The 12 checks I run before any company turns AI loose on real work. Covers data integrity, guardrails, user enablement, and rollout order — the things that separate a safe launch from an expensive mess.
✓What to lock down before anyone touches a prompt
✓The 4 questions that decide if your data is AI-ready
✓Which workflow to automate first (and which never to)
GET THE CHECKLIST
The 12 checks
YOURS · PRINT THIS PAGE
BEFORE ANYTHING · DATA
One source of truth exists. If two systems disagree on a number, fix that first. AI amplifies bad data.
You know where sensitive data lives. Client records, payroll, contracts — mapped before any tool touches them.
Access is scoped. The AI sees only what the task requires, never the whole drive.
GUARDRAILS
A written usage policy. One page: what's allowed, what's never allowed, who to ask.
Human sign-off on anything outbound. Nothing AI-drafted reaches a client unreviewed, until trust is earned per-workflow.
A kill switch. Someone owns turning any automation off in under a minute.
PEOPLE
Users are trained on failure modes, not just features — hallucination, stale data, confident nonsense.
Someone owns AI. A named person accountable for usage, spend, and outcomes. Not a committee.
The team knows why. "This removes your busywork" lands; silent rollouts breed fear and shadow usage.
ROLLOUT ORDER
Start internal, low-stakes, high-volume. Reporting, reconciliation, drafting. Never customer-facing first.
Measure before and after. Hours saved, error rates, cycle time — or you'll never know if it worked.
Expand only on evidence. Each new workflow earns its automation with data from the last one.
Want this done for you, with someone accountable for the result?