PlaybookProgram
The first 90 days of a governed AI program.
A phase-by-phase playbook for standing up a governed AI program that survives its first audit — from mapping the decisions to running the operating system.
01
Phase 1
Map the decisions
Before any model, list the consequential decisions and how each is made today.
- Inventory the decisions that move money, risk, or trust.
- For each, document the trigger, the owner, and the evidence behind it today.Most teams discover the evidence is thin — that is the point.
- Pick the first decision to govern: high value, bounded scope.
Watch for
If you cannot name the decision and its owner, you cannot govern it. Do not skip this to get to the model faster.
02
Phase 2
Measure the baseline
Capture the before-state so every later improvement is a measurement, not a claim.
- Record current accuracy, speed, and cost-of-error for the chosen decision.
- Define what "better" means in numbers, agreed with the owner.
- Store the baseline where it can be cited a year from now.
03
Phase 3
Compile the gates
Put scope, baseline, and value checks into the runtime as conditions of execution.
- Implement the gates the decision must pass before the system acts.
- Wire an evidence record for every decision — inputs, version, gate, owner.
- Make the governed path the only path; remove the fast lane.
Watch for
A gate that can be bypassed under deadline is not a gate. Verify there is no exception route.
04
Phase 4
Run and evaluate
Ship inside the gates and let the harness watch quality continuously.
- Deploy the decision under continuous evaluation against a blended sample.
- Review the first audit query as a query against the ledger, not a scramble.
- Expand to the next decision on the same governed spine.
When you’re done
- One consequential decision running governed, evidenced, and fast in production.
- A repeatable spine the next decision plugs into.
- An audit you answer with a query, not a panic.
Run your first 90 days with us.
We will run this playbook alongside your team, on a decision that matters.