Automation fails for predictable reasons: the wrong first process, no owner, and no way to measure whether it worked. Here is the 90-day shape we use to avoid all three.
Days 1–30: map and measure
Shadow the teams doing the work. Count volumes, time per task and error rates. You are looking for processes that are frequent, repetitive and painful — and where the data already lives in systems, not in someone’s head.
Days 31–60: build the first loop
- Automate one process end-to-end, including exceptions.
- Put a human-in-the-loop step where judgement matters.
- Instrument everything: runs, failures, time saved.
Days 61–90: make it a platform
Turn what you built into reusable components — connectors, prompts, approval flows — so the second automation takes a third of the time. Publish a simple dashboard so the business can see value accumulating week by week.
Ninety days is enough to prove the model. The compounding starts after.
