Did AI save time? Count the review.

Choose the real finish line

Imagine writing a client update. Producing a few paragraphs is only part of the work. Someone still needs to check the numbers, catch missing context and decide whether the message says what it should. An experiment that stops the clock before those checks tells you about drafting speed, not the whole task.

Define the finish line before you try a new tool. For this example, it could be an update that is factually checked, fits the audience and is ready for a person to send. Keep that definition the same for both approaches.

Count review as work

Use three simple notes: time spent preparing the input, time spent producing the draft, and time spent checking or correcting it. Keep a separate note for waiting time if another person needs to review it.

Here is a fictional example. A task used to take 40 minutes. The new approach takes 10 minutes to draft and 35 minutes to review. The finished task now takes 45 minutes. The draft arrived sooner, but this example did not save effort.

Give the tool a smaller job

A weak result does not always mean you should abandon the tool. It may mean you asked it to do too much. Try an outline instead of a final answer, or give it a checked source instead of asking it to supply facts. Then repeat the comparison.

Use several comparable tasks before making a broad claim. Record mistakes and rejected outputs as well as the good examples. A result that needs less correction may matter more than one that appears instantly.

Try this next

• Pick one task that you can finish in a single sitting.
• Write down what an acceptable result must contain.
• Time the existing approach, then the alternative—including review.
• Keep, adjust or stop the experiment based on the finished work.