Your Client Asked ChatGPT at Midnight. You Answered Tuesday.

There's a piece in Inside Public Accounting this month with a phrase I haven't been able to put down. The "Happy Historian" firm.

That's a firm whose product is the past. Collect the client's numbers, organize them, report what happened. How much did you make, what do you owe, sign here. It's honest work and it's most of the profession.

The argument is that those firms are about to get rolled, and not because a general model is a better accountant. Because the client can now ask a question at midnight and have an answer in four seconds.

And then they ask you why you never brought it up.

You're not losing the client. You're losing the question.

That distinction matters, because the two get solved differently.

Dext surveyed 500 UK accountants and bookkeepers through Censuswide in December, and the numbers are worth reading even though it's a UK sample and I wouldn't run them straight across to a US practice. 77% had seen an increase in clients using public AI tools for financial guidance. 72% had clients use AI output to question or challenge their professional advice. 68% had clients suggest the AI could replace the need for their services outright.

The mechanism travels even if the percentages don't, because none of that is about tax rules. It's about who was awake.

Your client had a question on a Sunday night. You have a process for that, and the process is that they email you and you get to it Tuesday. That was a perfectly good process for thirty years. It stopped being one when a competent-sounding answer became free and instant.

So the question goes somewhere else. And a question you never got asked is a relationship you're not in anymore, even while you're still doing the returns.

Then you clean it up for free

This is where it turns from annoying to actually expensive.

In that same survey, 31% of accountants were seeing client mistakes from bad AI advice on a weekly basis, 7% daily. The most common one, at 46%, was clients getting business expenses wrong. Then VAT, then personal tax planning, then payroll.

And somebody fixes those. 44% said they spend up to three hours a month cleaning up AI-generated advice, and 39% said four to ten hours. Half of them knew of a business that had taken a direct financial loss from it.

So walk the whole loop. The client asks the machine because you weren't available. The machine gets the expense treatment wrong because it doesn't know the business. You find it two months later and fix it, unbilled, because "you should have called me" isn't an invoice line.

You lost the question and then you ate the cleanup. That's the actual shape of what's happening, and it's not showing up as churn yet, which is why it's easy to miss.

It's not only accounting either. A survey of 1,000 business owners by UPrinting found 25% had lost business in the past year because customers used AI tools instead of paying for the service. Anybody whose product is knowledge is in this.

"Be more human" is not a strategy

The standard advice here is to lean into the relationship. Be the trusted advisor. Be the one who picks up the phone.

Yeah. That's true, and it's not enough, because it doesn't touch why they went elsewhere. Nobody opened ChatGPT looking for warmth. They opened it because it was open.

So the firms that win this don't out-human the model. They out-context it.

A general model answering a tax question knows tax. It does not know that this client took a big distribution in Q2, that their revenue is seasonal and Q4 is 40% of the year, that they've got equipment coming in the fall, that you already had this conversation in April and they said no. Your answer, with those facts in it, beats the generic answer every single time. It isn't close.

The problem is your answer takes three days to produce. Because producing it means opening the accounting system, then the practice management tool, then the spreadsheet somebody on your team maintains, then your own memory of a call five months ago.

That's the gap. Not knowledge. Retrieval speed. You're losing a race you'd win on the merits because the starting gun costs you three days.

Close the retrieval gap

That's the thing we're actually trying to solve. Liaison sits on top of the systems a firm already runs, so the client's whole picture becomes answerable in one place. The ledger stays where it is, the practice management tool stays where it is, nothing migrates, and none of your people have to learn a new system to keep doing their job.

What changes is that "what's changed for this client since last quarter" stops being a research project. You ask, you get the answer with the records attached, and it takes seconds.

Do that and the whole dynamic inverts. You're not the firm explaining why the ChatGPT answer was wrong in November. You're the one texting them in September about the distribution, before it occurs to them to ask a chatbot.

That's the game. The Happy Historian is in trouble because history became a commodity. Context didn't. You're sitting on years of it, and right now most of it is locked in four systems that don't talk, which is a tooling problem, not a talent problem.

If your firm knows more about your clients than you can get to on a Tuesday, that's worth a conversation.