Run your day · Lesson 6.3
Measuring AI Meeting Notes: The Few Numbers That Matter
It is easy to track the wrong thing with AI meeting notes. Counting how many calls you recorded or how many words the AI produced tells you about activity, not about results. This post names four numbers that show whether your notes are doing their job, and a weekly scorecard to read them.
You do not need software for this. A page of paper or a basic spreadsheet is enough. And there are no standard targets here, because good numbers depend on your work. The aim is to see your own trend.
Track time to send, errors in next steps, tasks completed, and how quickly you can find an old note. Review them once a week.
Why activity numbers mislead
Say you recorded twenty calls this month and the AI produced pages of notes for each. That sounds productive. But if the notes went out three days late, had wrong tasks and sit in a folder nobody opens, nothing improved. Meeting notes exist so people know what to do next. So measure that.
Metric 1: time from call end to follow up sent
Write down the time the call ended and the time the follow up left your outbox. The gap is the number. Shorter is better, because people respond to fresh messages and details are still clear in your head.
Choose a goal you can believe in, such as "same day" or "within the hour", and count how many calls meet it each week. A rising share tells you the routine works.
Metric 2: errors found in next steps
Each time you check the AI's next steps, count the corrections you had to make: wrong owner, wrong date, invented task, missed task. Keep a tally per call.
If the count is high, improve your instructions or your audio. If it falls over a few weeks, your changes are working. If it is always zero, check that you are really reading carefully, since that would be unusual.
Also record any error that reached someone else. That should be rare, and each one is worth a look at what went wrong.
Metric 3: tasks completed on time
This is the number that matters most. Of the next steps from your calls, how many were done by their date? You can count these at the weekly review. Put a tick next to each task when it is done.
Two cautions. Tasks with no date cannot be late, so count them separately; a lot of "no date" tasks is a sign your calls end without commitments. And tasks other people own depend on them, not only on your notes, so judge your own tasks most strictly.
Metric 4: time to find an old note
Once a month, pick a call from a few weeks back and time how long it takes to find what was agreed. If it takes seconds, your filing works. If it takes minutes, fix your naming or use one folder.
This number can feel odd, but it checks the reason you keep notes at all: using them later.
The weekly scorecard
Fill this in at the end of the week, in five minutes.
| Measure | This week | Last week | Note |
|---|---|---|---|
| Calls with a follow up sent within my goal | |||
| Total corrections to next steps | |||
| Errors that reached someone else | |||
| My tasks done by date | |||
| Tasks with no date | |||
| Time to find an old note |
Leave the cells empty until you have your own numbers. Compare to last week and to your own earlier months, not to other people.
How to read the scorecard
- Late follow upsIf follow ups go out late, move the notes step right after the call and block ten minutes for it.
- Many correctionsImprove audio and tighten the instructions, for example "use only the transcript".
- Missed tasksCheck whether the calls end with clear commitments. Ask "Who does what by when?" before you hang up.
- Slow searchingRename files with the date first and keep a single folder.
A small worked example
Say you review a week with four calls. Three follow ups went out within an hour and one next day. You made two corrections on the first call, none on the others. Of your six tasks, five were done by their date. You find an old note in under a minute. These numbers are made up to show how reading works, not benchmarks. You would conclude that the routine is steady, and the late one deserves a look: why did it slip? Maybe it was an evening call. A small change, like a ten minute block next morning, may fix it.
Do not overdo it
Four numbers are enough. If tracking takes longer than the notes themselves, drop to two: time to send and errors in next steps. You can add more once the habit sticks.
Also be careful about judging people by these numbers. They are for improving your own routine, not for ranking staff.
Connect it to your weekly review
These measures fit well in a weekly look back. The lesson Your weekly review shows a routine for that. And the notes routine itself is covered in Notes and next steps from every call. If you run the seven day trial in our trial post, the scorecard there feeds directly into these numbers.
The full routine is taught in the lesson Notes and next steps from every call.
Questions people ask
What is a good time to send the follow up?
Same day is a sensible goal for most work. Within the hour is better if you can manage it. Pick one you will keep.
How many errors are acceptable?
That depends on your work. Aim for a falling trend, and zero errors reaching other people.
Do I need a spreadsheet?
No. A paper page with the table above works fine.
Should I track how much time AI saves?
You can, but only by comparing with your own old timings. Do not rely on someone else's claim.