Which AI pilots are worth funding?
A practical way to scope an enterprise AI pilot, test it against the work people do today, and decide what deserves the next round of funding.
Before a team funds an AI pilot, I’d ask the sponsor to finish one sentence. “If this works, we’ll change the way we ____.” A clear answer tells you which workflow to test, who needs to be in the room, and what evidence is worth collecting.
Choose one decision people make today
Imagine a freight team reviewing carrier invoices. A dispatcher spots a charge for detention, checks the contract, finds the arrival record, and decides whether to approve it or ask for more evidence. That one decision makes a better pilot than a brief to automate freight operations.
Ask the dispatcher to walk through a recent invoice. Which record did they trust? What was missing? Who could approve a disputed charge? Those answers give engineering a boundary and give the sponsor a concrete change to evaluate.
Agree on the evidence before the demo
Pick a set of past invoices that includes routine approvals, genuine disputes, and cases with missing records. Have an experienced reviewer establish what should have happened. Hold some examples back for the final evaluation so the team sees how the pilot handles unfamiliar work.
For this workflow, I’d track minutes of reviewer time, incorrect approvals, and exceptions correctly sent to a person. Compare those with the current process on the same cases. Let the owner of the payment decision set acceptable error levels before anyone sees the results.
Follow the exception all the way through
Take the invoice with a missing arrival timestamp. A useful pilot can identify the missing evidence, show the reviewer the contract clause, and leave the payment on hold. Watch what happens next. Can the dispatcher find the source, correct the record, and finish without opening three more tools?
That handoff changes the economics. Time saved on routine invoices can disappear if reviewers spend the afternoon untangling exceptions. Include review time, integration work, and ongoing ownership in the funding case.
Write down the funding decision
End with a short decision memo that names the workflow owner, the observed result, the remaining risk, and the next commitment. I’d give the sponsor three choices.
- Fund
- The pilot meets the agreed thresholds. An owner can run the workflow, and the expected benefit covers the cost of operating it.
- Revise
- One identifiable gap is holding it back. Name the change, its owner, and the evidence that would justify another test.
- Stop
- The economics fall short, a critical error persists, or the team cannot obtain the evidence it needs to make the decision safely.