AI Operator Reliability Checklist

The 7-point checklist

1. One clear job + one clear success metric
   Define a closed loop (e.g., “inbound lead → qualify → schedule → log to CRM”) and one metric (e.g., “median follow-up < 10 minutes”).
2. Tight tool access (scopes, not god-mode)
   Minimum permissions. Separate operator accounts from founder accounts when possible.
3. State + idempotency (no double-sends)
   Every action needs a unique key and safe retries: “If already processed, do nothing.”
4. Monitoring + alerts (no silent failure)
   Logs for every run + alerts on repeated failures or “no runs in X hours.”
5. Human-in-the-loop where it matters
   AI drafts/queues; humans approve money/contracts/sensitive outbound.
6. Backoff + rate-limit discipline
   Exponential backoff under errors.
7. Weekly ops hygiene
   Kill broken hooks, rotate tokens, review top failures, fix root cause.

The quickest Week 1 install that pays off

Pick one workflow (don’t boil the ocean):

• Inbound (email/DM/form) → classify intent
• Follow-up drafted immediately
• Scheduling link or next step inserted
• CRM log + task created
• Daily summary to founder