ReviewDesk guide · By the Ingenious AI team · Published 14 November 2026 · Reviewed 14 November 2026
Who should own an ADMT programme — owner, reviewers and record-keepers?
Generate a free adverse-outcome letter — no sign-up Build a free pre-use notice See pricing
The short answer. One named programme owner who sees the whole estate, reviewers chosen for decision competence and armed with overturn authority, and a record-keeper who owns the trail, the vault tests and the retention schedule. At small scale one person may hold all three hats — but the hats are worn in distinct sittings, on the record. Diffusion of these roles, not malice or ignorance, is how most ADMT programmes actually fail.
What is ADMT programme roles?
ADMT programme roles are the accountability assignments that make the notice-explanation-review system run: the programme owner (estate, calendar, standards), the reviewers (case judgement with authority), and the record-keeper (audit trail, vault, exports, retention). In ReviewDesk the roles map to the workspace's operation — inbox ownership, case assignment and the audit function — and scale from one person wearing three hats to a staffed bench with pods.
Why do programmes fail by diffusion?
Because ADMT work sits between functions. Legal knows the duties, HR or operations runs the decisions, IT runs the tools, and the programme — the living system joining them — belongs to no one. The symptoms are consistent: notices drafted once and never re-read, an inbox that is everyone's side-task and no one's queue, a vendor change nobody mapped to the scope worksheet. Each function performed its part; the system failed at the joins, and joins are owned by roles, not by departments. Naming the three roles is the act that creates the joins.
The roles in this guide are deliberately few. Programmes drown in governance theatre — committees, RACI matrices with nine columns — while the actual work needs an owner who watches, reviewers who judge and a keeper who files. Three roles, named, with their rhythms written down, outperform a governance architecture that meets monthly to admire its own diagram. Scale adds people inside the roles, not more roles.
What does the programme owner actually do?
Owns the estate view and the calendar. The owner holds the workflow inventory and the scope determinations' currency, watches coverage and clocks across workflows, runs the recurring rhythms — weekly letter sampling, monthly aggregates, quarterly notice re-reads and vault tests — and owns the regime calendar: Illinois live, the 2027 starts, the April 2028 assessment, the December 2027 EU horizon, per the deadlines calendar. When a vendor announces a model change, the owner's desk is where its notice and scope consequences land. The owner is the programme's memory and its alarm system in one person.
Authority follows the view: the owner must be able to pause a workflow whose notice has gone stale, to escalate a bench capacity problem with data attached, and to say 'not recorded' in leadership's hearing when the records say so. An owner who can only report is a narrator. Choose someone whose judgement the organisation already trusts with bad news, because this role's core product is early, accurate bad news — the capacity the entire evidence philosophy depends on.
What makes a reviewer, as a role?
Competence for the decision type, independence from the original decision where achievable, explicit overturn authority, and protected time — the four properties from the review checklist, restated as a job description. The role's definition should be written before the first case: which decision types, what authority (restore, overturn, adjust terms — specified per type), what the outcome note must contain, and how quality is sampled. Reviewers recruited by availability rather than by the role definition produce the variance that sampling then has to discover the hard way.
The role also includes a duty of candour upward: reviewers see the decision system's failures first — the factor that keeps misfiring, the data source that keeps lying — and the programme needs a channel for that sight to become fixes. The monthly reviewers' pattern conversation, chaired by the owner, is that channel. A bench whose patterns die in case notes is a sensor network unplugged at the dashboard, and the organisation will keep paying for the same errors one applicant at a time.
What does the record-keeper own?
The integrity of the memory: the audit trail's completeness, the vault's horizons and export tests, the retention schedule's punctual deletions, the versions archive, the reconciliation between decision logs and traces where the engine runs. The record-keeper's rhythm is verification — the quarterly cold export, the access list's review, the sample case reconstructed from records alone. Where the owner asks 'is the programme working?', the keeper answers the prior question: 'can we prove what the programme did?' — the question every other question eventually becomes.
At scale the keeper also guards the honesty rules operationally: no backfills, no silent edits, corrections appended and attributed, exports logged. Culture phrases it as values; the keeper's checklist makes it a practice. The role pairs naturally with the inbox owner at small scale and separates as volume grows, because verification must eventually be independent of the work verified — the same separation logic as reviewer independence, applied to the programme itself.
How do the roles work at one-person scale?
