THE AI AUTOMATION CHECKLIST
UXLaunch Lab — uxlaunchlab.com

Most AI automation projects fail from scope, not technology — automating the
wrong thing, or shipping an agent nobody trusts enough to use. This checklist
is the sequence that avoids both.

BEFORE YOU AUTOMATE
[ ] The process is well-defined enough that two humans doing it reach the same result
[ ] You can name the specific hours or dollars currently lost to doing it manually
[ ] Someone with authority to change the process is involved from day one
[ ] You have ruled out that a simpler fix (a form, a template, a trigger) solves it

CHOOSING WHAT TO AUTOMATE FIRST
[ ] Highest-volume, most repetitive task is prioritised over the most technically interesting one
[ ] The task has clear success/failure criteria an agent can be measured against
[ ] Edge cases and exceptions are documented, not just the happy path
[ ] The task does not require judgment calls with legal or financial consequences — yet

BUILDING THE AGENT
[ ] Agent is trained on your actual content and data, not generic assumptions
[ ] Confidence threshold is defined for when the agent escalates instead of guessing
[ ] Every action the agent can take (book, update, reply) is explicitly scoped and permissioned
[ ] A human can see what the agent did and why, after the fact

TESTING & ESCALATION RULES
[ ] Tested against real historical requests, not just happy-path scripts
[ ] Escalation path is tested — a human actually receives and can act on handoffs
[ ] Agent has been tried by someone who was not involved in building it
[ ] Failure mode is graceful — a confused agent hands off, it does not loop or guess

LAUNCH & MONITOR
[ ] First 30 days have a named owner watching outcomes, not just uptime
[ ] There is a way to review a sample of agent conversations or actions weekly
[ ] A feedback loop exists for correcting the agent when it gets something wrong
[ ] Success is measured against the number named before you automated it

---
Need help building this? uxlaunchlab.com/services/ai-automation