That’s about the inflation rate. For a profession that has raised prices 4% to 5% every year like clockwork, it’s a break in the pattern.

The Financial Times credits AI. I think they’re right. And I think the fee slowdown is the first visible crack in a much bigger shift.

The old pricing playbook came with easy excuses like tighter regulation and too few CPAs to go around.

When clients have to buy your product and your competitors can’t hire enough people to undercut you, a price increase is an easy way to grow margins.

That playbook just stopped working.

A former PCAOB board member signed her first AI audit report

On Episode 506 of The Accounting Podcast, we talked about Christina Ho's next act. Christina is the former PCAOB board member I interviewed on the Earmark Podcast. She built a reputation as the board's most willing dissenter.

She joined Oath, a startup that works as both a software platform and a licensed CPA firm. She helps build the technology and audit methodology.

She just signed her first report and plans to sign an audit opinion later this month.

The model flips traditional audit upside down. Instead of testing a small sample once a year, Oath connects directly to client systems, tests 100% of the data, and verifies records alongside the monthly close.

Think about what that does to the worst part of the job. Audit is painful because firms cram a year of work into a few months (sometimes a few weeks).

Spread that work across 12 monthly closes and busy season disappears.

Saying "AI can't replace professional judgment" misses the point.

Most audit hours have little to do with judgment. We spend them on routine procedures performed the same way every time, with a small dose of judgment mixed in.

That small dose is why rules-based software never automated audit. AI can handle it.

The usual objection is reliability. AI is probabilistic, so it makes mistakes a careful human wouldn't.

Fair enough. But a human team tests a sample, and AI tests everything. Full-population testing, with a human reviewing exceptions and signing off on the opinion, should beat an overworked team checking a sliver of transactions.

We've seen this movie before

I started my firm about 15 years ago, when traditional firms still paid human bookkeepers to key in transactions by hand. Using cloud tools, I cut the labor cost by a factor of five.

Most of an accounting firm’s margin goes to labor. Cut the hours, and you can charge less and make more.

That's how cloud-first firms took market share. Friends of the podcast built practices from nothing into firms valued at $100 million or more over 10 to 20 years.

Audit missed that wave.

The technology that transformed client accounting services didn't apply to audit work. As David pointed out, cloud accounting made bookkeeping far more efficient, so AI adds a smaller bump there.

But audit has seen almost no innovation in decades, which means it has the most to gain.

How I think the market shakes out is the Big Four keep the multinationals. Those audits span multiple entities and still depend on people and relationships.

Everyone below them gets squeezed from underneath.

Small, tech-driven firms take midsize audits from regional firms. Solo CPAs take work from small firms.

That last group interests me most.

Picture a CPA running a team of AI agents, each with a defined role, working inside software built on a rigorous methodology. The agents do the fieldwork and bring questions to the CPA, who acts as engagement partner.

With enough insurance and a system you trust, one person could sign an opinion every week while working a couple of hours a day.

We already know what a tiny firm signing a mountain of opinions looks like. BF Borgers did it with fabricated work. Now you can do it with real audits.

Lower prices could also grow the market. Plenty of smaller companies settle for a compilation or review because an audit costs too much, even when an audit would help them raise money.

Cut the price and some of them will buy.

What to do about it

If you run an audit practice, you can't count on price increases anymore. Your cost structure is the only thing left to work on. Start by mapping which procedures are routine, low-judgment work, and test AI on those first.

If you're a CPA who assumed audit was off-limits without a big team, take another look.

Anyone who launched a cloud client accounting practice with a laptop and a Starbucks table will recognize this moment.

Christina Ho is betting her post-PCAOB career on it.

I prepare my own tax returns every year, including an S-corp and a partnership LLC for my media business.

This year, instead of working through the TaxAct interview myself like I always do, I handed the whole thing to Claude Cowork to see what it could do.

I shared the results on Episode 504 of The Accounting Podcast.

It started the way any good preparer would, pulling the GL from both entities, one in Xero and one in QuickBooks, and reconciling the trial balance and P&L against last year's return.

Along the way, it caught a QuickBooks bug I wasn’t aware of: the software was running an accrual report when it thought it was running cash, thanks to a broken dropdown.

Claude figured that out, reran the correct numbers, and kept going.

Then it built the book-to-tax workpaper, asked me the same adjustment questions any preparer would, and opened TaxAct to start the interview.

This is where it got interesting

It ran TaxAct's built-in error checks, as anyone would. Then it kept going and started catching mistakes in the software's calculations.

It found:

• Understated non-deductible expenses. TaxAct's autofill ignored the fact that entertainment is 100% disallowed under Section 274(a). There wasn't even a line on Schedule M-1 for that addback, so the agent built a custom one.

• Distributions defaulted to zero dollars when the real number was tens of thousands. Left alone, they wouldn’t have shown up on the K-1.

• Ending retained earnings auto-computed to the wrong figure. That one's serious because Schedule L wouldn't have balanced.

• A name-change box was checked incorrectly on the federal form. The agent caught it because the same field was unchecked on the state return, so it fixed the federal one.

• An Arizona e-filing checkbox resets to unchecked every time you run the software. Left unchecked, the state return wouldn't have transmitted.

Claude fixed it all, tied it out, and verified it against a fresh independent read of the GL before I transmitted the returns myself.

