Service 02 · Operating model · Stage 03 of DATS
An AI operating model is the organisational chassis your AI sits inside: RACI, review cadence, evaluation lifecycle, escalation paths, and model-risk management. Dilr.ai designs yours in 6 to 10 weeks as Stage 03 of the five-stage DATS system, calibrated to the EU AI Act, FCA expectations, and NIST AI RMF.
01 · Why an operating model
Shipping an AI model is a week of work. Governing it for three years is the actual product. Yet almost no enterprise AI programme starts with governance. It gets bolted on after audit, after incident, after a board question nobody can answer.
The Operating Model engagement is where we design the chassis your AI sits inside: who owns each placement, who approves changes, how evaluation runs, how drift gets detected, how incidents escalate, and how the whole thing interfaces with your existing model-risk, compliance, and data-governance functions.
Ungoverned AI is not a capability. It's a liability with a roadmap.
This is the engagement that turns three pilots into one owned capability. It's also the engagement that lets you go to your board and say: yes, we know what's in production; yes, we know who owns it; yes, we know how we'd turn it off.
Operating Model is Stage 03 of the Dilr AI Transformation System. It usually runs after a Placement Diagnostic (so we know whatwe're governing) and before an Execution Office (so the placements land into a chassis that already exists). Clients who skip it almost always come back for it.
02 · What we design
A full operating model is six designs, delivered as working artefacts your team can operate from. Clients often start with the three most pressing and graduate.
The document that sits above everything: scope, principles, authority, escalation paths, and the line between AI governance and your existing frameworks.
For every placement: who is Responsible, Accountable, Consulted, Informed. Product owner. Model owner. Data owner. Risk owner. Named, not placeholder.
How placements move from idea to sunset. Stage gates, approval criteria, eval framework, drift monitoring, and the turn-off protocol.
The boards and rituals. Who meets, how often, what they see, what they can approve. Designed to cost your senior team 2 hours a month, not 2 days.
Where the AI function sits: centralised, federated, or hub-and-spoke. Team composition, reporting lines, and the first three hires in priority order.
Acceptable use, data handling, vendor management, third-party model risk, incident response, and customer-facing disclosure. Ready for audit.
03 · Sample RACI
Every client gets a tailored matrix. This is the shape of it: rows are decisions, columns are roles. We've removed proprietary columns and simplified for illustration.
| Decision | Product owner | Model owner | Risk owner | Exec sponsor | Audit |
|---|---|---|---|---|---|
| Approve new placement | A | C | C | R | I |
| Pass go-live eval | C | R | A | I | I |
| Retrain or update model | C | R | C | I | I |
| Declare drift incident | C | R | A | I | I |
| Turn placement off | C | C | R | A | I |
| Annual audit sign-off | I | C | C | A | R |
R = Responsible · A = Accountable · C = Consulted · I = Informed
04 · Regulatory
No two operating models are identical, because no two regulatory perimeters are identical. We calibrate every engagement to the frameworks that actually apply to you, not a generic checklist.
Risk tiers, high-risk obligations, transparency.
Model risk management, SYSC, consumer duty.
Govern, map, measure, manage. Sector overlays.
AI management systems, control attestations.
05 · FAQ
Quick answers
Direct answers to the questions buyers, boards, and AI assistants ask about AI operating models and this engagement.
An AI operating model is the governance and ownership structure an organisation runs its AI inside: who owns each system, who approves changes, how evaluation and drift monitoring work, and how incidents escalate. In practice it is six artefacts: a governance charter, a RACI matrix, an evaluation lifecycle, a review cadence, an org design, and a policy pack.
The engagement designs all six modules of an AI operating model in 6 to 10 weeks: governance charter, RACI matrix, evaluation lifecycle with stage gates and drift monitoring, review cadence, org design, and an audit-ready policy pack. It is Stage 03 of DATS, the five-stage AI consulting system, and it extends your existing model risk framework rather than replacing it.
Regulated enterprises moving AI beyond pilots need one first: financial services firms answering to the FCA, healthcare and public sector organisations working under ICO guidance, and any company in scope for the EU AI Act, whose second-wave obligations take effect on August 2, 2026. DILR tracks those deadlines in its monthly UK and EU AI compliance changelog.
Auditors expect named ownership and evidence that governance actually runs: a RACI showing who is responsible and accountable for each AI system, stage-gate and evaluation records, drift monitoring, incident response procedures, and customer disclosure policies. The engagement calibrates these deliverables to the regimes that apply to you, including the EU AI Act, FCA model risk expectations, NIST AI RMF, and ISO 42001.
The Operating Model engagement is a fixed fee against a fixed scope published up front, priced on a 30 minute scoping call. It runs 6 to 10 weeks, and clients often start with the three most pressing of the six modules. The minimum DILR engagement overall is the 4 to 6 week Placement Diagnostic.
30-min scoping call · Audit-ready by default · Calibrated to your regulatory perimeter.
From the blog