Before you call an AI tool a win, ask the person who uses it every day

An owner asks how the new AI tool is going. The answer sits on their desk. Summaries arrive in minutes. Drafts show up before lunch. The work looks done.

The person who uses the tool every day sees something else. Figures they re-check against the source. Output they retype into another system because the layout does not fit. An extra screen to log into. The same correction, made every time.

Recent surveys keep finding this gap. Owners and managers are far more confident the tool delivers than the staff who use it. A tool can stay in use, look like a success from the owner's chair, and still make someone else's day longer.

The owner sees output, the person using it sees steps¶

The owner judges the tool by what reaches them. The person using it judges it by what it takes to get there.

Those are different measures. Output is visible and arrives in one piece. Steps are scattered. Two minutes checking here. Three minutes retyping there. A login after lunch. Spread across a day, nobody adds them up.

Fast output can sit on top of slow work.

A bookkeeping practice where the summaries look like a win¶

A 9-person bookkeeping practice uses AI to summarise the documents clients send in each month. The owner sees the summaries arrive within minutes. To the owner, the tool is a success.

The junior bookkeeper works from those summaries. She re-checks every figure against the source document. She retypes the totals into the ledger by hand, because the summary layout does not match the ledger. Each client file now takes her longer than it did before the tool.

She does not raise it. The owner is pleased with the tool, and she has a month of files to get through.

None of this is on her. The tool was set up to produce summaries, not to fit her work.

Why the extra steps never get reported¶

Three things keep the cost quiet.

• The tool was the owner's idea. Questioning it can feel like questioning the owner.
• A complaint sounds like resistance. Nobody wants to be the one who dislikes the new thing.
• Each step feels too small to mention. Two minutes is not worth raising. Forty of them a week is.

Silence is not the same as satisfaction.

Count the steps for one ordinary task¶

The test is plain. Sit with the person for one ordinary task. Not a demo. A normal file on a normal day.

Count the steps it takes now. Then count the steps it took before the tool. Include everything:

• Checking: comparing output against the source.
• Copying: moving output into another system.
• Fixing: correcting the same thing again.
• Switching: extra screens and extra logins.

If the new count is higher, the tool is adding work, whatever the output looks like.

Three questions to ask the person who uses it¶

Then ask them directly. Make it clear you want the real answer.

1. What do you do to the output before it is usable?
2. What do you check every single time?
3. Would you keep using it if the choice were yours?

The third question is the honest one. Listen to the pause before the answer.

Remove each added step, then set a date to decide¶

Treat every added step as a defect, not a quirk the person has to live with. Most can be removed by changing one of three things:

• The output layout, so it matches the ledger or system it feeds.
• The instructions, so the same correction stops being needed.
• The handoff, so the output lands where the work actually happens.

Then set a date, four to six weeks out. Recount the steps. Ask the three questions again. Decide whether the tool stays.

I would not call an AI tool a win until the person who uses it every day says it saves them time. Their answer is the measure, not the speed of what reaches your desk.

Keep exploring¶

If the same correction keeps showing up in your count, fixing the same AI mistake every week shows how to turn it into a standing instruction. If you want to talk through what your count turns up, contact FIT.