Each return took the agent roughly an hour, spread across a few hours of me checking in.

The concerning part for tax professionals

We've built an entire industry on the assumption that tax software gets the mechanics right and preparers supply the judgment calls.

That was never really true.

Preparers have been catching software bugs for years. It's just been invisible, folded into "reviewing the return" like it's part of the job description instead of a sign the tools aren’t truly trustworthy.

Something finally checked the software’s work with the rigor a senior preparer would, at every step, without getting tired and without phoning in the review because it's 9 p.m. during crunch week.

The value of AI is that it never skips a checklist item, no matter how many times it's run that checklist before.

That's what tax software has been missing all along: something that never gets complacent.

Firms treating AI as a way to speed up data entry miss the bigger opportunity

Sure, we can prepare returns faster, but the real value is double-checking returns without cutting corners when the software itself has a baked-in bug.

If your only quality control is a human skimming a printed return before it goes out the door, you're relying on the same weakness that made six separate mistakes in my tax software.

If you're in tax prep and you haven't handed a full return to an AI agent yet, not just for drafting but for working it end to end and checking its own output against the source data, you're missing something bigger than a productivity gain.

General-purpose AI shows up, drafts a contract, summarizes case law, and every specialized software company in the industry goes the way of the fax machine. Why pay for Westlaw when ChatGPT can do the job for free?

Thomson Reuters has an answer. And it's not the answer Silicon Valley expected.

We got into this on Episode 503 of The Accounting Podcast, and Hector Garcia, CPA, joined us for the conversation.

They built their own model

Thomson Reuters launched Thomson, a proprietary large language model (LLM) trained on 175 years of proprietary content from Westlaw, Practical Law, Checkpoint, Reuters, plus input from thousands of subject matter experts.

The project traces back to their 2024 acquisition of Safe Sign Technologies, a UK-based startup that developed legal-specific LLMs.

Thomson Reuters took an open-weight model and fine-tuned it on the library they've spent a century and a half building. It’s the stuff that's always been behind a paywall because it's actually valuable.

Think analysis from top tax and legal experts, not text scraped off the open internet.

The benchmarks are striking.

On instruction following, Thomson Reuters' model reportedly scored 0.914, ahead of Claude (0.861), Gemini (0.848), and ChatGPT (0.885).

On deep research using Westlaw and Practical Law, it hit 0.83 on factuality (meaning claims trace back to a correctly cited source), versus 0.65 and 0.68 for two leading frontier models.

These are self-reported numbers, so take the specifics with a grain of salt until someone independently verifies them. But the direction of the story is hard to argue with.

The "general AI wins everything" thesis assumes data doesn't matter

It does.

A model trained on the open web is a generalist. It knows a little about everything, meaning it's more likely to hallucinate the second you ask it something specific and technical.

As my cohost David Leary said on the podcast, he wants "dumb AI," a model that only knows accounting and tax. Narrow its knowledge, and it's less likely to make things up.

Thomson Reuters didn't stop at the data.

They brought in partner-level practitioners who spent months building rubrics for research tasks, tested outputs against them, and had the model improve from the feedback. Around 1,500 attorney-editors reviewed and rated outputs.

A startup with a slick interface and an OpenAI API key can't replicate that level of intelligence.

You can't rent 175 years of proprietary analysis and a standing bench of experts to grade your homework.

And Thomson Reuters says it's used less than 10% of its own content to train this model so far.

Imagine how the metrics move once they train on the rest.

Data ownership is the whole game

Everyone spent the last few years worrying AI would flatten specialized software’s advantage. Turns out the advantage is the data underneath the interface, and whether you own it.

Thomson Reuters owns theirs. So do other big legal and tax research providers who've spent decades building it. The frontier models, for all their compute and talent, train on what's freely available. And freely available data isn’t the hard part.

The model's entering production now inside CoCounsel, starting with high-volume structured document review, with broader legal and tax integration planned from there.

If you're evaluating AI tools for your firm, this changes the question you should be asking.

Does the tool know your domain, or is it guessing based on whatever it happened to read online?

Where did the training data come from?

Who's reviewing the outputs?

A specialist grounded in real authoritative content beats a generalist every time, especially when the answer has to be right, not just plausible.

When confronted, the AI denied doing any of it and asked for proof.

This story appeared in Accounting Today. Sage CTO Aaron Harris built an agent he named Arthur, and fortunately, this happened while he was testing Arthur on a spreadsheet for a fictional company.

But it’s not a one-off, as David Leary and I discussed on Episode 502 of The Accounting Podcast.

Every “rogue AI” story has a common thread

Ellen Choi, founder of Edgefield Group, built an AI chief of staff she calls TARS.

It mistook a legitimate but unusual purchasing pattern for duplicate payments and recommended auto-refunding thousands of dollars in real revenue. The only reason it didn't happen is that Choi never gave TARS the authority to execute refunds on its own.

Byron Patrick, Senior Product Manager at Karbon, asked his assistant to summarize his thinking after a customer call. Instead of a summary, it created a shared document and drafted a Slack message to his team.

Nothing left the company this time. But nothing stopped it from trying.

I'm not exempt from this, either. I set up Claude on my work account to draft emails and stage them for me instead of sending them. I never applied that same rule to my personal account. Last week I asked it for help with something, and it just sent an email as me.

No draft. No review.

