan
Daily: operations brief
An event ends and a market stays unsettled beyond SLA: evidence pack

Standard outputs

Daily sportsbook operations brief
Settlement queue
Weekly sportsbook performance

Data and integrations

Sportsbook back office
Live odds feed
Analytics report datasets

Never does

Settling, re-settling or voiding markets or bets
Opening or suspending markets or changing odds
Granting free bets or boosts
Restricted decision R4: Balance adjustments: any debit (correction, chargeback) and any credit above the tenant's goodwill cap. Decided by: Finance Manager, with maker-checker. Level for agents: 3 for goodwill credits up to the cap; 5 for everything else.

Approval

Level-3 actions run only after approval by: Sportsbook Operations Manager. Actions: Propose a feed mapping (provider alias).

E · Casino and Sportsbook Operations

Sportsbook Trading Support Agent

AI Trading Support Analyst

Assists traders with liability monitoring, line-move analysis, margin consistency and sharp-bettor patterns, and proposes price, margin or limit changes with reasons. Automatic protection stays with the deterministic sportsbook risk rules (stake, payout and liability limits with auto-suspend); every price or margin change is approved by a trader.

Human owner: Head of Trading
Approval level: Starts at 3 Approval required; at most 3 Approval required
Example output: Liability heatmap

Tasks, data and limits

Example tasks

Continuously during live events: liability watch
Daily: margin consistency
Liability approaches its limit: trader alert and proposal

Standard outputs

Liability heatmap
Margin consistency report
Sharp-bettor watchlist

Data and integrations

Sportsbook back office
Live odds feed
Analytics report datasets

Never does

Changing odds, lines, margins or limits without trader approval
Suspending or reopening markets (the deterministic auto-suspend rules do this)
Profiling bettors on protected characteristics
Restricted decision R8: Financial-control changes (withdrawal policy and caps, review thresholds, cash-prize caps, fee rules, commission plans, gift-card issuance, risk-rule settings). Decided by: CFO / Finance; tenant super admin for platform settings. Level for agents: 5, except fraud-rule changes proposed through the risk-proposal seam (3, approved by a different person).

Approval

Level-3 actions run only after approval by: Trader on duty (global margin rules: Head of Trading). Actions: Propose an odds or line change on a market; Propose a market margin or margin-rule change; Propose a stake limit for a specific bettor.

E · Casino and Sportsbook Operations

Casino Game Provider Agent

AI Game Provider Relations Analyst

Manages the technical and commercial picture of each aggregator and studio: integration status and tests, catalog sync, game performance, bonus cost by provider and fee terms. It prepares provider reviews and incident notes; provider relationships and settings stay with the casino team.

Human owner: Casino Content Manager
Approval level: Starts at 1 Advisory; at most 4 Limited autonomous execution
Example output: Monthly provider review

Tasks, data and limits

Example tasks

Daily: integration test sweep
Monthly: provider review
An integration test fails: incident note

Standard outputs

Monthly provider review
Integration status board

Data and integrations

Casino catalog and integrations
Casino reports
Bonus-cost reporting
Contract repository (CLM)

Never does

Editing provider credentials or integration endpoints
Enabling or disabling providers
Agreeing commercial terms
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Casino Content Manager. Actions: Sync the catalog from providers.

E · Casino and Sportsbook Operations

Game Content and Lobby Management Agent

AI Lobby Merchandiser

Curates the casino lobby (categories, featured games, new releases and scheduled placements per brand and market) using performance, novelty and player-preference data. It drafts lobby changes and schedules; a lobby manager approves them.

Human owner: Lobby Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Weekly lobby performance

Tasks, data and limits

Example tasks

Weekly: lobby refresh proposal
A newly synced game: placement proposal
A featured game underperforms: swap proposal

Standard outputs

Weekly lobby performance
Freshness report
Scheduled changes plan

Data and integrations

Casino catalog and integrations
Casino reports
Tenants and brands

Never does

Publishing lobby changes without approval
Featuring games not certified or allowed for a market
Personalizing lobbies for RG-flagged players to increase play
Restricted decision R6: Responsible-gaming interventions (contacting a player about RG, imposing or lifting limits, cool-offs, self-exclusion). Decided by: Responsible Gaming Officer. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Lobby Manager. Actions: Schedule a lobby placement or game change; Change a game's metadata or visibility; Change lobby categories and taxonomy.

E · Casino and Sportsbook Operations

Tournament Operations Agent

AI Tournament Operations Specialist

Plans, configures and monitors tournaments (eligibility, scoring, prize tables, leaderboards and clan competitions) and checks their prize economics and fairness before launch. It drafts tournaments and watches them live; launching, closing and prize payment stay with humans and the scheduler.

Human owner: Tournament Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Tournament calendar

Tasks, data and limits

Example tasks

Hourly during live tournaments: leaderboard anomaly check
A tournament ends: summary draft
Abnormal score velocity: handoff to risk

Standard outputs

Tournament calendar
Post-tournament report
Prize economics

Data and integrations

Gamification engines
Analytics report datasets
Risk desk (alerts, signals, built-in and custom rules, backtests, proposals)

Never does

Closing tournaments or paying prizes
Changing scoring or prize tables of a running tournament
Excluding players without a documented reason and approval
Restricted decision R4: Balance adjustments: any debit (correction, chargeback) and any credit above the tenant's goodwill cap. Decided by: Finance Manager, with maker-checker. Level for agents: 3 for goodwill credits up to the cap; 5 for everything else.

Approval

Level-3 actions run only after approval by: Tournament Manager. Actions: Activate an approved engagement definition.

E · Casino and Sportsbook Operations

Jackpot and Prize Management Agent

AI Jackpot and Prize Controller Assistant

Monitors jackpots, cash prizes, prize spins, raffles and marketplace fulfillment (contributions, drop conditions, held prizes, caps and liabilities). It reconciles and explains prize liabilities for Finance and flags anomalies; releasing, canceling or paying prizes is a human decision.

Human owner: Promotions Finance Controller
Approval level: Starts at 1 Advisory; at most 3 Approval required
Example output: Daily prize liability

Tasks, data and limits

Example tasks

Daily: liability snapshot and reconciliation
A prize is held for review: evidence summary
A jackpot contribution anomaly: alert

Standard outputs

Daily prize liability
Held-prize queue
Monthly prize reconciliation

Data and integrations

Gamification engines
Cash prizes
Bonus engine
Wallet and ledger
Analytics report datasets

Never does

Releasing, canceling or paying cash prizes
Triggering jackpot drops or raffle draws
Changing prize caps
Restricted decision R4: Balance adjustments: any debit (correction, chargeback) and any credit above the tenant's goodwill cap. Decided by: Finance Manager, with maker-checker. Level for agents: 3 for goodwill credits up to the cap; 5 for everything else.

Approval

Level-3 actions run only after approval by: Promotions Finance Controller. Actions: Attach a release or cancel recommendation to a held prize.

E · Casino and Sportsbook Operations

CMS and Content Operations Agent

AI Content Operations Coordinator

Keeps the brand sites' content operations running: page inventory, broken or stale content, menus, banners, pop-ups and expiring promotional content. It drafts fixes as unpublished versions and maintains the publishing checklist; editors publish.

Human owner: CMS Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Weekly content health

Tasks, data and limits

Example tasks

Daily: expiring content scan
Weekly: content health
An offer expires: affected content flagged and fix drafted

Standard outputs

Weekly content health
Publishing checklist

Data and integrations

Site Builder and CMS
Pop-ups and banners
Bonus engine
Tenants and brands

Never does

Publishing or deleting pages, menus, banners or pop-ups
Editing legal pages (terms, privacy, RG) without counsel approval
Restricted decision R7: Legal determinations and legal advice. Decided by: General Counsel or external counsel. Level for agents: 5.

Approval

Level-3 actions run only after approval by: CMS Manager (legal pages: General Counsel). Actions: Publish an approved page version; Restore an earlier page version.

E · Casino and Sportsbook Operations

Multi-Brand Operations Agent

AI Multi-Brand Operations Analyst

Compares configuration and performance across a tenant's brands (compliance settings, offers, lobby, payments, messaging) and flags inconsistencies and brand-pinning errors. It drafts brand-launch checklists and alignment proposals; brand owners decide.

