How a Royalty Accrual Went From Hours to Near Instant

Three weeks ago one of our architects walked a staff accountant at a licensed apparel company through the royalty accrual tool we had built with him. He watched a month calculate on screen.

Then he told the whole history of the task in a few sentences. Putting the accrual together used to take him three to four hours. A spreadsheet he had built brought it down to about half an hour. This, he said, was "like instant."

It is a good story, and it is easy to tell wrong.

Whose number is whose

The version you would put on a slide is three hours to instant. That is not what happened.

The middle step is his. A few months ago he built his own spreadsheet for the accrual, with an AI assistant helping. That spreadsheet took the job from hours to half an hour or less. He sent it to us, and what he asked for next was simple: stop having the information scattered all over the place.

Our part is the last step: half an hour to close to nothing. Rates held per licensor in one place, the accrual calculated from them, and an export to the general ledger so nobody retypes the result into accounting.

So the honest claim is smaller than the slide, and I would rather make it. Most of the time saved was saved by the person who does the work, before we built anything for him. If I had published three hours to instant, the first person to correct me would have been him, and he would have been right.

That is not a footnote. It is the best news in the story. A company where the accountant builds his own fix with an assistant is a company where the next few fixes are already half designed, by the people who know the work.

Why the last half hour was still worth building

From half an hour to nearly nothing sounds like a rounding error. Three reasons it was not.

It happens every month. A task that recurs every close, for as long as the company holds licenses, is worth getting right once.

The rates live in one place. When each licensor's rate sits in one table with one owner, the accrual uses the current rate every time. A copied spreadsheet uses whatever rate was right on the day it was copied.

The export removes the last re-key. The result goes into accounting in the format accounting expects, instead of being typed a second time by hand.

His own verdict was simpler: he had not thought we could save him even more time.

Where it stands today

It is not finished, and I am not going to pretend it is. He is running past months through the tool and checking them against his own methods. The ledger export is the last open piece, and it goes in this week. What reaches the chief executive so far is feedback from the people using it, which is the order I would want.

The three to four hours and the half hour are his numbers, and they are the right people's numbers: they come from the person who does the work, not from us. They are also recollections, and "like instant" is his impression rather than a timing. We did not time the original method with a stopwatch before it changed, which is the exact gap I wrote about in How to Measure the Return on AI Before You Launch.

So what I want from him next is two numbers from the parallel run. Minutes in the tool against minutes in his spreadsheet, for the same month. And whether the two totals match. When I have those, they replace everything above.

If you are measuring your own

Credit every step. A before and after hides the middle. Write down each version of the process between the old way and the new one, and who made each change. Somebody inside your company has often done part of the work already.

Measure against today, not the oldest version. The comparison that matters is the new tool against what the person does now. Comparing it to the way the job was done two versions ago flatters the tool and quietly erases the person who improved it.

Let the person who does the work hold the stopwatch. They know where the clock starts. They know the checks at the end that nobody else sees. And their number is the one your finance team will believe.

Run both for one close. Same month, both methods, compare the minutes and the totals, then retire the old one. It costs one extra half hour, and it is the only way the number means anything.