There was nothing bad in it, thankfully. But it wasn't really my voice, and I didn't get a say.

This is a controls problem

We talk about these incidents like they're bugs in the model's judgment. They're not.

Arthur didn't hallucinate a fake invoice. It made a reasonable-sounding call with real access and no one checking its work before it acted.

TARS pattern-matched on legitimate data and had the authority to act on a bad guess until a human thought to take that authority away.

These stories have the same root cause: nobody separated "figure out what to do" from "actually do it."

We already solved this problem. Just not in accounting software.

Coding tools figured it out fast. Claude Code and similar tools default to an "ask, then build" pattern.

The agent proposes the plan, you approve it, then it executes.

After his Slack incident, Patrick’s fix was to adopt a personal standing rule that the AI gives him a plan before it acts. That's the same principle, applied by hand, because the tools don't enforce it for him.

Accounting software hasn't caught up.

I see plenty of "draft this" and "post this" prompts in the tools coming to market. I don’t see the deliberate pause between them built into the product.

It's on the accountant to remember to ask for a plan first, just as it was on me to remember which of my own accounts had guardrails turned on.

That's backward.

An intern doesn't get to void an invoice or email a vendor without a second set of eyes, no matter how confident they sound.

We don't hold AI agents to that same basic discipline.

Your control can’t be hoping the agent behaves

When you evaluate AI tools for your firm or your finance team, the question isn't "does it make good decisions?"

Every agent looks great until the one time it doesn't.

The real question is, what happens between the decision and the action? Is there a checkpoint? Can a human see the plan before it executes? Can you revoke authority for one task without turning off the whole tool?

If the answer is no, you have a liability waiting for its moment. It’s like giving an intern the admin password and not reviewing their work.

We've spent decades building segregation of duties into processes humans touch. AI agents don't get a pass just because they're software.

If anything, they need more separation between judgment and execution, not less. Because it’ll act on a bad guess with total confidence and zero hesitation.

Nobody guesses their way to a 98% hit rate.

My co-host, David Leary, and I got into this on Episode 505 of The Accounting Podcast, and the more we dug into the story, the less it looked like a curiosity.

The numbers behind the pattern

Blockchain forensics firm Bubblemaps traced the activity across dozens of yes-or-no contracts tied to KPMG audit clients, including Wells Fargo, Home Depot, and DoorDash.

The pattern started in November 2025 and netted roughly $22,000.

The same trader also bet on companies audited by other firms and won 82% of those. That’s a strong number, but nowhere near the KPMG rate.

Bubblemaps traced the money flow across 19 different Polymarket accounts. That suggests one person or team deliberately spread bets around to stay under the radar.

Federal investigators are now looking into whether a KPMG employee is behind it.

The Wall Street Journal reported that charges against a KPMG employee could come as soon as this fall. KPMG, meanwhile, says it has zero tolerance for trading on nonpublic client information.

This isn’t just another crazy crypto story

It’s tempting to file this under “weird blockchain gambling thing” and move on. I get it. Prediction markets still feel like a novelty to many professionals.

But strip away the fact that Polymarket is a cryptocurrency-based platform. If the pattern is what it looks like, someone with access to material nonpublic information found a low-friction, semi-anonymous way to monetize it.

And an outside analytics firm had to notice the pattern.

That’s a controls problem.

Audit firms built insider trading safeguards around the channels they know. Employees can’t trade in their clients' stock. They sign an acknowledgment every year.

But those controls assume the leak happens through a phone call, stock trade, or Signal message to a friend.

They weren’t designed with a permissionless betting market in mind. You don't even need a brokerage account, just a wallet and an opinion on whether Home Depot beats estimates next quarter.

This should worry every firm

The same transparency that let someone place these bets is also what exposed the pattern. Every wager on Polymarket is on a public blockchain.

Bubblemaps didn’t need a subpoena to spot the pattern. It needed public on-chain data and the patience to trace money across 19 accounts.

It's reassuring that this kind of activity is more traceable than a whispered stock tip. But it’s also a warning: traceability only works if somebody's looking.

An outside firm only spotted this because the pattern got weird enough to attract attention.

So, what needs to change?

Banning employees from Polymarket is a good first step for firms that haven't already.

But they also need to recognize that new channels for monetizing information keep showing up. Prediction markets won't be the last one.

They need to monitor for this stuff, not just prohibit it and hope.

The old model assumed employees might leak information through the channels you know about, and you watched those channels.

The new model has to assume new channels will appear faster than policy manual updates, and build monitoring that doesn't rely on an outside blockchain analytics firm to do your job for you.

He's not wrong. And the reason he's not wrong should worry every accountant.

My co-host, David Leary, and I dug into this on Episode 501 of The Accounting Podcast.

Here's the deal

Nvidia is backing a massive financing structure with some of the biggest names in private capital: Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. The money, $500 billion of it, is meant to fund AI data centers. Nvidia needs those data centers built because it needs somewhere for its chips to go.

No data centers, no chip sales.

The structure uses special purpose vehicles (SPVs): separate entities that own the data centers, buy the chips, and lease compute power back to Nvidia's real customers, Anthropic, OpenAI, and the rest.

The debt is secured by the compute itself. And Nvidia is guaranteeing roughly 25% of it.

Burry called this an attempt to use "unnatural credits to prolong momentum late in the bull phase," and a "sign of desperation."

