Say this sentence out loud and notice how unfamiliar it sounds.
My bookkeeper noticed something.
I’ve run businesses for a long time, across a lot of industries, with a lot of different people handling the books. I can count on no hands the number of times someone came to me unprompted and said: hey, this looks off, you should take a look.
It isn’t because they were bad at their jobs. Most of them were good. It’s because nobody hired them to notice. They were hired to record.
Those are not the same job, and the gap between them is where almost all of the money goes.
Recording is a job with an end. Noticing isn’t.
Recording has a shape. Transactions come in, they get categorised, the month gets closed, a report gets produced. There’s a defined start, a defined finish, and a deliverable at the end that either balances or doesn’t. It’s honest work and it’s necessary and it can be done extremely well.
But it’s fundamentally backward-facing and it’s fundamentally finished. Once the month is closed, the box is shut. Nothing in the process asks the question is this normal. It only asks is this in the right column.
Noticing has no shape and no end. It’s not a deliverable. It’s the thing that happens when something has been watching long enough to know what your ordinary looks like — and then something stops being ordinary and it says so.
You can't schedule noticing. You can't close the books on it. It either happens continuously or it doesn't happen at all.
And it’s the only part of the whole arrangement that would actually have changed anything for me.
The one that got me
Here’s the example that made me build this, and I’m going to be careful about how I tell it because I’m not naming the company involved.
A software subscription. A small one — the kind of monthly amount that never registers as a decision. Something we’d been paying for a long time, entirely legitimately, actually using it, no complaints about the product at all.
Then one month it roughly doubled.
On its own, that’s nothing. A small number became a slightly less small number. We’re billed in US dollars, so a move like that reads as exchange rate before it reads as anything else, and nobody launches an investigation into a currency conversion.
The month after that, it jumped again — many times over, far past anything a currency swing could explain.
That one was an error. Not a price increase we’d missed the announcement for, not a plan change, not seats somebody added — an increase with nothing behind it. We found it and we disputed it, because it was not a change anyone had agreed to.
And here’s the part that still bothers me. Nothing in the books was wrong.
The charge went through — correct. It hit my account — correct. It was categorised under software subscriptions — correct. It appeared on the profit and loss under software subscriptions, along with everything else in that category — correct. It was reconciled — correct. The books balanced, every month, perfectly.
An error that balances is invisible.
That’s the whole problem in one sentence. Reconciliation asks whether a charge matches the bank, not whether the charge makes any sense. A thousand-percent jump matches the bank exactly as well as the right amount would.
And if the process can’t raise its hand about an error that size, it’s worth being honest about what it does with everything that isn’t an error at all — the legitimate increases, the quiet more, the changes that produce no discrepancy for anybody to chase.
The gap isn’t skill. It’s position.
There’s a second thing that example taught me, and it isn’t a criticism of anyone who has ever done my books.
I caught it because I’m inside the business. I know what that subscription is, what it’s for, what it has always cost, and that nothing changed on our end to justify the jump. A bookkeeper working from outside has none of that. They see a line item, in a category, that reconciles. There is no version of their job where they’re supposed to know that this particular vendor should not have moved.
So the only person positioned to catch it is the person running the company and the books. Which sounds like an answer until you notice it’s a trap — that person has no spare hours, and the reason they hired a bookkeeper in the first place was to stop being that person.
That’s the real gap. Not a skill gap. A position gap. And the only thing I’ve found that closes it is pattern recognition: something holding the history of every recurring charge and comparing against it, so that knowing what’s normal doesn’t have to live in one overloaded person’s head.
Why a monthly review can’t catch it
I’ve thought about this a lot, and the honest answer is that no reasonable person reviewing the books monthly would have caught the first move. The second one got found because it was extreme. The first one is the one that matters, because the first one is what almost everything else looks like.
Look at any single month in isolation. Software subscriptions: a number. Is that number correct? Yes. Is it wildly different from what you’d expect? No — it’s a bit higher, but categories move around, you signed up for something in the spring, one vendor bills annually, there was a seat added. Every month contains enough normal variation that a small permanent step disappears into the noise.
The step is only visible across time. You need to lay eighteen months next to each other and see the line change level and stay there. Almost nobody does that. It’s not a report anyone runs, because running it requires already suspecting something, and you don’t suspect anything, because nothing looks wrong.
That’s the structural failure of monthly review as a way of watching money. It’s a series of snapshots, and the thing you’re looking for only exists in the motion between them.
The same blind spot applies to a lot more than subscriptions. A supplier who quietly adjusted their rate. A charge that was supposed to be a one-time thing and is somehow still arriving. Two entities paying for the same tool because two different people set it up. Something you cancelled that didn’t actually cancel. A category that’s been drifting upward at three percent a month for a year, which is invisible monthly and enormous annually.
None of these are errors. All of them balance. Every one of them is found by comparison and none of them are found by review.
What noticing actually requires
Once I understood the problem this way, the design of the fix became fairly obvious.
Today’s charge versus every previous instance of that charge. This month’s category versus the same category over the last two years. This vendor’s rhythm versus this vendor’s rhythm.
That means the watching has to be continuous, which means it has to be automatic, which means it can’t be a person. Not because people can’t do it — because no reasonable person is going to run a cross-referenced comparison of every recurring charge in your business against its own history, every day, forever, for the fee you’re paying. It’s not a human-shaped job. It’s tedious in a way that machines are specifically good at and people specifically are not.
This is also why it has to be a daily rhythm rather than a monthly one. I wrote about what changed when checking my books became one morning email instead of a login — the habit is what makes the watching worth anything. Something can notice all it likes, but if it tells you in a quarterly report, you’ve already paid for the thing it noticed eleven more times.
The flag, not the decision
I want to be clear about where this stops, because “AI catches things in your finances” is the kind of claim that gets oversold immediately.
The system notices and raises it. That’s the whole job. It says: this charge is higher than it used to be, here’s what it was before, here’s when it changed. Short. Specific. Not an alarm — most of what it surfaces is completely fine and you’ll clear it in two seconds.
What happens next is entirely you. Maybe the price increase was fair and you’d have agreed to it anyway. Maybe you forgot you added seats. Maybe it’s the one that makes you cancel something you stopped using two years ago. The decision is judgment and judgment is yours, the same way the grey-area calls stay with your accountant and always should.
It doesn’t need to be right about what to do. It needs to be right that something changed, and it needs to tell you near the time it changed.
That’s a much smaller promise than most of what’s sold in this space. It’s also, as far as I can tell, the only part anybody actually needed.
What noticing feels like
The first time something got flagged that I genuinely hadn’t known about, my reaction wasn’t gratitude. It was a slightly sick feeling. Because if that one had been sitting there unnoticed, what else had been?
That feeling passed and what replaced it is the actual product. Not savings — although there were some. What replaced it was the end of a specific kind of low-grade suspicion I’d carried for years: the sense that money was leaving my businesses in ways I couldn’t see and would probably never find, and that being a diligent person wasn’t going to fix it because I’d been diligent the whole time.
I was right about that, by the way.
Diligence was never going to fix it. Continuity was.
There’s a version of running a business where you’re the only thing standing between your money and everything that quietly wants a piece of it, and you’re doing that job in the gaps between all your other jobs, from memory, once a month, tired. I did that for fifteen years. It doesn’t work and it isn’t your fault that it doesn’t.
You deserve to have someone say I noticed something about your own money. At least once. Ideally on a Tuesday, before it’s been happening for two years.
Ledger is the money-watcher I built to be the thing that notices. Not a bookkeeper. A pair of eyes that never looks away from your books and tells you when something changes.