Human owner: Multi-Brand Operations Manager
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Monthly cross-brand configuration audit

Tasks, data and limits

Example tasks

Monthly: configuration audit
A new brand is created: launch checklist

Standard outputs

Monthly cross-brand configuration audit
Brand launch checklist

Data and integrations

Tenants and brands
Compliance configuration
Payments (deposits and withdrawals)
Bonus engine
Casino catalog and integrations
Analytics KPIs and ops summary

Never does

Changing any brand configuration
Copying player data between brands
Any access outside the tenant
Restricted decision R8: Financial-control changes (withdrawal policy and caps, review thresholds, cash-prize caps, fee rules, commission plans, gift-card issuance, risk-rule settings). Decided by: CFO / Finance; tenant super admin for platform settings. Level for agents: 5, except fraud-rule changes proposed through the risk-proposal seam (3, approved by a different person).

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Brand owner.

E · Casino and Sportsbook Operations

Market and Jurisdiction Configuration Agent

AI Jurisdiction Configuration Analyst

Checks each brand's market configuration (restricted countries, age, KYC thresholds, RG limits and tools, game and payment availability, advertising rules) against a counsel-maintained jurisdiction rules library and reports gaps. Configuration changes are compliance decisions made by authorized staff.

Human owner: Compliance Officer
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Market configuration checklist

Tasks, data and limits

Example tasks

Monthly: configuration audit
A regulatory change is logged: impact assessment

Standard outputs

Market configuration checklist
Regulatory change impact assessment

Data and integrations

Compliance configuration
Tenants and brands
Casino catalog and integrations
Payments (deposits and withdrawals)
Jurisdiction and regulatory rules library

Never does

Changing KYC policy, RG settings, restricted countries or availability
Interpreting law as legal advice
Restricted decision R7: Legal determinations and legal advice. Decided by: General Counsel or external counsel. Level for agents: 5.
Restricted decision R8: Financial-control changes (withdrawal policy and caps, review thresholds, cash-prize caps, fee rules, commission plans, gift-card issuance, risk-rule settings). Decided by: CFO / Finance; tenant super admin for platform settings. Level for agents: 5, except fraud-rule changes proposed through the risk-proposal seam (3, approved by a different person).

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Compliance Officer.

F · Player Operations

Customer Support Agent

AI Customer Support Agent

Works the support desk and live chat with the human team: triages and categorizes tickets, drafts accurate replies grounded in the player's account context and approved help content, and hands off to a human whenever money, verification, complaints or responsible gambling are involved. In a later phase, after shadow evaluation, it may answer purely informational live-chat questions directly, always with a visible route to a human.

Human owner: Support Team Lead
Approval level: Starts at 2 Drafting; at most 4 Limited autonomous execution
Example output: Daily desk summary

Tasks, data and limits

Example tasks

Continuously: triage new tickets and chats
Daily: desk summary
A new ticket or chat: categorize and draft

Standard outputs

Daily desk summary
Weekly contact-reason analysis
Deflection and handoff report

Data and integrations

Support desk
Live chat
Player accounts and profiles
Payments (deposits and withdrawals)
Bonus engine
Compliance KYC

Never does

Promising or processing refunds, bonuses, withdrawals or account changes
Disclosing account data before the defined identity check
Handling complaints, RG or self-exclusion requests without handing off
Closing complaints

Approval

Level-3 actions run only after approval by: Support agent on shift. Actions: Send a player-visible reply.

F · Player Operations

Support Quality Assurance Agent

AI Support QA Analyst

Reviews a sample (and over time all) of tickets and chats against the QA scorecard (accuracy, tone, policy adherence, RG and complaint handling) and gives coaching notes to team leads. Scores inform coaching; they are never used on their own for disciplinary decisions.

Human owner: Support QA Lead
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Weekly QA scorecard

Tasks, data and limits

Example tasks

Daily: score the previous day's sample
A severe breach (RG or complaint mishandled): same-day alert

Standard outputs

Weekly QA scorecard
Coaching notes
Policy-breach log

Data and integrations

Support desk
Live chat
Knowledge base (articles, macros, approved answers)
Operators and roles

Never does

Disciplinary recommendations or performance ratings for HR use
Sharing QA scores outside the support management chain
Restricted decision R12: Employee termination and other adverse employment decisions (discipline, pay, rating). Decided by: Head of HR with management. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Support QA Lead.

F · Player Operations

Customer Complaint Resolution Agent

AI Complaints Case Handler Assistant

Manages formal complaints for the complaints team: logs them, assembles the account evidence (bets, transactions, bonus terms, chat history), tracks regulatory response deadlines and drafts responses. Complaint outcomes and any compensation are decided by the complaints manager, and ADR referrals follow the license rules.

Human owner: Complaints Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Complaints register

Tasks, data and limits

Example tasks

Daily: deadline tracker
A ticket is categorized as a complaint: evidence pack and deadline
A deadline is 48 hours away: escalation

Standard outputs

Complaints register
Deadline tracker
Monthly root-cause report

Data and integrations

Support desk
Live chat
Casino reports
Payments (deposits and withdrawals)
Wallet and ledger
Bonus engine

Never does

Deciding complaint outcomes or compensation
Admitting liability
Sending final responses without approval
Closing complaints

Approval

Level-3 actions run only after approval by: Complaints Manager. Actions: Propose a goodwill credit up to the tenant cap.

F · Player Operations

Training and Knowledge Base Agent

AI Knowledge Base Editor

Builds and maintains the help content that support staff and the Customer Support Agent rely on: articles, macros and policy summaries drawn from resolved tickets, product changes and approved policies. Every article is reviewed and approved before use.

Human owner: Support Training Lead
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Knowledge-base coverage and freshness

Tasks, data and limits

Example tasks

Weekly: gap analysis from tickets
Monthly: freshness review
A new release note: affected articles flagged

Standard outputs

Knowledge-base coverage and freshness
Gap list from new contact reasons

Data and integrations

Support desk
Live chat
Operator Academy and platform docs
Site Builder and CMS
Knowledge base (articles, macros, approved answers)

Never does

Publishing articles without approval
Writing policy (it only summarizes approved policy)
Copying personal data from tickets into articles

Approval

Level-3 actions run only after approval by: Support Training Lead (policy wording: Compliance Officer). Actions: Publish an approved article.

F · Player Operations

Payments and Cashier Agent

AI Cashier Operations Analyst

Watches the cashier (deposit and withdrawal flows, method availability, payment outbox and dead letters, crypto valuations and manual adjustments) and explains failures and backlogs. It prepares fixes and retries for approval; it moves no money.

Human owner: Payments Manager
Approval level: Starts at 1 Advisory; at most 3 Approval required
Example output: Daily cashier health

Tasks, data and limits

Example tasks

Every 15 minutes: outbox and stuck-payment scan
Daily: cashier health
A dead-letter item: diagnosis and retry proposal

Standard outputs

Daily cashier health
Stuck-payment list
Weekly payments KPIs

Data and integrations

Payments (deposits and withdrawals)
Wallet and ledger
Payment service providers
Analytics report datasets

Never does

Approving, rejecting or executing payouts
Posting balance adjustments
Changing withdrawal policy, limits or provider settings
Restricted decision R4: Balance adjustments: any debit (correction, chargeback) and any credit above the tenant's goodwill cap. Decided by: Finance Manager, with maker-checker. Level for agents: 3 for goodwill credits up to the cap; 5 for everything else.

Approval

Level-3 actions run only after approval by: Cashier Lead. Actions: Retry or resolve a payments outbox item.

F · Player Operations

Deposit Conversion Agent

AI Deposit Conversion Analyst

Analyzes deposit attempts by method, amount, country and device to find declines, abandonment and slow methods, and proposes routing, method-ordering and messaging improvements. Provider routing and cashier settings are changed by the payments team.

Human owner: Payments Manager
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Weekly deposit conversion

Tasks, data and limits

Example tasks

Daily: success-rate check
A method's success rate drops by more than 10 points: alert

Standard outputs

Weekly deposit conversion
Decline analysis

Data and integrations

Payments (deposits and withdrawals)
Payment service providers
Analytics report datasets

Never does

Changing routing, method availability or limits
Contacting players about failed deposits without CRM approval
Restricted decision R8: Financial-control changes (withdrawal policy and caps, review thresholds, cash-prize caps, fee rules, commission plans, gift-card issuance, risk-rule settings). Decided by: CFO / Finance; tenant super admin for platform settings. Level for agents: 5, except fraud-rule changes proposed through the risk-proposal seam (3, approved by a different person).

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Payments Manager.

F · Player Operations

Withdrawal Review Agent

AI Withdrawal Review Assistant

Prepares every withdrawal that the deterministic auto-approve policy sends to manual review: why the policy declined to auto-approve (its reason codes), the player's KYC level, deposit and withdrawal pattern, bonus status, open risk alerts and prior decisions, plus what to check. Approval, rejection and the second review on high-value withdrawals remain human decisions under the existing maker-checker.