He's increased his short positions against major AI companies as a result.

Why does this remind him of Enron?

Same basic move.

Enron used off-balance-sheet entities to keep debt away from its financial statements while inflating the appearance of a whole new investable asset class: wholesale power contracts.

When it collapsed in 2001, its $60 billion bankruptcy was the largest in US history at the time. It's also the reason most of the SOX compliance work you do exists.

Nvidia could build these data centers itself. It's choosing not to. Because if Nvidia builds and owns them, it eats the construction costs and the depreciation.

Structure it through SPVs instead, and Nvidia books the chip sale revenue up front while someone else carries the asset and the depreciation schedule.

That's not the only lever they’re pulling.

AI chips are conservatively useful for two to three years before they're obsolete. Companies are already stretching that estimate to five or six years for depreciation purposes, which lowers annual depreciation expense and inflates reported profits.

The SPV structure is just the next tool in the same box.

None of this is fraud

When David asked whether this points to a problem tied to a specific firm, the honest answer is no. This is happening in plain sight, fully disclosed, structured by some of the most sophisticated capital allocators on the planet.

If anything changes here, it won't be an enforcement action. It'll be a change to GAAP, because GAAP is what currently allows it.

That was true at Enron, too. The scandal wasn't really that Enron broke the rules. It's that the rules had gaps big enough to drive a $60 billion collapse through.

We spent decades building post-Enron infrastructure, PCAOB oversight, SOX controls, new disclosure requirements, specifically to close those gaps.

Now, here's a $500 billion structure doing something that looks a lot like the original problem, and it's legal.

This is bigger than one company's balance sheet

The stock market's growth over the past couple years has been heavily concentrated in a handful of companies, Nvidia among them, that are also deeply entangled in circular AI investment arrangements.

AI companies spend with data center owners, who turn around and invest back into the AI companies. They recognize revenue on both sides of that circle.

Millions of ordinary people hold these stocks through index funds without knowing how much of that growth depends on financing structures like this one.

If accountants and standard setters wait for a restatement or a bankruptcy filing before asking whether GAAP should allow this kind of structure, we learned nothing from Enron except how to recognize the pattern after it's too late to matter.

IAS 1 is gone.

IFRS 18 replaces it, and it doesn't just tweak the edges. It rewrites how entities present the statement of profit or loss, which changes how you read financial statements for years to come.

Start with the classification framework, because it drives everything else. Every item of income and expense now has to land in one of five categories: operating, investing, financing, income taxes, or discontinued operations. Everything goes somewhere.

And here's the part that will bite people. You can't pull something out of operating just because it's volatile, unusual, or non-recurring. If it doesn't meet the criteria for another category, it's operating. Full stop. For any company that's been flexible about defining its own version of operating profit, that's a real change in mindset.

There are also two mandatory subtotals now: operating profit, and profit before financing and income taxes. If you've ever tried to compare two companies in the same industry and found they defined "operating" differently, this is the fix. Both were complying with IFRS. You were still comparing apples to oranges.

This is the first episode of a new show we just launched at Earmark called Standard Practice. Standards aren't just boxes to check, and this show digs into the audit, accounting, and tax rules that shape the profession and what they actually mean when you're the one applying them.

Meredith Mednick, CPA, CA, and I got into all of it in episode 1. And you can earn free CPE credit just for listening.

Register here: https://earmark.app/c/3645

Here's what else we covered:

• A naming trap: The investing and financing categories in IFRS 18 are not the same as investing and financing activities in IAS 7. The IASB acknowledged this would cause confusion and kept the names anyway.

• Management-defined performance measures: Your adjusted, non-GAAP style metrics now come with real disclosure requirements. This pulls management's own numbers into the financial statements in a controlled, transparent way.

• Aggregation and disaggregation: New rules on how much you group and how much you break out. A lot of the judgment lives here.

• What doesn't change: Recognition and measurement are untouched. This is purely presentation and disclosure, which doesn't make the impact small.

• Goodwill and EPS: Goodwill now gets its own line in the statement of financial position, and there are amendments to IAS 33 on what numerators you can use for additional EPS disclosures.

• It takes effect for periods beginning on or after January 1, 2027. That sounds far away. It isn't, if you're the one restating comparatives and rebuilding the mapping from your chart of accounts to those five categories.

How to earn free CPE

Listen to the episode on Earmark, take the quiz, and get your credit. It's free.

Register here: https://earmark.app/c/3645

That's the real question hiding behind the "diamond" org chart everyone in accounting firm management is suddenly talking about.

In a recent piece for CPA Practice Advisor, Luke Frye, CPA and Accountant in Residence at Canopy, argues AI is cutting the traditional firm pyramid into a diamond. A smaller entry-level layer. A fat middle of managers who review what the staff using AI produce. A small layer of partners at the top doing strategic advisory work.

We picked this apart on Episode 500 of The Accounting Podcast with guest co-hosts Adam Zaki (CFO.com) and Adrienne Gonzalez (Going Concern.)

The more we looked at it, the more it didn’t add up.

The setup

Most versions of the diamond chart assume roughly equal numbers of people at the staff and manager levels.

That only holds if nobody leaves the firm.

In a normal firm, people quit, get poached, or burn out. If you shrink the entry-level base and still expect the middle to stay just as fat, you need something to keep refilling it.

