Use AI without lowering the quality of the work

Pick a bounded task

First drafts, meeting summaries, internal procedures and reformatting are easier to test than anything a client relies on. A good test task has three things: it happens often, someone already owns the result, and a person can tell a good version from a bad one quickly.

Avoid starting with work where an error is expensive and hard to spot, such as figures in a client report or advice someone will act on. Those can come later, once people have a checking habit.

Try it freeWant to see a session first? The free AI Hour is online on the first Friday of every month. Next: Fri, Nov 6, 12 PM ET.

Write the baseline before you change anything

Take five normal examples of the task done the current way. For each one, note:

• How long it took, roughly, start to finish
• Whether it needed rework, and why
• What usually goes wrong

A rough, honest baseline beats a precise number invented afterwards. Without it, every AI test "feels faster", because nobody counts the checking.

Decide where AI stops

Write down which steps the tool may do and which stay with a person. A typical split: AI may structure notes, draft, reword and summarize. A person chooses the sources, handles exceptions, confirms anything involving money, dates or commitments, and gives the final yes.

A next stepRather have someone run it? See the public workshop or private team training, or book a free 30-minute call.

Write the review checklist first

Agree the checklist before the test, so you judge every output the same way:

• Every fact, figure, date and name matches the source, or it's removed
• Nothing was added that the source doesn't say
• The required details are all there
• The tone fits the person who will read it
• No personal or confidential detail slipped in
• The owner can explain every sentence

A worked example

This is an illustration with made-up numbers, to show the arithmetic. It isn't a measured result.

Read it honestly. In week one, the AI version was faster but less accurate, because the draft invented an action item nobody agreed. The fix wasn't a better tool. It was one more line on the checklist: "every action item has an owner who was in the meeting". By week three, the quality matched the old way and each summary took about 12 minutes less. That is a keep.

If the checked version had been slower, or kept needing corrections, the right answer would have been to stop.

Standardize or stop

After two to four weeks, make a decision:

• Keep: write down the steps, the approved tool, the prompt that works, the checklist and the owner, so anyone on the team can repeat it.
• Change: adjust one thing (the prompt, the source material, the checklist) and test again for two weeks.
• Stop: if the checked result isn't better or faster. Don't keep a tool to justify the subscription.

Related: How to train employees to use AI · AI readiness checklist · Free guide