Human owner: Withdrawals Team Lead
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Manual queue summary

Tasks, data and limits

Example tasks

Continuously: summarize each newly queued withdrawal
A withdrawal is routed to manual review: review summary
A second review is required: summary for the checker

Standard outputs

Manual queue summary
Weekly decision-consistency report
SLA report

Data and integrations

Payments (deposits and withdrawals)
Compliance KYC
Wallet and ledger
Bonus engine
Risk desk (alerts, signals, built-in and custom rules, backtests, proposals)
AML cases and configuration

Never does

Approving, rejecting, canceling or delaying any withdrawal
Acting as maker or checker in a two-person review
Contacting the player
Restricted decision R2: Final AML suspicion determinations and SAR/STR filings. Decided by: MLRO. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: the withdrawals reviewers, who make every decision in the cashier.

F · Player Operations

Payment Reconciliation Agent

AI Payment Reconciliation Analyst

Reconciles money across the platform's own ledgers (wallet, payments, game ledger, bonus and prize ledgers) and, once connected, against PSP settlement reports and bank statements, and explains every break. It proposes resolutions for Finance; it posts nothing.

Human owner: Finance Controller
Approval level: Starts at 1 Advisory; at most 3 Approval required
Example output: Daily reconciliation

Tasks, data and limits

Example tasks

Daily 04:00: internal ledger reconciliation
A break above threshold: investigation note

Standard outputs

Daily reconciliation
Break register
Month-end reconciliation pack

Data and integrations

Wallet and ledger
Payments (deposits and withdrawals)
Casino reports
Bonus engine
Cash prizes
PSP settlement reports and bank statements

Never does

Posting adjustments or corrections
Changing ledger data
Restricted decision R4: Balance adjustments: any debit (correction, chargeback) and any credit above the tenant's goodwill cap. Decided by: Finance Manager, with maker-checker. Level for agents: 3 for goodwill credits up to the cap; 5 for everything else.
Restricted decision R8: Financial-control changes (withdrawal policy and caps, review thresholds, cash-prize caps, fee rules, commission plans, gift-card issuance, risk-rule settings). Decided by: CFO / Finance; tenant super admin for platform settings. Level for agents: 5, except fraud-rule changes proposed through the risk-proposal seam (3, approved by a different person).

Approval

Level-3 actions run only after approval by: Finance Controller. Actions: Propose a resolution for a reconciliation break.

F · Player Operations

KYC Case Preparation Agent

AI KYC Case Preparer

Prepares KYC cases in the pending queue for human reviewers: checks document completeness, legibility and consistency with the profile, highlights mismatches and expiry, and drafts the request for missing items. Verification decisions, KYC level changes and the two-person rule for operator uploads remain human.

Human owner: KYC Team Lead
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Queue status with pre-check results

Tasks, data and limits

Example tasks

Continuously: pre-check new submissions
A document is uploaded: pre-check
A document expires within 30 days: refresh request draft

Standard outputs

Queue status with pre-check results
Weekly KYC throughput

Data and integrations

Compliance KYC
Player accounts and profiles
Sanctions and PEP screening

Never does

Approving, rejecting or changing KYC status or levels
Editing identity fields
Storing document images outside the compliance service
Using document data for any other purpose

Approval

Level-3 actions run only after approval by: KYC Team Lead. Actions: Request step-up KYC for a player.

F · Player Operations

AML Case Preparation Agent

AI AML Case Preparer

Assembles AML case files for the MLRO: transaction patterns, source-of-funds indicators, linked accounts, screening results, risk alerts and a chronology, with the typologies they may match. It never decides whether activity is suspicious, never files a SAR, and never tips off the player.

Human owner: MLRO
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Case file per AML case

Tasks, data and limits

Example tasks

Daily: refresh open case files
An AML case opens: case file
A new transaction on a player with an open case: addendum

Standard outputs

Case file per AML case
Weekly AML queue
Monthly typology trends for the MLRO

Data and integrations

AML cases and configuration
Payments (deposits and withdrawals)
Wallet and ledger
Risk desk (alerts, signals, built-in and custom rules, backtests, proposals)
Sanctions and PEP screening
Compliance KYC

Never does

Determining suspicion, closing cases or filing SARs or STRs
Placing or lifting holds
Any player contact about AML matters (tipping-off)
Sharing case data outside the MLRO's team

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: the MLRO, who makes every determination.

F · Player Operations

Fraud and Risk Agent

AI Fraud and Risk Analyst

Watches deposit, bonus, gameplay and tipping patterns against the risk desk's built-in and custom rules, investigates alerts, backtests thresholds and proposes rule changes on Risk Alerts > Proposals for a different operator to approve. It is capped at suggest: its only platform write is a proposal, and alert status, cases, holds and freezes are human-only by route policy.

Human owner: Fraud and Risk Manager
Approval level: Starts at 3 Approval required; at most 3 Approval required
Example output: Daily alert digest

Tasks, data and limits

Example tasks

Every 15 minutes: triage new alerts
Weekly: backtest all active rules
A new alert: summary and recommended next step

Standard outputs

Daily alert digest
Weekly rule performance and proposals
Monthly fraud trends

Data and integrations

Risk desk (alerts, signals, built-in and custom rules, backtests, proposals)
Analytics report datasets
Player accounts and profiles
Wallet and ledger
Affiliate module

Never does

Changing rule settings, alert status or cases directly
Freezing accounts, placing holds or requiring step-up KYC on its own
Contacting players
Approving its own proposals

Approval

Level-3 actions run only after approval by: A different operator with risk-desk rights (finance or super admin). Actions: File a rule proposal (create, tune, enable, disable, retire) with rationale and backtest.

F · Player Operations

Responsible Gaming Agent

AI Responsible Gaming Analyst

Detects markers of harm (escalating deposits, chasing losses, long or late sessions, reversed withdrawals, limit changes, distress in chats) and prepares cases with evidence for the RG team. It assists the RG professional; every contact with the player and every limit, cool-off or exclusion action is decided and taken by a human.

Human owner: Responsible Gaming Officer
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Daily RG marker queue

Tasks, data and limits

Example tasks

Hourly: marker scan
Self-exclusion language in chat or tickets: immediate RG handoff
A deposit spike with late-night sessions: case

Standard outputs

Daily RG marker queue
Monthly RG trends and interventions for the RG officer

Data and integrations

Responsible-gaming data
Analytics report datasets
Payments (deposits and withdrawals)
Support desk
Live chat
Community chat and moderation

Never does

Contacting players, or setting or lifting limits, cool-offs or exclusions
Using RG data for marketing or value scoring
Labeling a player a problem gambler (it reports markers and evidence only)
Restricted decision R1: Player account closure or permanent ban. Decided by: Head of Player Operations with the Compliance Officer. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: the Responsible Gaming Officer, who decides and acts.

F · Player Operations

Chargeback and Dispute Agent

AI Chargeback and Dispute Analyst

Prepares responses to chargebacks and payment disputes (evidence of identity, deposits, gameplay, terms acceptance and communications) within the PSP's deadlines. It drafts representment packs; submissions and any account actions are decided by the payments team.

Human owner: Payments Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Dispute register

Tasks, data and limits

Example tasks

Daily: deadline tracker
A dispute arrives: evidence pack

Standard outputs

Dispute register
Monthly chargeback analysis

Data and integrations

PSP dispute and chargeback feeds
Payments (deposits and withdrawals)
Wallet and ledger
Compliance KYC
Casino reports

Never does

Submitting representments without approval
Closing accounts or blocking players
Posting chargeback adjustments
Restricted decision R1: Player account closure or permanent ban. Decided by: Head of Player Operations with the Compliance Officer. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Payments Manager. Actions: Submit a dispute response to a PSP.

G · Data, Finance and Reporting

CFO and Financial Planning Agent

AI Finance Analyst to the CFO

Gives the CFO a sourced view of revenue, costs, cash and liabilities (player balances, bonus and prize liabilities, affiliate payables, gaming taxes) by combining platform data with the accounting system. It drafts analyses and plans; financial decisions and controls belong to the CFO.

Human owner: CFO
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Monthly flash

Tasks, data and limits

Example tasks

Third business day of the month: flash
A liability ratio breaches its limit: alert

Standard outputs

Monthly flash
Liability statement
Cash forecast

Data and integrations

Analytics report datasets
Wallet and ledger
Accounting / ERP and bank
Budget and plan figures

Never does

Approving payments, journal entries or budgets
Changing financial controls
External financial communications
Restricted decision R3: Regulatory reporting and submissions to regulators. Decided by: Compliance Officer or MLRO. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: CFO.

G · Data, Finance and Reporting

FP&A Agent

AI FP&A Analyst

Builds budget-versus-actual and variance analyses by brand, product and channel, and maintains driver-based forecasts with Finance. It explains variances with evidence and drafts commentary; budgets and reforecasts are approved by Finance leadership.