A shrinking base can't refill an equal-sized middle forever. That's the math problem.

The rebuttal

A commenter on our podcast livestream had a smart answer to this. Firms don't actually need to hold onto every person to make the math work. They just need to stop promoting managers.

People still enter each stage. People still leave each stage. The firm just stops advancing them upward.

Keep someone at manager for 12 years instead of 5, and the middle of the diamond stays fat without the firm needing anything close to 100% retention.

That fixes the arithmetic. But creates an even bigger problem.

The cost

If a firm stops making partners, why would people stick around?

Managers already have options. Recruiters actively target experienced staff, and AI only makes that talent more portable.

Ask someone to stay at the manager level indefinitely while their skills get more valuable outside the firm, and you shouldn't be surprised when they leave.

But the bigger issue is succession.

Why this matters

The traditional partnership model dies because the partners at the top retire, and nobody underneath them is ready to take over.

That's what happens when a firm stops promoting people through the pipeline that used to produce its next generation of partners.

Once you break that pipeline, a firm's succession plan is selling to private equity.

This is why I think we've been looking at PE roll-ups the wrong way.

Private equity isn't the root cause of what's happening to CPA firms. It's the symptom.

PE shows up because firms broke their own succession pipeline first, whether they meant to or not. A firm that can't promote its way to a next generation of leaders doesn't have many options left when the founding partners want out.

The diamond chart looks like a forward-looking response to AI. Underneath it, for many firms, it's the last few years of a slow-motion decision to sell rather than pass the firm on.

What to watch for

If your firm is talking about a diamond structure, ask what's happening to the promotion timeline for managers. Ask who's being groomed to eventually run the place.

If the honest answer is "nobody, we're just going to keep good people at manager longer," that's not a staffing strategy.

That's a firm already on a path to being acquired - it just hasn't announced it yet.

So where’s the pipeline breaking down? New salary data from Accounting Today has the answer, and it's not the story we usually tell about pay in accounting.

My co-host, David Leary, and I dug into the numbers on Episode 499 of The Accounting Podcast.

The promotion that used to mean something doesn't anymore

Moving from associate to senior used to come with a real pay bump. Not this year.

In New York, associate pay now averages $105,000 and senior pay averages $108,000. That's a 3% raise for taking on more responsibility. It’s the same story in Ontario, where professionals get about a 4% jump from associate to senior.

Compare that to the manager level, where salaries jump by $20,000 to $40,000. Connecticut, Virginia, and New York all report manager averages above $139,000.

That's not a promotion bump. That's a different pay grade entirely.

So what’s going on?

Entry-level salaries had to rise because firms couldn't attract talent without it. Partners keep getting richer too, with equity partner compensation now averaging $839,000 industry-wide.

But the senior level, the people who've already put in a few years and are grinding toward manager, are stuck in between. They’re not new enough for the entry-level raises and not senior enough for the manager money.

As David put it during the episode, partners pocket higher profits, entry-level staff get slightly higher salaries, and the senior staff stay where they are.

Two forces are squeezing that middle tier

Offshoring and AI eat into the work seniors used to do. Less pressure on that work means less pressure on firms to pay up to keep the people doing it.

Meanwhile, those seniors are the people who should be developing into the controllers companies desperately need.

But employers say they can't find controllers and assistant controllers with institutional knowledge, technical depth, and the judgment to influence executive decisions.

That's not an entry-level skill set. It takes years to build, and accounting firms underpay those years, driving talent out of the profession.

Think of it like a bridge with a weak middle span. You can reinforce both ends all you want.

Strengthen the anchor points, widen the approach lanes, make the entrance as inviting as possible. None of it matters if the section connecting them can't bear the weight.

Firms are so focused on winning the entry-level hiring war and keeping partners happy that they're ignoring the span holding the whole structure together.

Corporate finance already figured this out

Average salary increases in corporate finance average 6.7% for executives, 6.2% for directors, and 5.7% for clerical staff, according to Controllers Council.

Thirty-eight percent of finance leaders say they plan to increase hiring this year, up from 24% last year.

Sixty-one percent report a talent shortage, up from 46%.

Corporate pay is up across the board because they recognize the shortage. Public accounting still pays up at the edges and hopes the middle holds.

It won't.

If you're a senior with three or four years in, why wait around for a manager title that may or may not exist by the time you get there? You'd rather jump to corporate now, where the raises are already showing up, or leave the profession altogether.

Every senior who leaves the profession entirely is one less future controller, meaning the talent shortage only gets worse.

What firms need to do

Stop treating the associate-to-senior promotion as a formality. If that promotion helps retain the people who will eventually run your client engagements or become your best clients' controllers, the pay needs to reflect it.

Right now, the data shows it doesn't.

It logs into Xero, pulls the bank statements, and runs the reconciliation report. Matches opening and closing balances on its own.

It works, about 80% of the time. The other 20%, I'm babysitting it. It hits a permissions wall. It gets confused mid-task. It needs me to click through a login flow it can't handle by itself. I'm the only one who knows how to run it, which makes it automation with a chaperone.

That's the DIY tradeoff. Claude and other general-purpose agents are incredibly flexible. You can point them at almost anything. But flexible and reliable are two different problems, and building an agent for one specific job inside one specific system is a different exercise than pointing a general agent at your browser and hoping.

So which one should your firm actually use? A DIY agent you build yourself, or one built specifically for accounting work?

