15 questions.
Before you spend a dollar on AI.
The same self-assessment we run before we quote a build. Five sections — Data, Team, Stack, Compliance, Ops. If you answer no to more than four of these, you don't have an AI problem yet. You have a prerequisites problem.
Data
- Q01
Do you have a single source of truth for the data your AI will touch — customer records, transactions, call logs, inventory — or does it live in 4 disconnected systems?
- Q02
Is that data clean enough that a new hire could read it without asking three clarifying questions?
- Q03
Do you have written permission (or the legal basis) to use customer data for automated decisions, retrieval, or training?
Team
- Q01
Is there one person on the team whose job description will change after this AI ships, and do they know it?
- Q02
Does your operator actually want AI in this workflow, or did someone upstairs decide for them?
- Q03
Who owns the AI after launch — monitoring, retraining, cost control, escalation paths?
Stack
- Q01
Can your current systems send and receive webhooks, or expose an API, without a 3-month integration project?
- Q02
Is your CRM / ops system the bottleneck you're trying to replace, or the thing the AI will plug into?
- Q03
Do you already know which model family — Claude, OpenAI, open-source — has to be avoided for compliance, cost, or vendor lock-in reasons?
Compliance
- Q01
Is the data the AI will touch subject to HIPAA, SOC 2, GDPR, PCI, or state privacy laws, and do you have a sign-off path for those obligations?
- Q02
Do you have an audit trail requirement — every decision logged, reproducible, and reviewable — or is best-effort enough?
- Q03
If the AI made a mistake that cost a customer money, could you explain the decision in writing within 48 hours?
Ops
- Q01
What exact metric improves when this AI works — calls recovered, hours freed, revenue per lead — and who reviews it weekly?
- Q02
What's your rollback plan if the AI starts producing bad output at 2am on a Saturday?
- Q03
Are you willing to let the AI take full action on low-risk work in month one, or does every action need human approval until trust is built?
— Take it with youKeep it handy for the team review.
Send yourself a link to this complete checklist so you can return to it during scoping. This is a one-time resource delivery, not a newsletter subscription.