Human owner: FP&A Manager
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Monthly budget versus actual

Tasks, data and limits

Example tasks

Monthly: budget versus actual
Weekly: forecast refresh
A mid-month variance above 10%: early note

Standard outputs

Monthly budget versus actual
Rolling forecast

Data and integrations

Analytics report datasets
Budget and plan figures
Accounting / ERP and bank
Forecasting engine

Never does

Changing budgets or approving reforecasts
Sharing plan data outside Finance
Restricted decision R8: Financial-control changes (withdrawal policy and caps, review thresholds, cash-prize caps, fee rules, commission plans, gift-card issuance, risk-rule settings). Decided by: CFO / Finance; tenant super admin for platform settings. Level for agents: 5, except fraud-rule changes proposed through the risk-proposal seam (3, approved by a different person).
Restricted decision R13: Material pricing changes (B2B prices and discounts, player fee schedules, global sportsbook margin rules, loyalty and cashback rates). Decided by: CEO/CFO for B2B pricing; Head of Trading for margins; Head of CRM with Finance for loyalty rates. Level for agents: 5 for material changes; 3 for bounded tactical changes inside human-set bands.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: FP&A Manager.

G · Data, Finance and Reporting

Revenue Assurance Agent

AI Revenue Assurance Analyst

Checks that every revenue and cost figure is complete and correct: game rounds against the wallet ledger and provider reports, bonus and free-spin costs, sportsbook settlements, affiliate commissions and fees. It explains leakage and proposes fixes; corrections are approved by Finance.

Human owner: Revenue Assurance Manager
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Monthly revenue assurance report

Tasks, data and limits

Example tasks

Daily: cross-source GGR check
Monthly: revenue assurance report
A cross-source GGR difference above threshold: investigation

Standard outputs

Monthly revenue assurance report
Leakage register

Data and integrations

Casino reports
Wallet and ledger
Bonus-cost reporting
Sportsbook back office
Affiliate module
Analytics report datasets

Never does

Posting corrections or adjustments
Changing commission or fee rules
Restricted decision R4: Balance adjustments: any debit (correction, chargeback) and any credit above the tenant's goodwill cap. Decided by: Finance Manager, with maker-checker. Level for agents: 3 for goodwill credits up to the cap; 5 for everything else.
Restricted decision R8: Financial-control changes (withdrawal policy and caps, review thresholds, cash-prize caps, fee rules, commission plans, gift-card issuance, risk-rule settings). Decided by: CFO / Finance; tenant super admin for platform settings. Level for agents: 5, except fraud-rule changes proposed through the risk-proposal seam (3, approved by a different person).

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Revenue Assurance Manager.

G · Data, Finance and Reporting

Business Intelligence Agent

AI Business Intelligence Analyst

Answers business questions by choosing the right reports from the platform's live dataset catalog and explaining what they show, with 'what to fix' recommendations.

Human owner: BI Lead
Approval level: Starts at 1 Advisory; at most 4 Limited autonomous execution
Example output: On-demand analyses

Tasks, data and limits

Example tasks

Monday: KPI narrative
Daily: anomaly scan
A KPI anomaly: explanation

Standard outputs

On-demand analyses
Weekly KPI narrative
Anomaly explanations

Data and integrations

Analytics report datasets
Analytics KPIs and ops summary

Never does

Inventing metrics, datasets or numbers the report engine did not return
Player-level personal data in answers (aggregates only)
Changing report definitions

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: BI Lead.

G · Data, Finance and Reporting

Data Analyst Agent

AI Data Analyst

Performs multi-step ad-hoc analyses (cohorts, funnels, segment comparisons, experiment readouts) across the report datasets and a read-only query sandbox with tenant and brand filters. It shows its queries and assumptions so analysts can verify every result.

Human owner: BI Lead
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Analysis memos

Tasks, data and limits

Example tasks

On demand
A campaign A/B test completes: readout

Standard outputs

Analysis memos
Experiment readouts

Data and integrations

Analytics report datasets
CRM segments
CRM campaigns
Read-only query sandbox

Never does

Querying outside the tenant and brand scope
Exporting player-level data outside the platform
Writing to any table

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: BI Lead.

G · Data, Finance and Reporting

Player Profitability Agent

AI Player Profitability Analyst

Measures profitability by player segment, product and source (GGR, bonus and free-spin cost, payment fees, affiliate commission) and explains what drives unprofitable segments. Findings guide offer and acquisition policy; they never drive decisions about an individual account.

Human owner: Head of Finance
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Monthly profitability by segment, source and product

Tasks, data and limits

Example tasks

Monthly: profitability pack
A segment turns loss-making: note

Standard outputs

Monthly profitability by segment, source and product

Data and integrations

Analytics report datasets
Bonus-cost reporting
Affiliate module

Never does

Using profitability to restrict an individual player's withdrawals, limits or play (such decisions belong to trading and risk staff)
Showing per-player profitability outside Finance and CRM leadership
Restricted decision R13: Material pricing changes (B2B prices and discounts, player fee schedules, global sportsbook margin rules, loyalty and cashback rates). Decided by: CEO/CFO for B2B pricing; Head of Trading for margins; Head of CRM with Finance for loyalty rates. Level for agents: 5 for material changes; 3 for bounded tactical changes inside human-set bands.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Head of Finance.

G · Data, Finance and Reporting

Bonus Cost Agent

AI Bonus Cost Controller

Tracks bonus, free-spin, cashback, free-bet and prize cost against budget by offer, brand and segment, explains cost spikes and estimates outstanding bonus liability. It is read-only: it recommends pausing or tuning offers, and the bonus team approves.

Human owner: Bonus Manager
Approval level: Starts at 1 Advisory; at most 4 Limited autonomous execution
Example output: Daily bonus cost monitor

Tasks, data and limits

Example tasks

Daily 06:00: cost monitor
Weekly: offer cost ranking
An offer's cost exceeds 150% of forecast: pause recommendation

Standard outputs

Daily bonus cost monitor
Weekly cost by offer and brand
Monthly liability estimate

Data and integrations

Bonus-cost reporting
Analytics report datasets
Bonus engine
VIP and loyalty program
Cash prizes
Budget and plan figures

Never does

Pausing, revoking or changing offers
Changing wagering or contribution rules
Restricted decision R8: Financial-control changes (withdrawal policy and caps, review thresholds, cash-prize caps, fee rules, commission plans, gift-card issuance, risk-rule settings). Decided by: CFO / Finance; tenant super admin for platform settings. Level for agents: 5, except fraud-rule changes proposed through the risk-proposal seam (3, approved by a different person).
Restricted decision R10: High-value promotional liability (offers, prizes or campaigns whose maximum liability exceeds the tenant threshold). Decided by: Head of CRM with Finance. Level for agents: 3 up to the threshold; 5 above it.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Bonus Manager with the Finance Controller.

G · Data, Finance and Reporting

Payment Performance Agent

AI Payment Performance Analyst

Benchmarks payment methods and providers on success rate, speed, cost and chargeback ratio by market. It recommends routing and method-mix changes to the payments team.

Human owner: Payments Manager
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Monthly payment performance

Tasks, data and limits

Example tasks

Monthly: performance report
A provider's success rate drops: alert

Standard outputs

Monthly payment performance

Data and integrations

Payments (deposits and withdrawals)
Payment service providers
PSP settlement reports and bank statements
Analytics report datasets

Never does

Changing routing or provider configuration
Restricted decision R8: Financial-control changes (withdrawal policy and caps, review thresholds, cash-prize caps, fee rules, commission plans, gift-card issuance, risk-rule settings). Decided by: CFO / Finance; tenant super admin for platform settings. Level for agents: 5, except fraud-rule changes proposed through the risk-proposal seam (3, approved by a different person).
Restricted decision R13: Material pricing changes (B2B prices and discounts, player fee schedules, global sportsbook margin rules, loyalty and cashback rates). Decided by: CEO/CFO for B2B pricing; Head of Trading for margins; Head of CRM with Finance for loyalty rates. Level for agents: 5 for material changes; 3 for bounded tactical changes inside human-set bands.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Payments Manager.

G · Data, Finance and Reporting

Affiliate Profitability Agent

AI Affiliate Profitability Analyst

Measures the net value each affiliate brings (referred players' NGR after bonus cost and fees, against commission paid) and models proposed deal changes with the commission-preview tools. It recommends deal reviews; deal terms are decided by affiliate managers and Finance.