I'm digging into that live with Sasha Orloff from Puzzle on August 13 at 10:00 AM PT. We're comparing DIY Claude agents against purpose-built accounting agents, sharing real examples of both, and talking through where DIY works, where it breaks, and how firms actually get value from AI today.

If you're trying to grow without adding headcount, this is worth an hour.

Register here: https://streamyard.com/watch/hQ4pUFYw7CJq

A growing share of it is AI-generated. And the agency has no reliable way to tell which comments come from real people and which come from a chatbot with an agenda.

My co-host, David Leary, and I discussed the story on Episode 498 of The Accounting Podcast.

The IRS feedback system used to work just fine

Rulemaking in this country runs on public comment. Congress passes a law. The Treasury and the IRS write the regulations that actually implement it. Before those regulations become final, the public gets to weigh in.

That feedback should shape the outcome.

For years, the system had a built-in safeguard against manipulation: laziness.

If someone wanted to flood the comment period, they'd copy and paste the same message over and over. The IRS could spot duplicates instantly and discount them. Bots did this, too. Same text, different name, easy to filter out.

That safeguard is gone.

AI doesn't copy and paste

Feed a large language model one opinion and a prompt, and it will generate a thousand comments that all say the same thing in a thousand different ways. Each one reads as an original, individually written submission.

None of them is.

As David and I talked about on the show, this is bigger than the IRS. Every agency that relies on public comment has this problem: the SEC, PCAOB, and FASB.

So does every professional body that collects feedback from its own members, including NASBA and the AICPA. The comment box has become a place where one person can show up as an army.

Think about what that means for rulemaking

A single person with a laptop can generate thousands of unique-sounding comments arguing for one outcome. Regulators use those comments to gauge public opinion.

If one side looks organic but isn't, the rule gets written to reflect a position almost nobody actually holds. One person, artificially amplified, can outweigh an entire industry's real feedback.

The GAO's response so far is a recommendation, not a fix. It's telling Treasury and the IRS to create policies for reviewing large volumes of identical or near-identical comments. That's a reasonable start. But it assumes the comments will still look similar enough to flag.

AI-generated text doesn't have to.

Here’s an analogy

Twitter, now known as X, used to be a place for actual conversation.

Now, post anything with real engagement potential, and you're buried under bot replies. Many people have stopped reading replies altogether because they know most of them aren't real.

Public comment periods are heading down the same road. Once regulators can't trust the comments, the whole point of collecting them disappears. And once the public suspects the process is already gamed, they stop participating.

That's the real cost here. A process built on individual, sincere feedback breaks down the moment feedback can be manufactured at scale.

Once that trust is gone, it doesn't come back by hiring more reviewers.

So what needs to happen?

Flagging duplicate comments, which is all the GAO is recommending, won't be enough on its own. Agencies need a way to verify that a comment came from an identifiable person or organization, not a script.

That could mean requiring authenticated submissions. It could mean building tools that catch AI-generated patterns even when the wording varies. Either way, agencies need to start treating public comments as something that can be gamed.

We keep hearing that, as AI does more of the work, firms will need fewer people. But look at what's actually happening at companies investing heavily in AI, and the story falls apart.

We got into this on Episode 497 of The Accounting Podcast.

The data says the opposite of the doom narrative

A study from Ramp and Revelio Labs tracked AI spending and workforce records at nearly 22,000 US companies from 2021 through 2026.

Firms that spent more on AI increased total headcount by an average of 10.2% in the two years after rollout.

The heaviest AI adopters (companies in the top third of spending per employee) expanded entry-level hiring by 12%.

Yes, the companies betting hardest on AI hire more entry-level people, not fewer.

So why does the "AI eliminates jobs" story feel so true?

Because it explains half the equation and ignores the other half. AI genuinely does replace tasks like data entry, first-pass categorization, and pulling numbers into a return.

But it doesn't replace judgment. And judgment is the actual bottleneck.

The bottleneck moved

For decades, the constraint in accounting was data entry. Getting the numbers in accurately took most of the time.

AI collapses that step. But someone still has to verify the output.

You can't skim an AI-generated workpaper the way you'd skim a junior preparer's draft, because AI is confidently wrong in ways that look identical to confidently right.

So the bottleneck shifted from preparers to reviewers. And we don't have enough reviewers.

We're short on managers, directors, and partners who have the experience to catch what AI gets wrong. We're short on controllers and CFOs who can sign off. Speeding up the front end of the process just piles more work onto the back end, where qualified people are already scarce.

This is the same mistake the profession made with Excel

We've had macros in Excel for decades, but most accountants never learned to use them.

AI is a much bigger leap in capability, and a much bigger leap in the skills gap it exposes.

If your firm's plan for AI is "let it do the work so we need fewer people," you're solving for a problem the data doesn't support.

The plan should be "train our people to review AI output fast and well," because that's the actual constraint now.

What this means for hiring right now

If you're a firm leader cutting entry-level headcount because "AI will handle it," you're making a decision that the data doesn't back up.

Forward-thinking firms are hiring more junior people and putting them to work reviewing and refining AI output under more experienced oversight.

That's a different job than the one juniors used to do. But it's still a job.

The accounting shortage isn't going away because AI showed up. If anything, AI raised the skill floor for entry into the profession.

