← AI Transformation Suite Roles & Responsibilities

Roles & Responsibilities

Download Word

1. Why This Document Exists

At 262 people across 25 teams, no individual can know what the program is doing. That is not a failure of management attention; it is arithmetic. A program of this size either substitutes structure for personal knowledge or it substitutes optimism, and the second option is how AI programs put an unvalidated model in front of a member.

So this document is not a directory. It states what each function owns, and it states the thing an org chart actively obscures: which decisions have been deliberately placed beyond the reach of the person accountable for delivery.

The design principle that governs everything below. ⚠⚠ No model reaches production without clean Independent Model Validation sign-off AND AI Governance Board approval — and that requirement cannot be overridden by schedule pressure, by an Executive Sponsor request, or by any other governance body. ***The Program Director cannot approve his own work, and neither can the Executive Sponsor. That is not an oversight in the authority structure; it is the authority structure.***

2. The Shape of the Organization

Four authority levels, and the most informative thing about them is where the Program Director is not.

LevelAuthorityHeld byDecides
1StrategicM. Kavanagh, ACME COO (Executive Sponsor)Program viability, budget re-baseline, regulatory escalation, board deadlock, contingency >$1M
2GovernanceThree boards — see §3Phase-gate sign-off, model production release, architecture approval, changes beyond Program Director authority
3ProgramC. Tyrrell, Program DirectorDay-to-day delivery coordination, schedule movement within ±2 weeks, resource changes
4ExecutionTeam and BRD leadsTask sequencing and method within their own scope
Read Level 3 carefully. On a $69M, three-year, 262-person program the Program Director's unilateral schedule authority is two weeks. Everything beyond that is a board decision. ***That is not a diminished role — it is what accountability looks like when the failure mode is a bad model in production rather than a late release.***

3. The Three Boards — and Why There Are Three

One board would be simpler and would concentrate exactly the authority this design is trying to separate. The Program Director chairs none of them. He assembles evidence, submits gate packages, and receives decisions.

BoardChairOwnsBlocking power
Executive Steering BoardM. Kavanagh (COO)Phase-gate sign-off, scope for Years 2 and 3, regulatory and reputational riskA phase gate without ESB approval does not advance.
AI Governance BoardS. Khurana (AI Governance Director)Model production release, fairness and bias posture, clinical and privacy acceptabilityNo model to production without its approval.
Enterprise Architecture Review BoardD. ChenTechnical fit, platform standards, security sign-offNo architecture change without EARB sign-off.

No board may overrule another. Where two disagree, the conflict escalates to Level 1 for resolution rather than being settled by whichever chair is more senior or more insistent. That rule is what stops a three-board structure collapsing into one board with two advisory committees.

4. The Functions That Exist To Be Able To Say No

Four of the 25 teams are structurally independent of delivery. They are worth naming together, because their independence is the mechanism rather than a courtesy.

FunctionLeadIndependence mechanism
Independent Model Validation (6)P. OkaforReports through a different authority chain than the BRD delivery teams — it does not report to the Program Director, and its sign-off is a precondition, not an input
Data Privacy Office (5)E. Sato, Chief Privacy OfficerSits on the AI Governance Board; privacy determinations are not delivery trade-offs
SOX / Financial Controls (6)Control testing scope is not set by the program
Clinical / Medical Policy (6)Dr. N. Patel, CMOClinical acceptability is a stop, not a factor to be balanced against velocity
The test used throughout this portfolio: if the program can change the answer by asking harder, it was never an independent function. ⚠ Independent Model Validation is the sharpest instance in the entire portfolio — it is the only function whose reporting line was deliberately routed away from the program in order to protect its ability to refuse.

5. The Delivery Organization

The remaining teams do the work. Three BRDs carry the AI capability itself; the rest are the platform, quality and enablement functions that any program of this size needs.

GroupingTeamsOwnsDoes not own
Program Leadership & PMO16 peopleIntegrated plan, reporting, gate packages, dependency managementAny gate decision
The three BRDsClaims & Prior Auth; Member/Provider UX; Underwriting & RiskRequirements, build, and the model within their domainWhether their own model is fit for production
Data & Cloud AI Platform20 peopleThe platform every BRD builds onBRD-level scope
QA & Testing18 peopleTest strategy, execution, quality evidence at each gateThe gate decision itself
Cybersecurity, IT Ops, Enterprise Architecture28 peopleSecurity posture, environments, technical standards
Change, Training, Comms, UX, HR26 peopleAdoption, readiness, workforce impactWhether the business is ready — they report it, they cannot create it

A BRD team does not certify its own model. It builds it, evidences it, and hands it to a validation function it does not control. ***This is the single most transferable idea in the suite: the people who build a model are the worst-placed people to judge whether it should ship, and no amount of integrity fixes that — only structure does.***

6. Who Decides What

DecisionRecommendsDecidesCan be overridden by
Task sequencing within a teamTeam leadProgram Director
Schedule movement ≤ 2 weeksTeam leadsProgram DirectorThe relevant board
Schedule movement > 2 weeks, or any scope/budget changeProgram DirectorGovernance boardLevel 1, on deadlock only
Architecture changeEnterprise ArchitectureEARBLevel 1, on deadlock only
Phase gateProgram Director (assembles the package)Executive Steering Board
Model to productionBRD teamIndependent Model Validation + AI Governance Board, bothNobody. Not the Sponsor, not another board, not schedule pressure.

7. How This Compares To the Other Programs Here

QuestionCatalyst (this suite)Lighthouse (stage-gate)PM / Agile (commercial)
Who chairs the deciding body?Not the Program Director — three separate chairsThe Program Manager, without a voteThe Program Manager, as chair
What stops the work?A failed validation, unappealablyA gate decisionA sponsor decision
PM's unilateral schedule authority±2 weeksWithin the gate windowWithin the approved envelope
What is being protected?The member on the other side of the modelThe capital decisionThe delivery commitment
Catalyst and Lighthouse solve the same problem two different ways. Lighthouse lets the Program Manager chair the deciding board but removes his vote. Catalyst lets him keep a vote nowhere, by never seating him on the boards at all. ***Both designs start from the same premise: the person who assembles the evidence must not be the person who rules on it.*** Which is right depends on how much the organization values the chair's context against the appearance of independence — and both answers are defensible.