Human owner: Head of Affiliates
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Monthly affiliate ROI

Tasks, data and limits

Example tasks

Monthly, after the commission run
An affiliate's ROI is negative two months running: review memo

Standard outputs

Monthly affiliate ROI
Deal simulation memos

Data and integrations

Affiliate module
Analytics report datasets

Never does

Changing deals, fee rules or payouts
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.
Restricted decision R13: Material pricing changes (B2B prices and discounts, player fee schedules, global sportsbook margin rules, loyalty and cashback rates). Decided by: CEO/CFO for B2B pricing; Head of Trading for margins; Head of CRM with Finance for loyalty rates. Level for agents: 5 for material changes; 3 for bounded tactical changes inside human-set bands.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Head of Affiliates with Finance.

G · Data, Finance and Reporting

Revenue Forecasting Agent

AI Revenue Forecaster

Forecasts GGR, NGR, bonus cost and deposits by brand and product for the coming weeks and months, with ranges and the drivers behind them, and tracks forecast accuracy openly. It informs planning; it does not set targets.

Human owner: FP&A Manager
Approval level: Starts at 1 Advisory; at most 4 Limited autonomous execution
Example output: Weekly revenue forecast

Tasks, data and limits

Example tasks

Monday: forecast refresh
Actuals below the lower band 3 days running: shortfall note

Standard outputs

Weekly revenue forecast
Accuracy tracker

Data and integrations

Analytics report datasets
CRM campaigns
Gamification engines
Forecasting engine

Never does

Publishing forecasts externally
Setting budgets or targets

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: FP&A Manager.

G · Data, Finance and Reporting

Billing and Collections Agent

AI Billing and Collections Assistant

Prepares B2B invoices from contract terms and metered usage (license fees, revenue share, sourced-content share, hired agents), reconciles payments and drafts collection reminders. Invoices and collection actions are approved by Finance.

Human owner: Billing Manager
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Monthly billing run

Tasks, data and limits

Example tasks

First business day: billing draft
Weekly: receivables
An invoice is 15 days overdue: reminder draft

Standard outputs

Monthly billing run
Aged receivables

Data and integrations

B2B billing and invoicing
Tenant usage
Contract repository (CLM)
Accounting / ERP and bank

Never does

Issuing invoices, credits or write-offs without approval
Suspending client services
Restricted decision R8: Financial-control changes (withdrawal policy and caps, review thresholds, cash-prize caps, fee rules, commission plans, gift-card issuance, risk-rule settings). Decided by: CFO / Finance; tenant super admin for platform settings. Level for agents: 5, except fraud-rule changes proposed through the risk-proposal seam (3, approved by a different person).
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Billing Manager (credits and write-offs: CFO). Actions: Issue an approved invoice or reminder; Send an external email from the company workspace.

G · Data, Finance and Reporting

Management Reporting Agent

AI Management Reporting Analyst

Produces the recurring management pack (weekly and monthly KPIs by brand and product, with commentary) from the report engine, keeping definitions consistent and every number sourced. It drafts the pack and, once the template is approved, distributes it internally on schedule.

Human owner: BI Lead
Approval level: Starts at 2 Drafting; at most 4 Limited autonomous execution
Example output: Weekly management pack

Tasks, data and limits

Example tasks

Monday 07:00: weekly pack
Second business day: monthly pack
A data refresh fails: delay notice

Standard outputs

Weekly management pack
Monthly management pack

Data and integrations

Analytics report datasets
Analytics KPIs and ops summary
Other agents' published outputs (agent workspace)

Never does

Changing KPI definitions
Distributing outside the approved internal list

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: BI Lead for the template; the COO and CFO receive the pack.

H · Product and Technology

Chief Product Officer Agent

AI Product Strategy Analyst

Synthesizes player feedback, usage data, competitor insight and commercial goals into product strategy options and roadmap trade-offs for the CPO. It drafts roadmap proposals; prioritization decisions are the CPO's.

Human owner: Chief Product Officer
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Quarterly roadmap review

Tasks, data and limits

Example tasks

Quarterly: roadmap review pack
A significant new cluster of contact reasons: product opportunity note

Standard outputs

Quarterly roadmap review
Opportunity assessments

Data and integrations

Support desk
Public API v1
Other agents' published outputs (agent workspace)
Project and issue tracking

Never does

Committing roadmap dates to customers
Re-prioritizing engineering work directly
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.
Restricted decision R13: Material pricing changes (B2B prices and discounts, player fee schedules, global sportsbook margin rules, loyalty and cashback rates). Decided by: CEO/CFO for B2B pricing; Head of Trading for margins; Head of CRM with Finance for loyalty rates. Level for agents: 5 for material changes; 3 for bounded tactical changes inside human-set bands.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Chief Product Officer.

H · Product and Technology

Product Manager Agent

AI Product Manager Assistant

Turns problems into well-formed product specs (problem statement, user stories, acceptance criteria, success metrics and edge cases) grounded in data and support evidence. Product managers own scope and decisions.

Human owner: Head of Product
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Spec drafts

Tasks, data and limits

Example tasks

Before sprint planning: spec readiness check
A new feature request: spec draft

Standard outputs

Spec drafts
Feature success reviews

Data and integrations

Project and issue tracking
Email, calendar and documents
Support desk
Operator Academy and platform docs

Never does

Changing priorities or committing scope
Publishing specs externally
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Product manager. Actions: Update issues and project fields.

H · Product and Technology

Product Owner Agent

AI Product Owner Assistant

Keeps the backlog healthy: refines stories, checks acceptance criteria, merges duplicate tickets, prepares sprint candidates and tracks delivery against the sprint goal. The product owner accepts work and sets priorities.

Human owner: Product Owner
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Backlog health

Tasks, data and limits

Example tasks

Twice weekly: backlog refinement draft
A duplicate ticket is created: merge proposal

Standard outputs

Backlog health
Sprint readiness

Data and integrations

Project and issue tracking
Code repository and CI/CD

Never does

Accepting stories or changing priorities
Closing tickets without approval
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Product Owner. Actions: Update issues and project fields.

H · Product and Technology

UX Research Agent

AI UX Research Assistant

Synthesizes qualitative and quantitative evidence (support contacts, complaints, funnels, surveys and usability notes) into user problems ranked by severity and frequency. It plans studies and drafts discussion guides; researchers run sessions with consenting players.

Human owner: UX Research Lead
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Monthly user problems report

Tasks, data and limits

Example tasks

Monthly: user problems report
A new complaint theme: problem note

Standard outputs

Monthly user problems report
Study syntheses

Data and integrations

Support desk
Live chat
Analytics report datasets
Surveys and research tooling
Public API v1

Never does

Contacting players for research without consent and approval
Using personal data beyond the study's consent

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: UX Research Lead.

H · Product and Technology

UX/UI Design Agent

AI UX/UI Design Assistant

Produces design variants, UI copy and accessibility checks against the brand's design system, and reviews screens for consistency, visibility of responsible-gambling tools and mobile usability. Designers own final designs.

Human owner: Head of Design
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Design reviews

Tasks, data and limits

Example tasks

Each release: accessibility check
A new screen enters review: design-system check

Standard outputs

Design reviews
Accessibility audit

Data and integrations

Design tools and brand asset library
Site Builder and CMS
Operator Academy and platform docs

Never does

Shipping UI changes
Dark patterns
Hiding responsible-gambling tools or terms

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Design lead.

H · Product and Technology

Engineering Manager Agent

AI Engineering Manager Assistant

Gives engineering managers a view of delivery flow (cycle time, review latency, CI health, incident load and on-call fatigue) and drafts sprint reports and capacity plans. It does not assess individuals for performance decisions.

Human owner: Engineering Manager
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Weekly delivery report

Tasks, data and limits

Example tasks

Weekly: delivery report
The CI failure rate spikes: note

Standard outputs

Weekly delivery report
Monthly engineering health

Data and integrations

Code repository and CI/CD
Project and issue tracking
Cloud and infrastructure monitoring

Never does

Individual performance ratings
Merging, deploying or changing branch protections
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.
Restricted decision R12: Employee termination and other adverse employment decisions (discipline, pay, rating). Decided by: Head of HR with management. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Engineering Manager.

H · Product and Technology

Software Architecture Agent

AI Software Architect Assistant

Reviews designs and changes for architectural consistency (service boundaries, tenant and brand isolation, idempotency of money paths, data contracts and security patterns) and drafts architecture decision records. Architects decide.

Human owner: CTO
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Design review notes

Tasks, data and limits

Example tasks

On each design submission
A change touches a money path or tenant scoping: mandatory review note

Standard outputs

Design review notes
ADR drafts

Data and integrations

Code repository and CI/CD
Email, calendar and documents
Operator Academy and platform docs

Never does

Approving designs or merging code
Changing infrastructure
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: CTO or lead architect.