That will make the shortage worse unless we start training people for the job that exists now, not the job we assumed AI would eliminate.

Another founder and CEO needed an audit solutions platform. The vendor quoted him $300,000 a year. He built his own with Claude and Retool for under $30,000.

These stories from Accounting Today sound like serious cost savings. But there’s another cost that doesn’t show up on any invoice.

We got into this on Episode 496 of The Accounting Podcast.

The one-person problem

The part of the story that gets glossed over is that the person vibe-coding the solution is often the only one who knows how it works.

Not "most familiar with it." The only person.

That's the risk of vibe-coded software.

I get the counterargument. My co-host David Leary made it to me on the show. These AI coding tools document themselves shockingly well.

Flip into ask mode, and the tool explains how a feature works, even if you never wrote a single comment. The knowledge isn't trapped in someone's head. It's trapped in the code, and the code can explain itself to anyone who asks.

In theory.

In practice, someone still has to know to ask. Someone still has to notice the thing is broken, know where to look, and trust what the AI tells them about a system they've never opened.

How many firms have built that skill? I’m guessing not many.

We found this out the hard way at Earmark

Our AI course generator runs on a Zapier workflow I built years ago. Seventy-five steps. It worked great until it didn't.

It broke, and nobody else on the team could fix it. Nobody even knew it was broken, because nobody but me was positioned to notice.

That's a no-code tool, not custom software. It still created a single point of failure. Now scale that up to an audit platform that a Top 100 firm depends on every busy season.

Think of it like the financial model nobody wants to touch

You've seen this before. Someone builds an Excel model so complex, with so many interlinked workbooks, that eventually they're the only one who understands it.

Everyone else is scared to click the wrong cell. The model works fine, right up until that person is out sick during crunch time or has left for a different firm entirely.

Vendors sell you something else along with the software: a team that maintains it, documents it, and answers the phone when it breaks. That's the actual product, not just the code.

Vibe-code your own replacement, and you get the code but skip the team. You save the fee but inherit the redundancy problem.

To be fair, buying from a vendor isn't automatically safer. We recently lost data because our podcast streaming platform’s interface was poorly designed, and its support team spent more time deflecting blame than fixing it.

There are vendor risks, too. But there's a company you can point to, a contract, and, in theory, more than one person who understands the product.

So what do you actually do with this?

If you're going to vibe-code a tool your firm depends on, keep in mind the $30,000 isn't the whole cost.

Budget for redundancy from day one. Document it properly, not with AI-generated comments buried in the code, but with a plain-English explanation a second person could follow under pressure.

Better yet, train a backup person to actually understand it, not just to ask the AI what it does.

The tools are now good enough that almost anyone can build something. They aren’t so good that just anyone can maintain what someone else built once that someone is gone.

People are asking, “If AI does all the grunt work, how will junior staff ever learn?”

It's a fair question, but it's the wrong one.

The right question is, “Now that we have this powerful new tool to train our staff, how do we use it?”

KPMG might have the answer

KPMG is testing a new training tool called TaxSIM, developed with Centaurian AI.

Instead of having new hires grind through thousands of hours of real client prep work to build judgment, TaxSIM throws them into rapid-fire simulated tax scenarios.

Centaurian's CEO describes it like a racing simulator. Users hit scenario after scenario, get feedback, and improve fast. The system is designed to make staff reason through the problem themselves before it lets them lean on AI assistance.

It creates friction on purpose because difficult practice builds skill.

This grind was never a good teacher

The profession has spent the last year treating the loss of "grunt work" as an existential threat to staff development. If AI drafts the memo, runs the reconciliation, and flags the exceptions, what's left for a first-year to do?

The assumption baked into that question is that grunt work was a good training method in the first place.

I don't think it was. It was just the only training method we had.

Think about how we used to teach anything before we knew better

A couple hundred years ago, if you wanted to learn Latin, history, or rhetoric, you translated one text into another language, over and over, for years. That's how John Adams, our second U.S. President, educated his son, John Quincy Adams. It worked, eventually. It also took a lot of time, and it worked mainly because nobody had invented a faster way yet.

Public accounting's version of that is the up-or-out staff accountant model. Grind through years of prep work, absorb judgment by osmosis, and hope you survive long enough to become the person reviewing someone else's grunt work instead of doing your own. It's slow and tough on retention, but we kept it because we hadn’t seen a better alternative.

Now we have one.

Simulation training already works everywhere else

Pilots don't build judgment by flying thousands of real passengers and hoping for the best. They train in simulators that compress years of edge cases into weeks.

Chess players use rapid drilling tools to practice specific endgames, not just play full matches.

Poker players study solved hands instead of only learning from live games over a decade.

Tax work is no different.

A junior staffer using TaxSIM could encounter more varied, high-stakes tax scenarios in a few months than they'd see over three years of real engagements, because real engagements are mostly the same handful of situations over and over.

Simulation lets you compress the hard cases into a training loop instead of waiting for them to randomly show up on a real client's return.

So why hasn't this happened already?

Honestly, it should have started in college. Accounting programs have every opportunity to build simulation-based judgment training into the curriculum, and mostly haven't. That's on the institutions, not on AI.

KPMG building this internally signals firms are done waiting for academia to catch up.

The transition still isn’t painless

If AI does the sampling, drafting, and reconciliations, firms have to be deliberate about what replaces the reps that work used to provide.

