The point of this playbook
A shared operating model that makes funding, approval, evaluation, and escalation explicit.
01
Mandate and outcomes
Begin with the change the operating model exists to produce. Tool adoption is an activity, not the outcome.
- What business or workforce outcomes matter this year?
- Which types of AI work are in scope-and which are not?
- What evidence will leadership review each quarter?
02
Decision rights
Write down who can start, approve, pause, and retire AI-enabled work. Ambiguity here creates either uncontrolled experimentation or permanent gridlock.
- Who owns the workflow and its result?
- Who approves data use, risk level, and production access?
- Which decisions stay local, and which require a central review?
- Who can stop a system when the evidence changes?
03
Portfolio and funding
Use one visible portfolio instead of disconnected pilots. Fund the next piece of evidence, then expand only when the case improves.
- How are opportunities proposed and compared?
- What must be true before a prototype becomes a pilot?
- Who pays for shared infrastructure and team-specific work?
- When is an experiment retired or absorbed into normal operations?
04
Evaluation and escalation
Every production workflow needs a quality bar, monitoring owner, and clear path when something goes wrong.
- What is measured before launch and after launch?
- Where is human review mandatory?
- How can employees or customers report a problem?
- Who investigates, communicates, and decides whether to resume?