H · Product and Technology

Front-End Development Agent

AI Front-End Developer

Implements and fixes front-end changes (player site, back office, PWA) in branches with tests and screenshots, following the design system and accessibility rules. Every change goes through human code review and CI; it never deploys.

Human owner: Front-End Lead
Approval level: Starts at 3 Approval required; at most 4 Limited autonomous execution
Example output: Pull-request summaries

Tasks, data and limits

Example tasks

On ticket assignment
A ticket is assigned: branch and pull request

Standard outputs

Pull-request summaries

Data and integrations

Code repository and CI/CD
Project and issue tracking
Design tools and brand asset library

Never does

Merging to protected branches
Deploying
Accessing production data or secrets
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Code reviewer (a human engineer). Actions: Push a branch and open a pull request for human review.

H · Product and Technology

Back-End Development Agent

AI Back-End Developer

Implements service changes (APIs, engines, migrations) in branches with contract tests, respecting tenant and brand scoping, idempotency and the migration rules (for example, never deleting money rows in a migration). Every change goes through human code review and CI; it never deploys or runs migrations against shared databases.

Human owner: Back-End Lead
Approval level: Starts at 3 Approval required; at most 4 Limited autonomous execution
Example output: Pull-request summaries

Tasks, data and limits

Example tasks

On ticket assignment
A ticket is assigned: branch and pull request

Standard outputs

Pull-request summaries

Data and integrations

Code repository and CI/CD
Project and issue tracking

Never does

Merging to protected branches
Deploying or running migrations against shared databases
Accessing production data or secrets
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Code reviewer (a human engineer). Actions: Push a branch and open a pull request for human review.

H · Product and Technology

Full-Stack Development Agent

AI Full-Stack Developer

Delivers small end-to-end features across the BFF, services and UI in one branch, with tests at each layer and screenshots. Every change goes through human code review and CI; it never deploys.

Human owner: Engineering Manager
Approval level: Starts at 3 Approval required; at most 4 Limited autonomous execution
Example output: Pull-request summaries

Tasks, data and limits

Example tasks

On ticket assignment
A ticket is assigned: branch and pull request

Standard outputs

Pull-request summaries

Data and integrations

Code repository and CI/CD
Project and issue tracking
Design tools and brand asset library

Never does

Merging to protected branches
Deploying
Accessing production data or secrets
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Code reviewer (a human engineer). Actions: Push a branch and open a pull request for human review.

H · Product and Technology

Database Agent

AI Database Engineer Assistant

Reviews migrations and queries for safety and performance (locks, indexes, partitioning, row-level tenant and brand predicates, immutability of money rows) and drafts migration plans and query optimizations. Migrations run only through CI/CD with human approval.

Human owner: Data Engineering Lead
Approval level: Starts at 1 Advisory; at most 3 Approval required
Example output: Migration review notes

Tasks, data and limits

Example tasks

On each migration pull request
A slow query appears in production logs: optimization draft

Standard outputs

Migration review notes
Monthly query performance

Data and integrations

Code repository and CI/CD
Cloud and infrastructure monitoring

Never does

Running DDL or DML on shared or production databases
Deleting or rewriting money or audit rows
Accessing production personal data
Restricted decision R4: Balance adjustments: any debit (correction, chargeback) and any credit above the tenant's goodwill cap. Decided by: Finance Manager, with maker-checker. Level for agents: 3 for goodwill credits up to the cap; 5 for everything else.

Approval

Level-3 actions run only after approval by: Database reviewer (a human engineer). Actions: Push a branch and open a pull request for human review.

H · Product and Technology

API and Integration Agent

AI Integration Operations Analyst

Monitors the tenant's integrations (outbound signed webhooks, public-API usage and idempotency conflicts, game aggregator and studio connections), diagnoses failures and retries failed webhook deliveries within limits. Credential, endpoint and secret changes stay with humans.

Human owner: Integration Lead
Approval level: Starts at 1 Advisory; at most 4 Limited autonomous execution
Example output: Daily integration health

Tasks, data and limits

Example tasks

Every 10 minutes: failed-delivery scan
A delivery fails repeatedly: diagnosis, then a capped retry
Idempotency conflicts spike: note

Standard outputs

Daily integration health
Webhook failure analysis

Data and integrations

Outbound signed webhooks
Public API v1
Casino catalog and integrations
Tenant usage
Service health

Never does

Creating, editing or deleting webhook endpoints
Rotating secrets or API keys
Changing integration credentials
Restricted decision R15: Access control and secrets (operator roles and module scopes, API keys, webhook secrets, any agent's permissions, autonomy or budget). Decided by: Tenant super admin. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Integration Lead.

H · Product and Technology

QA and Test Automation Agent

AI QA Engineer

Writes and maintains automated tests (unit, contract and end-to-end), triages flaky tests and failing CI runs, and drafts test plans from specs. It runs only in non-production environments.

Human owner: QA Lead
Approval level: Starts at 3 Approval required; at most 4 Limited autonomous execution
Example output: Test plan drafts

Tasks, data and limits

Example tasks

Nightly: flaky-test triage
CI fails on the main branch: triage note

Standard outputs

Test plan drafts
Weekly CI health

Data and integrations

Code repository and CI/CD
Project and issue tracking

Never does

Disabling tests or CI gates to make builds pass
Running tests against production
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Code reviewer (a human engineer). Actions: Push a branch and open a pull request for human review.

H · Product and Technology

Release Management Agent

AI Release Coordinator

Prepares releases: change lists, migration checks, release notes, rollback plans and go/no-go checklists, and tracks health after release. Production deployment is approved and executed by authorized people through the CI/CD gate.

Human owner: Release Manager
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Release checklist

Tasks, data and limits

Example tasks

Each release: checklist and notes
Health degrades after a release: rollback recommendation to the Release Manager

Standard outputs

Release checklist
Release notes draft
Post-release health report

Data and integrations

Code repository and CI/CD
Service health
Operator Academy and platform docs

Never does

Deploying to production or merging to the production branch
Skipping CI gates or approvals
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Release Manager.

H · Product and Technology

DevOps and SRE Agent

AI SRE Assistant

Watches service health, error rates, latency, queues and resource usage, correlates alerts and drafts runbook steps and post-incident timelines. Restarts, scaling and configuration changes in production are human actions.

Human owner: SRE Lead
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Weekly reliability report

Tasks, data and limits

Example tasks

Continuously: alert correlation
Weekly: reliability report
Correlated alerts: probable cause and runbook step

Standard outputs

Weekly reliability report
Post-incident timelines

Data and integrations

Service health
Cloud and infrastructure monitoring
Code repository and CI/CD

Never does

Restarting, scaling or reconfiguring production infrastructure
Accessing secrets
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.
Restricted decision R15: Access control and secrets (operator roles and module scopes, API keys, webhook secrets, any agent's permissions, autonomy or budget). Decided by: Tenant super admin. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: SRE Lead.

H · Product and Technology

Cloud Infrastructure Agent

AI Cloud Infrastructure Assistant

Reviews infrastructure cost, capacity, backups and configuration drift, and drafts infrastructure-as-code changes for review. Changes are applied by authorized engineers.

Human owner: Cloud Lead
Approval level: Starts at 1 Advisory; at most 3 Approval required
Example output: Monthly cost and capacity review

Tasks, data and limits

Example tasks

Monthly: cost review
Weekly: backup check
A backup fails: alert

Standard outputs

Monthly cost and capacity review
Backup verification report

Data and integrations

Cloud and infrastructure monitoring
Code repository and CI/CD

Never does

Applying infrastructure changes
Accessing secrets or production data
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.
Restricted decision R15: Access control and secrets (operator roles and module scopes, API keys, webhook secrets, any agent's permissions, autonomy or budget). Decided by: Tenant super admin. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Cloud Lead. Actions: Push a branch and open a pull request for human review.

H · Product and Technology

Security Operations Agent

AI Security Analyst

Monitors security signals (operator privilege changes, unusual bulk actions in the audit log, API-key usage, webhook secret age, failed authentications) and drafts investigation notes and hardening recommendations. Access changes and incident decisions are made by humans.

Human owner: Security Lead
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Weekly security signals report

Tasks, data and limits

Example tasks

Continuously: audit-log anomaly scan
Weekly: hygiene report
A bulk export or privilege change outside normal patterns: investigation note

Standard outputs

Weekly security signals report
Hygiene report (keys, secrets, roles)

Data and integrations

Audit log
Operators and roles
Tenants and brands
Outbound signed webhooks
Security tooling

Never does

Disabling accounts, revoking keys or changing roles
Accessing secret values
Running intrusive scans without authorization
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Security Lead.

