The Receipt Test: show what AI actually changed
A usage report tells you whether people opened the tool. A receipt tells you what changed in the work.
Published: November 9, 2026
When AI spending gets questioned, "we use it a lot" is not an answer. Prompt counts, license counts and enthusiasm don't show what the money bought. A receipt does: one task, measured before and after, with the checking time included and the limits written down.
This guide shows you how to write one in about ten minutes. You don't need a data team. You need one recurring task and an honest note of what you haven't measured.
The six fields
Copy this:
How to fill it in
- Pick one recurring task. A weekly report, a monthly reconciliation, a standard client email. Not "AI in general." One task you do often enough to compare.
- Write down the before. How long did an acceptable version take without AI, including review? If nobody timed it, write "not measured." That's a finding, not a failure.
- Name the change. Which tool, and which part of the task it handles.
- Measure the result across more than one run. One good afternoon proves little. Note how many times you measured.
- Count checking and correction time. AI drafts fast. If fixing the draft eats the savings, the receipt should say so.
- Write the limit. One person, one task type, a handful of runs. Say it plainly. A receipt that admits its limits is easier to trust.
A worked example (hypothetical)
Four weekly status reports. Before AI, drafting and review averaged 100 minutes per report. With AI, drafting averaged 20 minutes and checking and correction averaged 25, for 45 total.
- Result: 55 minutes saved per report.
- Quality: the same reviewer used the same checklist and recorded no increase in errors.
- Limit: one person, one report type, four reports.
Notice what the example doesn't claim: that the whole team saves 55 minutes, or that every report type works the same way.
Three mistakes that weaken a receipt
- Counting usage instead of results. "We ran 200 prompts" describes activity, not value.
- Leaving out checking time. The draft got faster; did the finished work?
- Measuring once. A single good run is an anecdote. Note the count.
What to do with your receipt
Bring it to the next budget or tool review. One honest receipt, with its limit stated, gives your team a clearer case than usage counts or enthusiasm. It also shows where AI didn't help, which is worth knowing before renewal.
Keep going
- The cost side: our Budget Math series, starting with Your AI Subscription Is Becoming a Utility Bill.
This guide is a habit, not an audit. It isn't financial, legal or compliance advice. A receipt shows what you measured; it can't prove what you didn't. From The AI Minute. We read 47 AI and tech newsletters so you don't have to.