Private equity bought accounting because it looked like a fortress. Clients that never leave, work that law requires, recurring revenue measured in billable hours.

Then those same funds started installing AI on top of it. And AI erodes the exact thing that made the fortress worth the price.

They bought the asset right before dissolving it themselves

This is the contradiction almost nobody is saying out loud, and Footnote lays it out well: the multiple a fund pays for an accounting firm is justified by stable cash flows generated by labor-intensive work. That labor is the asset. The AI the same fund is financing turns that labor into a software function.

AI is not going to bankrupt accounting firms. It is going to make cheap what was bought expensive.

And who absorbs the gap is already decided. It is the partner who left 20% to 40% of their stake in the platform as rollover equity, tied to EBITDA targets over two to four years and a three- to five-year employment agreement. If the multiple compresses because the work got cheaper, the seller already banked the cash at closing and the capital that stayed inside takes the hit.

The actual scale of the consolidation

The numbers matter because they explain why this is not one isolated deal.

Between 2015 and 2025 there were 177 direct private equity investments in accounting, and those triggered 875 follow-on acquisitions, touching more than a thousand firms. In 2025 alone, roughly 200 direct investments spawned about 900 downstream deals. The multiplier effect has risen about fourfold since 2021. By early 2026, around half of the 30 largest US accounting firms already held PE money or an alternative practice structure.

The vehicles have names. Thrive Holdings, the spinout from Joshua Kushner’s Thrive Capital, committed $1 billion to buying local practices. Current, formerly Crete Professionals Alliance, assembled close to 50 practices backed by SoftBank, Altimeter and D1, and is seeking roughly $2 billion.

The most telling detail is elsewhere. OpenAI took an ownership stake in Current and assigned a team to train agents on tax work. The model maker owns a slice of the firm it is going to replace with that model, and its stake grows as the portfolio grows. Current reports 31% time savings per return and 7,000 returns processed through AI last season.

Thirty-one percent time savings sounds like good news right up until you remember how the business charges.

The legal structure of these deals shows exactly where the risk sits. A fund cannot own the audit practice. That work is reserved for licensed CPAs. So the firm gets split in two. The attest entity stays with the licensed partners, and everything else, tax, advisory, payroll, consulting, is sold to the platform. Capital buys the side that regulation does not protect, which happens to be the side that automates most easily. That is not a coincidence. It is the same line drawn twice.

There are already signs the arithmetic is tightening. Per The Finance Story, the EBITDA multiples being paid in these deals have climbed to levels that leave little room for error, at the exact moment the cost base of the purchased work is about to move down. Paying up for an asset that is getting cheaper is a defensible bet if volume covers the difference. It stops being one if the volume also depends on billing by the hour.

Why hourly billing makes AI investment irrational

Here is the mechanism, and it applies equally to accounting, legal, and consulting.

If your revenue is tied to time worked, every hour AI saves you is an hour you stop billing. You buy the tool, pay for integration, train the team, and the direct result is charging your client less. Under that model, automating is not an investment. It is a decision that destroys your own revenue.

No firm that bills by the hour solves this with better technology. It gets solved by changing how you charge, and that is an owner’s decision, not an IT decision. Which is exactly why permanent capital walks in so easily: a fund can absorb gross margin compression in exchange for volume and consolidation, because it does not live on this quarter’s hours.

We wrote about this from the other side of the counter when Microsoft added Accenture to its Frontier program and the stock fell anyway. The market was not punishing the quality of the work. It was repricing a revenue model.

What changes for you when your provider gets acquired

This stops being industry news the moment your accountant, your law firm, or your consultancy joins one of these platforms. Three things change that actually touch you.

The person you talk to still signs the deliverable. But a growing share of the execution now runs through an agent, and that is not bad by definition. It is something you should know before you renew, not after.

Then there is where your data ended up. Your books now sit inside a platform running models across hundreds of clients’ information at once. Ask what gets used for training and what does not, and get the answer in writing.

Pricing is where almost nobody looks. If the firm keeps billing you by the hour while automating internally, it keeps the entire efficiency gain. That may be completely fine. What it should not be is a default nobody ever discussed.

When we covered Blackstone’s bet on Norm AI in legal work and the 21 skills that turn Claude into a strategy consultant, the pattern was already visible: professional work is being repackaged as software, and price follows the new cost structure on a lag.

What we look at with a firm that sells services

If you run a firm that bills by the hour, the question is not which AI tool to buy. It is which of your deliverables you can sell at a fixed price without going broke, and that gets answered with data you probably already have.

You need to know how long each repeatable deliverable actually takes today, the spread between the fast case and the slow one, and what share of that time is professional judgment versus data capture and verification. Only the second part automates cleanly right now. The first is what you keep selling, and at a better price, if you manage to separate them.

Separating them is literally what the AI Maestro discovery phase does: measure the real processes before deciding what gets automated, with a Process Reality Map and a go or no-go gate at the end. Not because the technology is hard, but because nobody should repackage their revenue model on a hunch.

Consolidation is going to continue. What is not decided yet is whether your firm arrives at that table with a pricing model of its own, or with a multiple calculated on hours AI is already erasing.

Measure your processes before you reprice them