H · Product and Technology

Incident Response Agent

AI Incident Coordinator

During incidents, assembles the timeline from alerts, deploys and logs, drafts status updates and stakeholder notes, tracks actions and prepares the post-incident review. The incident commander makes decisions, and communications to players and regulators are approved by humans.

Human owner: Incident Commander (CTO)
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Incident timeline

Tasks, data and limits

Example tasks

During incidents: status update drafts at the agreed cadence
An incident is declared: open the timeline and task list

Standard outputs

Incident timeline
Status update drafts
Post-incident review

Data and integrations

Service health
Cloud and infrastructure monitoring
Code repository and CI/CD
Other agents' published outputs (agent workspace)

Never does

Executing remediation in production
Sending external or regulatory notifications
Restricted decision R3: Regulatory reporting and submissions to regulators. Decided by: Compliance Officer or MLRO. Level for agents: 5.
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Incident commander.

H · Product and Technology

Data Engineering Agent

AI Data Engineer Assistant

Maintains data pipelines and datasets: freshness checks, impact of schema changes, data-quality tests and documentation of the report datasets. Pipeline changes go through code review.

Human owner: Data Engineering Lead
Approval level: Starts at 1 Advisory; at most 3 Approval required
Example output: Data-quality report

Tasks, data and limits

Example tasks

Daily: freshness and quality checks
A schema change touches a report dataset: impact note

Standard outputs

Data-quality report
Dataset documentation

Data and integrations

Analytics report datasets
Code repository and CI/CD
Cloud and infrastructure monitoring

Never does

Changing production pipelines or schemas directly
Copying personal data to non-approved stores
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Data Engineering Lead. Actions: Push a branch and open a pull request for human review.

H · Product and Technology

AI and Machine Learning Agent

AI ML Operations Assistant

Supports the lifecycle of the platform's predictive models (churn, value, fraud scoring) with feature checks, drift monitoring, backtests, fairness checks and model cards, and keeps the agent evaluation metrics (shadow agreement, acceptance and override rates). Model promotion and changes to any agent's autonomy are approved by humans.

Human owner: Head of AI/ML
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Model cards

Tasks, data and limits

Example tasks

Weekly: drift check
Monthly: agent evaluation report
A model's drift crosses its threshold: alert

Standard outputs

Model cards
Monthly drift report
Agent evaluation report

Data and integrations

Model training and registry
Analytics report datasets
Other agents' published outputs (agent workspace)

Never does

Promoting models to production
Changing any agent's autonomy, permissions or budget
Training on another tenant's data
Restricted decision R9: Production deployment. Decided by: Release Manager / CTO through CI/CD. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Head of AI/ML.

H · Product and Technology

Technical Documentation Agent

AI Technical Writer

Drafts and updates technical documentation (API references, integration guides, release notes, runbooks and operator Academy articles) from code changes, specs and support questions. Documentation is reviewed before publication.

Human owner: Documentation Lead
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Documentation drafts

Tasks, data and limits

Example tasks

Each release: docs update drafts
An API change is merged: reference update draft

Standard outputs

Documentation drafts
Documentation coverage

Data and integrations

Code repository and CI/CD
Operator Academy and platform docs
Support desk
Public API v1

Never does

Publishing documentation
Exposing internal security details or secrets

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Documentation Lead.

I · Legal, Compliance and Corporate Administration

Legal and Contract Review Agent

AI Legal Assistant (Contract Review)

Reviews contracts against the company's playbook (liability, IP, data protection, termination, governing law and gaming-specific clauses) and produces issue lists and redline suggestions for counsel. It gives no legal advice or determinations; counsel decides.

Human owner: General Counsel
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Issue lists

Tasks, data and limits

Example tasks

On upload of a contract for review
A contract is uploaded for review: issue list

Standard outputs

Issue lists
Contract register summary

Data and integrations

Contract repository (CLM)
Email, calendar and documents
Policy and procedure repository

Never does

Legal determinations or advice
Sending redlines to counterparties
Signing or accepting terms
Restricted decision R7: Legal determinations and legal advice. Decided by: General Counsel or external counsel. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Legal counsel.

I · Legal, Compliance and Corporate Administration

Privacy and Data Protection Agent

AI Privacy Analyst

Supports the DPO: maintains the record of processing, drafts DPIAs (including one for every AI agent before go-live), tracks data-subject requests and their deadlines, and checks that agents and features minimize personal data. Decisions on lawful basis, breaches and DSAR responses are made by the DPO.

Human owner: Data Protection Officer
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: DSAR tracker

Tasks, data and limits

Example tasks

Daily: DSAR deadline tracker
Quarterly: record-of-processing review
A DSAR arrives: data inventory and draft response

Standard outputs

DSAR tracker
DPIA drafts
Record of processing

Data and integrations

Player accounts and profiles
Data-subject request tooling
Policy and procedure repository
Operators and roles
Audit log

Never does

Deciding breach notification or lawful basis
Releasing personal data to requesters without DPO approval
Erasing data
Restricted decision R3: Regulatory reporting and submissions to regulators. Decided by: Compliance Officer or MLRO. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Data Protection Officer. Actions: Compile a player's data export for DPO review.

I · Legal, Compliance and Corporate Administration

Regulatory Readiness Agent

AI Regulatory Analyst

Tracks regulatory requirements for each license and market, maps them to platform controls and evidence, and reports readiness gaps ahead of audits. It informs the Compliance Officer; it makes no legal determinations and submits nothing to regulators.

Human owner: Compliance Officer
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Readiness dashboard

Tasks, data and limits

Example tasks

Monthly: readiness review
A regulatory update: impact assessment

Standard outputs

Readiness dashboard
Gap register

Data and integrations

Jurisdiction and regulatory rules library
Compliance configuration
Responsible-gaming data
Audit log
Analytics report datasets

Never does

Regulatory submissions or correspondence
Presenting legal interpretations as determinations
Restricted decision R3: Regulatory reporting and submissions to regulators. Decided by: Compliance Officer or MLRO. Level for agents: 5.
Restricted decision R7: Legal determinations and legal advice. Decided by: General Counsel or external counsel. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Compliance Officer.

I · Legal, Compliance and Corporate Administration

Licensing Preparation Agent

AI Licensing Coordinator

Prepares license applications and renewals: document checklists, corporate and ownership information, policies, technical system descriptions and certification evidence, while tracking every regulator deadline. Submissions and declarations are made by authorized officers.

Human owner: Licensing Manager
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Application checklist

Tasks, data and limits

Example tasks

Weekly: deadline scan
A regulator requests information: response pack draft

Standard outputs

Application checklist
Deadline calendar

Data and integrations

Regulator licensing portals
Email, calendar and documents
Policy and procedure repository
Operator Academy and platform docs

Never does

Submitting applications or signing declarations
Representing the company to regulators
Restricted decision R3: Regulatory reporting and submissions to regulators. Decided by: Compliance Officer or MLRO. Level for agents: 5.
Restricted decision R7: Legal determinations and legal advice. Decided by: General Counsel or external counsel. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Licensing Manager with General Counsel.

I · Legal, Compliance and Corporate Administration

Compliance Monitoring Agent

AI Compliance Monitoring Analyst

Runs continuous control testing on the platform: KYC backlog and KYC-level coverage before withdrawals, RG tool usage, self-exclusion enforcement, AML case ageing, sanctions-check coverage, marketing suppression and audit-log completeness. It reports exceptions to the Compliance Officer and changes no controls.

Human owner: Compliance Officer
Approval level: Starts at 1 Advisory; at most 2 Drafting
Example output: Daily control exceptions

Tasks, data and limits

Example tasks

Daily 06:00: control tests
A withdrawal paid below the required KYC level: critical exception
A message sent to a suppressed player: critical exception

Standard outputs

Daily control exceptions
Monthly compliance monitoring report

Data and integrations

Compliance KYC
Responsible-gaming data
AML cases and configuration
Sanctions and PEP screening
CRM campaigns
Payments (deposits and withdrawals)

Never does

Changing compliance settings, holds or cases
Regulatory reporting
Restricted decision R3: Regulatory reporting and submissions to regulators. Decided by: Compliance Officer or MLRO. Level for agents: 5.
Restricted decision R8: Financial-control changes (withdrawal policy and caps, review thresholds, cash-prize caps, fee rules, commission plans, gift-card issuance, risk-rule settings). Decided by: CFO / Finance; tenant super admin for platform settings. Level for agents: 5, except fraud-rule changes proposed through the risk-proposal seam (3, approved by a different person).

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Compliance Officer.

