You have had AI in your practice for a year or more now. Ask where the time went, and see whether anyone can answer. The monthly close still takes about as long as it did. Somewhere in your firm, someone has already said the quiet version out loud: I am not sure this is actually doing anything.
My wife and I were married in June of 1996. We celebrated thirty years this June, and I am writing this from Alaska, hours from boarding a ship with no signal for a week. Thirty years is long enough to have learned something about investments whose return you cannot point at.
I watched a firm decide it was finished
In the late 1990s I worked on a global audit platform for one of what was then the Big Six. My part was policy: sitting with international partners while they set it, building it into the software, then helping roll it out.
Understand what that changed. Before, policy was a manual you hoped people read. After, it was in the form: an auditor in a city I had never visited, who had never heard of me, could not skip the step. That is not the old work done faster. It is a capability that did not previously exist at any speed, and it takes as long to execute, or longer.
We did speed some things up. But the larger share of what we gained went somewhere else: into work we should always have been doing and never had, because it was too hard to do by hand. The quality of the audit went up. The clock did not move.
The meter said zero, and the meter was not lying
Leadership was watching the clock. They were looking at a very large investment: development, licences, hardware at a time when firms were still absorbing one laptop per staff member, and distribution on floppy disks and then CD-ROMs to tens of thousands of auditors. Against all of that sat a cycle time that had not improved.
They were not being stupid. On the instrument they had, the reading was genuinely zero. No report anywhere has a line for "the same judgment applied by someone you will never meet."
So the budget was scaled back. By the time I left in 2001 the platform was in maintenance mode, which is a polite way of saying the firm no longer intended to go anywhere with it. I left before the end of the story, so I cannot tell you how it finished. I can tell you how the decision was made, because I was arguing the losing side.
Two explanations that feel identical from the inside
Here is the part I have thought about for twenty-five years. When the gain goes into new work rather than saved time, there are two possibilities.
The first is that the new work was genuinely worth doing and was previously impossible. The return is real; you are simply reading the wrong instrument. The second is that you became more thorough because you could, not because anyone needed it. There is no return, and you have quietly raised your cost base.
From the inside, those two feel exactly the same. Both feel like doing better work, and both are defended sincerely by the people doing them. Nobody in that room could tell them apart, including those of us arguing hardest for the money. The case for funding did not lose to ignorance. It lost to an absence of evidence we had failed to produce.
Your firm is running the same experiment right now
The hours are refilling: with review, with verification, with standing behind a number a machine produced, and with client expectations that rise to meet whatever you have just become capable of. That is not waste. Some of it is the most valuable work your firm does, and where that freed capacity goes is a separate question from whether you can see it.
But it means your meter reads zero. And somewhere in the next few quarters, a reasonable person at your firm is going to look at the subscriptions, the training, the time, and the unchanged capacity, and conclude that this did not work.
Name the currency before you spend it
There is one piece of discipline I would ask of you, and it takes a single conversation. Before you deploy anything, decide out loud which currency you intend to take the return in: margin, capacity, or quality. If the answer is quality, name the specific thing you will now do that you do not do today.
Do that, and someone who was not in the room can check it in ninety days. Skip it, and every AI investment in your firm eventually resolves to "it feels better" against "the hours did not drop," and the second one wins, because it has a number attached.
Thirty years of marriage is the other kind of investment: nobody can show you the return, and you do not fund it because the return is provable. You fund it because the alternative is a disadvantage you cannot measure either. The difference with your practice is that you get to choose the meter first. Choose it before someone else reads it for you.
Not sure which currency your firm is actually banking? The AI Readiness Scorecard walks you through where your practice stands across five dimensions, in about ten minutes: theaiaccountant.ai/scorecard
Next Friday I will get to the harder question, and the one I have failed twice: whether any of it survives your busy season.