Most failed automation projects fail at selection, not construction. The workflow chosen was too ambiguous, too political, or too rare to justify the build. Here is the filter we apply — in our own businesses first — before anything gets built.
Start with frequency and structure
The best first automation is a workflow that happens often and follows a recognizable shape. Client intake. Appointment scheduling. Proposal preparation. Document collection. If the process runs dozens of times a month and the steps are broadly the same each time, automation has something to grip.
Rare, bespoke work is a poor first target even when it's expensive — there isn't enough repetition to recover the build effort, and every run is an exception.
Look for the transfer points
Time rarely disappears inside a step; it disappears between steps. The inquiry that waits for a callback. The call notes waiting to become a proposal. The signed agreement waiting for someone to start onboarding. Map where information waits for a human to move it, and you've found the automation candidates.
Check for a clear judgment boundary
A good first workflow has an obvious line between structured motion and human judgment. Capturing a lead's details is structured; deciding whether to take the client is judgment. Drafting a proposal from templates is structured; approving the price is judgment. If you can draw that line cleanly, you can automate one side and protect the other with an approval point.
Demand a measurable baseline
Before building anything, measure the current state: how many times a month, how long it takes, how long things wait, how often exceptions occur. Without a baseline, you'll never know whether the automation worked — and "it feels faster" is how budgets get wasted. If a workflow can't be measured, it isn't ready to be automated.
The shortlist test
- Happens at least weekly, ideally daily
- Follows a recognizable structure most of the time
- Has visible transfer points where work waits
- Has a clean boundary between structured steps and judgment calls
- Can be measured before and after
A workflow that passes all five is a strong first project. One that fails two or more should wait — automate something else first and come back when the process itself is more settled.
This selection work is exactly what an AI Opportunity Audit produces: the map, the baseline, and the ranked shortlist.