The five-day tally, and why memory is not a baseline

The measurement everyone skips

Half the businesses that adopt AI cannot say afterwards whether it helped. Not because the effect was small, but because nobody wrote down what things were like before. What they have instead is a feeling, and a feeling in week one of a new tool is worth nothing: every new tool feels good in week one, including the ones that get abandoned in week five.

So before you change anything, spend five working days counting. It costs about four minutes a day.

What to count

One line per instance of the task, on your phone or on a sheet by the till. Four columns:

1. What it was. Two or three words. “Price enquiry, WhatsApp.”
2. Minutes. Rounded to the nearest five. Do not agonise; a rough number honestly recorded beats a precise number invented on Friday.
3. Did it need you? Yes or no. Could a competent new employee with your price list have done it?
4. Outcome, where there is one. Booked, quoted, ignored, refunded.

At the end of the week you have four numbers that will still be true in six months: how many, total minutes, the proportion that needed you specifically, and the conversion rate.

Those four are your baseline. Write them at the top of a page with the date. That page is what makes any later claim of improvement checkable by somebody other than you.

Why memory does not work

People do not misremember randomly; they misremember in a consistent direction, and knowing the direction is what makes the tally worth doing.

Hated tasks inflate. Ask an owner how long they spend on invoice chasing and you will typically hear two to three times the tallied figure. Unpleasant work is remembered by how it felt, and it feels long.

Frequent short tasks vanish. Forty ninety-second interruptions do not register as an hour of work. They register as nothing at all, which is exactly why they are the biggest opportunity and the last thing anyone names.

The recent week overwrites the year. Whatever happened in the last three days feels typical. It usually is not.

The combination is unfortunate: owners tend to attack the task that annoys them most, which is often not the task that costs the most. The tally corrects both errors at once, and it is the only instrument in this course that does.

Five days, and when five days is not enough

Five working days is enough for anything that happens several times a day. It is not enough for weekly work, and it is badly wrong for seasonal work.

If your candidate task is a month-end or a quarter-end job, count one full cycle instead, or reconstruct it from records: invoices raised, emails sent, appointments in the diary. Reconstruction from records is far better than recollection, because the records do not have feelings about the task.

And pick an ordinary week. A festival week, a week you were away, a week the main supplier failed: none of these are your baseline. If the week you tallied was strange, say so on the page and tally another.

The two numbers you will want later

When you run the trial, you will be comparing against this page. Two of the numbers do most of the work.

Total minutes per week is the size of the prize. Multiply by 48 to see the year. A task at 30 minutes a week is 24 hours a year, and no tool is worth a week of setup to halve it.

The proportion that needed you is the ceiling. If two thirds of enquiries could have been handled by a competent stranger with your price list, then two thirds is roughly the most any tool can take off you, and a vendor promising more is promising to remove your judgement rather than your workload.

Keep tallying, occasionally

One week a quarter, repeat it. It takes twenty minutes across five days and it tells you three things nothing else will: whether the saving held after the novelty wore off, whether the work quietly grew back in a different shape, and whether the thing you automated is still the biggest item on the list.

Usually, after a successful change, it is not. Something else has moved to the top, and you would not have known.