Honestly, with the hats distinguished by sitting and by record. The owner-reviewer-keeper who configures the tool in the morning, reviews a case at noon and files the record in the afternoon is performing three roles whose separation exists to create checks; at one-person scale the checks become disciplines instead: the review conducted on the full file in a separate session with a fresh note, per the small-business guide; the quarterly export test done cold, as if by a stranger; the scope determination written before the outcome is known. The records must show the distinctions the headcount cannot.
Name a deputy for absence even at this scale — a person who can acknowledge requests and keep the clocks from breaching during holidays — because coverage does not respect annual leave. The deputy needs access, a one-page runbook and the owner's rhythms written down. Programmes that live entirely inside one person's habits are one resignation from archaeology; the runbook is the cheapest continuity instrument in this entire fleet of guides.
How do roles evolve as the programme grows?
By specialisation inside the roles, in a predictable order. Reviewers pod by decision type and gain leads who sample quality — the growth path in the assignment guide. The keeper gains the traces reconciliation and the pack calendar. The owner gains the vendor-management and assessment duties at full weight. What never changes is the count of accountabilities: one owner for the estate, named reviewers for cases, one keeper for the record. Growth that adds coordinators between the roles, rather than capacity inside them, reintroduces the diffusion the roles exist to prevent.
Write the role definitions into the programme's one-page constitution and review them annually alongside the risk assessment. Roles drift as people change; the annual refresh re-names the humans, re-states the authorities and re-commits the rhythms. An ADMT programme is, in the end, three promises with names attached — we watch it, we judge it fairly, we can prove it. Everything else in these guides is machinery for keeping those promises at scale.
Your five actions for ADMT programme roles
- Name the three roles — programme owner, reviewers, record-keeper — in writing, with a deputy for absence.
- Give the owner the estate view, the calendar and the authority to pause a stale workflow and escalate with data.
- Define reviewers by competence, independence, specified overturn authority and protected time — before the first case.
- Charge the record-keeper with verification: quarterly cold exports, reconciliation, retention executed on schedule.
- At one-person scale, separate the hats by sitting and by record; at any scale, refresh the role sheet annually.
Worked example: the programme that survived its founder leaving
A fictional employer's ADMT programme was, for its first year, one operations manager — Priya Nair — who built the workflows, cleared the inbox and kept the records in her own meticulous habits. When she resigned, the handover took two days, because the programme's constitution existed: role sheet with her three hats itemised, the rhythms in the calendar, the runbook for the deputy, the vault test results filed quarterly. Her successor's first month was unremarkable, which was the triumph: coverage held, clocks held, and the first sampled letters read exactly like the last of Priya's.
The counterfactual is the common story: programmes that live in a person's head leave with the person, and the organisation discovers the departure three months later, in an unanswered inbox and a notice describing a retired tool. Roles written down are continuity insurance with a premium of one page. Priya's employer paid it in her first month, on the advice of nobody in particular — just the discipline this guide describes, applied before it was needed, which is the only time discipline ever gets applied cheaply.
Frequently asked questions
What are the three roles every ADMT programme needs?
A programme owner (estate view, calendar, standards, authority to act), reviewers (decision competence, independence, overturn authority, protected time), and a record-keeper (trail completeness, vault tests, retention, export discipline). Named people, written definitions, a deputy for absence.
Can one person hold all three roles?
Yes, at small scale — if the hats are separated by discipline: reviews as distinct sittings on the full file, verification done cold, scope written before outcomes. The records must show the distinctions the headcount cannot.
What is the programme owner's core product?
Early, accurate bad news — plus the estate's rhythms: weekly sampling, monthly aggregates, quarterly re-reads and tests, and the regime calendar held against the operation. The owner needs authority to pause and escalate, not just to report.
Why must record-keeping eventually separate from doing?
Verification must be independent of the work verified — the same logic as reviewer independence. At scale, the keeper who tests exports cold should not be the person whose cases are being tested.
Is this guide legal advice?
No. ReviewDesk is templates and workflow tooling, not legal advice. The roles are ReviewDesk's operating model; consult counsel for your governance obligations.
Keep reading: What are the key ADMT deadlines — Illinois now, Colorado and California 2027, the EU in December 2027? · What changed with Colorado SB 26-189 — and what does it keep from the earlier AI Act approach? · What is the California ADMT compliance timeline — from the 2026 regulations to the 2028 assessment? · What does ReviewDesk's free tier actually cover — and when do you need Pro or Team? · What does a good adverse-outcome letter actually look like? (Worked examples) · All ReviewDesk guides · ReviewDesk home · Try the free desk · Free letter generator · Free notice builder · Pricing
ReviewDesk is built by Ingenious AI. Templates and workflow tools — not legal advice.