The most expensive AI mistake is improving the wrong workflow first. A strong first choice scores well on five simple tests.
Then one question decides how you improve it: is the bottleneck the doing (you need it faster) or the deciding (you need a layer that shows the impact of a choice before someone makes it)? Both are valid. Naming which one sharpens the whole build.
What a bad first choice looks like
The same five tests tell you what to avoid for a first project. Skip anything where a wrong answer does real damage before a person can catch it. Skip anything that has to plug into five systems on day one. And skip anything that comes down to open-ended judgment with no written rules and no clear right answer. None of these are off-limits forever. They are just the wrong place to start, before the approach is proven and trust is built.
Score two candidates
Run two through the tests. Matching invoices to purchase orders: heavily manual, a clear dollar return, low risk because a person still approves each payment, one or two systems to start, and visible in weeks. It scores. Setting next year's pricing strategy: high stakes, no fixed rules, judgment all the way down, and no way to see it working in weeks. It does not, at least not first. The test is quick on purpose, so you can run every candidate on your list in an afternoon.
The tests, applied to the real case
The problem. A nonprofit runs recuperative-care beds: medical respite for homeless patients leaving the hospital. Hospitals send the same referral to four or five facilities at once, and the first to respond wins the patient. The operator's window was about fifteen to twenty minutes. Before, a staffer had to notice the email, open a dozen pages of attachments, read them, and decide. Every empty bed is lost revenue against a fixed payroll.
What we built. About a month after their first email to us, they had a live system running one workflow end to end:
The AI reads the packet in minutes and never guesses. A fixed rulebook, not the model, applies the operator's own admission policy, so every decision can be explained. A person makes the final call, with the source document on screen. Beds filled is the number it moves.
It scored on all five: heavy manual load, a filled bed is direct revenue, the system only recommends, ninety-eight percent of referrals arrive in one inbox, and it was live in a month. That was not luck. That is how you choose.