I · Legal, Compliance and Corporate Administration

Policy Management Agent

AI Policy Coordinator

Maintains the policy library (AML, RG, KYC, privacy, security, complaints, bonus terms) with owners, versions and review dates, compares policies with the platform's configuration, and drafts updates for approval. It never approves or publishes policy.

Human owner: Compliance Officer
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Policy register

Tasks, data and limits

Example tasks

Monthly: review-date scan
A regulatory change affects a policy: update draft

Standard outputs

Policy register
Mismatch report

Data and integrations

Policy and procedure repository
Compliance configuration
Jurisdiction and regulatory rules library

Never does

Approving or publishing policies
Changing platform settings to match a draft policy
Restricted decision R7: Legal determinations and legal advice. Decided by: General Counsel or external counsel. Level for agents: 5.
Restricted decision R8: Financial-control changes (withdrawal policy and caps, review thresholds, cash-prize caps, fee rules, commission plans, gift-card issuance, risk-rule settings). Decided by: CFO / Finance; tenant super admin for platform settings. Level for agents: 5, except fraud-rule changes proposed through the risk-proposal seam (3, approved by a different person).

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Policy owner, then the Compliance Officer.

I · Legal, Compliance and Corporate Administration

Human Resources Agent

AI HR Coordinator

Answers employees' HR questions from approved policies, prepares HR documents and tracks HR processes (leave, contract renewals, probation dates). Employment decisions are made by HR and management, and it never handles disciplinary or termination decisions.

Human owner: Head of HR
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: HR process calendar

Tasks, data and limits

Example tasks

Weekly: process date scan
A probation end is 30 days away: task for the manager

Standard outputs

HR process calendar
Question themes

Data and integrations

HRIS and payroll
Policy and procedure repository
Email, calendar and documents

Never does

Employment decisions (hiring, discipline, termination, pay)
Sharing employee personal data
Legal advice on employment law
Restricted decision R7: Legal determinations and legal advice. Decided by: General Counsel or external counsel. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: HR partner.

I · Legal, Compliance and Corporate Administration

Recruitment Agent

AI Recruiting Coordinator

Drafts job descriptions, screens applications against stated must-have criteria, schedules interviews and keeps candidates informed. Shortlists are reviewed by recruiters and hiring managers, and it never rejects a candidate on its own.

Human owner: Talent Acquisition Lead
Approval level: Starts at 2 Drafting; at most 3 Approval required
Example output: Pipeline report

Tasks, data and limits

Example tasks

Daily: new application screening summaries
A new application: screening summary against must-haves

Standard outputs

Pipeline report
Screening fairness check

Data and integrations

Applicant tracking system
Email, calendar and documents

Never does

Rejecting candidates or making offers
Using protected characteristics or their proxies
Automated decisions with legal effect on candidates
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.

Approval

Level-3 actions run only after approval by: Recruiter. Actions: Create internal calendar holds and internal reminders; Send an external email from the company workspace.

I · Legal, Compliance and Corporate Administration

Employee Onboarding Agent

AI Onboarding Coordinator

Runs new-starter onboarding: checklists, account-request drafts, training plans (including the operator Academy and mandatory compliance training) and check-ins during the first weeks. Access is granted by the owners of each system.

Human owner: People Operations Lead
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Onboarding tracker

Tasks, data and limits

Example tasks

Daily: onboarding task reminders
A new hire is created: onboarding plan and access-request drafts

Standard outputs

Onboarding tracker

Data and integrations

HRIS and payroll
Operator Academy and platform docs
Learning management system
Email, calendar and documents

Never does

Granting system access or roles
Collecting more personal data than onboarding requires
Restricted decision R15: Access control and secrets (operator roles and module scopes, API keys, webhook secrets, any agent's permissions, autonomy or budget). Decided by: Tenant super admin. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: People Operations Lead.

I · Legal, Compliance and Corporate Administration

Performance Management Agent

AI Performance Review Assistant

Supports the performance cycle with reminders, goal-tracking summaries and review templates drafted from inputs provided by managers and employees. It never rates people or recommends pay, promotion or termination.

Human owner: Head of HR
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Cycle progress

Tasks, data and limits

Example tasks

During the cycle: weekly reminders
A review is overdue: reminder draft

Standard outputs

Cycle progress

Data and integrations

HRIS and payroll
Email, calendar and documents

Never does

Rating employees
Pay, promotion or termination recommendations
Using monitoring data (QA scores, productivity) as sole evidence
Restricted decision R12: Employee termination and other adverse employment decisions (discipline, pay, rating). Decided by: Head of HR with management. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: The employee's manager.

I · Legal, Compliance and Corporate Administration

Training and Development Agent

AI Learning and Development Coordinator

Plans and tracks staff training (mandatory AML, RG and data-protection training, product training on the operator Academy, and role skills) and drafts learning content for review. It reports completion to managers.

Human owner: Learning and Development Lead
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Training compliance report

Tasks, data and limits

Example tasks

Monthly: completion report
A new feature ships: Academy update draft

Standard outputs

Training compliance report
Content review calendar

Data and integrations

Operator Academy and platform docs
Learning management system
HRIS and payroll

Never does

Certifying staff as trained without evidence
Publishing training content without approval

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Learning and Development Lead (compliance content: Compliance Officer).

I · Legal, Compliance and Corporate Administration

Executive Assistant Agent

AI Executive Assistant

Manages an executive's calendar, inbox triage, travel and meeting preparation under explicit delegation rules. It drafts replies and holds time; anything sent externally in the executive's name needs approval unless the executive has allow-listed that type of message.

Human owner: The executive served
Approval level: Starts at 2 Drafting; at most 4 Limited autonomous execution
Example output: Daily agenda brief

Tasks, data and limits

Example tasks

Daily 07:00: agenda brief
A meeting request arrives: proposed slot and brief

Standard outputs

Daily agenda brief
Weekly follow-ups

Data and integrations

Email, calendar and documents
Other agents' published outputs (agent workspace)

Never does

Sending external messages without approval (except allow-listed types)
Accepting commitments
Opening confidential HR or board folders unless delegated
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.

Approval

Level-3 actions run only after approval by: The executive served. Actions: Send an external email from the company workspace.

I · Legal, Compliance and Corporate Administration

Administrative Operations Agent

AI Office and Administration Coordinator

Handles corporate administration: entity records, registered-office and filing calendars, intake of supplier invoices, office logistics and document filing. Statutory filings and payments are executed by authorized people.

Human owner: Office and Administration Manager
Approval level: Starts at 2 Drafting; at most 2 Drafting
Example output: Filing calendar

Tasks, data and limits

Example tasks

Weekly: filing deadline scan
A supplier invoice arrives: route for approval

Standard outputs

Filing calendar
Invoice intake log

Data and integrations

Email, calendar and documents
Accounting / ERP and bank
Procurement and vendor register

Never does

Making payments or statutory filings
Signing documents
Restricted decision R3: Regulatory reporting and submissions to regulators. Decided by: Compliance Officer or MLRO. Level for agents: 5.
Restricted decision R11: Contract execution (signing, accepting terms, committing the company). Decided by: Authorized signatory. Level for agents: 5.

Approval

No level-3 actions: nothing the agent produces executes on its own. Outputs are reviewed by: Company secretary for filings; Finance for invoices.

Which agents would you hire first?

Build your package from the departments that matter most, and add agents as you grow.

Book a 20-minute live demo AI Workforce pricing

AI Agent Directory

Agents

CEO and Strategy Agent

Example tasks

Standard outputs

Data and integrations

Never does

Approval

COO and Operations Agent

Example tasks

Standard outputs

Data and integrations

Never does

Approval

Chief of Staff Agent

Example tasks

Standard outputs

Data and integrations

Never does

Approval

Project Management and PMO Agent

Example tasks

Standard outputs

Data and integrations

Never does

Approval

Business Forecasting Agent

Example tasks

Standard outputs

Data and integrations

Never does

Approval

Board and Investor Reporting Agent

Example tasks

Standard outputs

Data and integrations

Never does

Approval

Market Expansion Agent

Example tasks

Standard outputs

Data and integrations

Never does

Approval

CMO and Marketing Strategy Agent

Example tasks

Standard outputs

Data and integrations

Never does

Approval

Acquisition Strategy Agent

Example tasks

Standard outputs

Data and integrations

Never does

Approval

Paid Media Agent

Example tasks

Standard outputs

Data and integrations

Never does

Approval

Affiliate Management Agent

Example tasks

Standard outputs

Data and integrations

Never does

Approval

SEO Strategy Agent

Example tasks

Standard outputs

Data and integrations

Never does

Approval

Content Marketing Agent

Example tasks

Standard outputs

Data and integrations

Never does

Approval