AI OPERATING MODEL CANVAS
Applied AI Series reference design by João Frederico Neto · Hosted by Alsvior Global

1. JOB AND ACCEPTED OUTCOME
What business job is being completed? Who can accept it? What evidence proves completion?

2. CURRENT WORK
Map each handover, system, decision, recurring exception and human minute. Establish a baseline from actual cases.

3. DESIGN CHOICE
Compare conventional workflow, one bounded agent and coordinated responsibilities. Use the simplest design that meets acceptance and recovery needs.

4. EVIDENCE AND KNOWLEDGE
For each material fact, record source ID/version, claim, scope, access, status, reviewer, expiry and dependants. How is a contradiction resolved?

5. AUTHORITY
List permitted tool actions, approval roles, policy checks and actions that remain outside model authority. Name an exception owner.

6. ACTION AND RECOVERY
What happens if an action times out after reaching the target? How is the target reconciled? When is a bounded retry authorised?

7. COMPLETION CONTRACT
List required checks, independent verification, current evidence rules and named acceptance. Do not equate task activity with an accepted result.

8. ECONOMICS
Track attempted and accepted cases separately. Include all attempts, human review, fixed operation and recovery cost. Distinguish released capacity from realised cash.

9. EVALUATION
Use held-out cases and the same acceptance rules across designs. Measure failures and cost per accepted outcome; record versions and sample sizes.

This canvas is a planning aid, not a compliance certification or measured client case.