Bolting a simulator onto an otherwise unchanged pyramid structure won’t automatically produce better judgment. It has to build reasoning, not just look like a video game with a CPE label on it.

But the fear driving most of this conversation, that AI is going to strand a generation of accountants without a way to develop skills, assumes the old way was working well.

It wasn't.

Job satisfaction already drops the further you climb the traditional ladder. The grind was never the point. Building judgment was the point, but we didn't have a faster way to do it.

Now we might.

If simulation training compresses years of pattern recognition into months, KPMG can build better auditors and tax pros faster than the traditional model ever could. And they'll do it without asking staff to sacrifice their twenties, their sleep, and their sanity to climb the ladder.

That should worry the firms still betting on the old apprenticeship model. It should excite everyone else.

My co-host David Leary and I covered the recent IRS announcement on Episode 494 of The Accounting Podcast.

If you’re still charging by the hour for tax prep while using AI, you might accidentally be violating Circular 230.

The IRS drew a line on AI billing

The IRS Office of Professional Responsibility recently issued its first-ever AI guidance for tax practitioners. Most of it is unsurprising. Verify what the AI produces. Don't hand off your professional judgment to a chatbot. Protect client data. Circular 230 already required all of this.

But buried in the bulletin is a sentence that should put the nail in the coffin of hourly billing for tax work. The OPR said charging for time not actually spent because AI made you faster could amount to "an unconscionable fee." Cost savings from AI, it said, need to be reflected in what you bill the client.

In other words, if AI cuts your return prep time in half, and you still bill the client for the hours it used to take, you're now on the wrong side of Circular 230.

Timesheet padding has always been unethical. But with this new bulletin, the IRS explicitly states AI efficiency gains belong to the client, not the firm.

Nobody bills hourly anyway. So why does this matter?

You might think, fine, I don't bill hourly anyway. Neither does almost anyone else.

Last year, Ignition surveyed firms using its platform and found that just 3% of tax prep engagements are still billed hourly. The rest have moved to fixed fees, value pricing, per-form pricing, or some blend of these.

Admittedly, that result is skewed because it only included Ignition customers, and firms using online proposal management software tend to be more forward-thinking.

A separate National Society of Accountants survey puts hourly billing higher, around a quarter of more traditional firms, but even there, it's a minority practice.

So why does an IRS bulletin about hourly billing matter? Because it removes the last argument for the holdouts.

Firms still billing by the hour usually say it's simpler, it's what they've always done, or they're worried fixed fees leave money on the table on complex returns.

The IRS just changed the math. Keep billing hourly, keep using AI, and you now have a regulator on record saying your billing model itself creates compliance exposure.

Fixed fees were never just a pricing philosophy

I've made this case before, and the IRS just handed me another reason to make it again.

Fixed and value-based pricing are the only pricing models that don’t put your firm in direct conflict with your technology.

When you price the outcome instead of the hours, getting faster with AI is a win for your margin and your client's experience. When you bill by the hour, efficiency threatens your revenue.

Where to start

If you're one of the firms still billing hourly for tax prep, you don't need a consultant or a six-month transition plan. Pick your highest-volume, most standardized return type. Individual 1040s are a good place to start. Quote it as a fixed fee for next season. Use this year's actual hours as your baseline, and price to a margin that works even after AI shaves time off the job.

Hourly billing in tax prep was already a shrinking minority practice. Now it's a compliance risk, too.

Here's what he means. AI is going to eat the mechanical work — the bookkeeping, the categorization, the reconciliation. That's not a threat, it's the point. Once the mechanical work is gone, what's left is the stuff that's always been the real product: trust, judgment, and relationships.

The firm he's building, United Accountants, runs on an AI operating system that layers on top of whatever software a firm already uses. No rip-and-replace. CPAs keep their own local brand and hand off the grunt work to the machine.

Karl and I dug into all of it in episode 115 of the Earmark Podcast. And now you can earn free CPE credit just for listening.

Register here: https://earmark.app/c/3507

What else is in this episode

• The "first big small firm": Karl's model for scaling a practice without turning it into a faceless national brand.

• Outcomes-based pricing: Why the billable hour makes less and less sense once AI is doing the mechanical work, and what should replace it.

• The pyramid problem: What happens to a firm's structure when the junior tasks that fed the bottom of the pyramid get automated away.

• Who actually gets replaced: It's not the accountant who uses AI. It's the one who pretends it isn't coming.

How to earn free CPE

Listen to the episode, take a short quiz in the Earmark app, and claim your credit. That's it — free CPE for the time you'd spend learning anyway.

Register here: https://earmark.app/c/3507

Something doesn't add up.

On Episode 493 of The Accounting Podcast, I walked my co-host, David Leary, through a write-up project I'd been dreading. Two years of transactions, nine bank accounts, a mix of PDFs and CSVs. The kind of job that used to take days.

I pointed Claude Cowork at a folder of statements, gave it the chart of accounts, and asked it to start coding.

It pulled every transaction. It scanned and converted the PDF bank statements when no CSV files were available. It merged everything into a single file per account, then coded 2,200 transactions and flagged the ones it wasn't confident about, so I could review only those.

It even found a business line of credit I didn't know existed, buried inside a clearing account, and asked me what to do about it before fixing it.

The entire project took about four hours, including gathering the statements. Out of 2,200 transactions, it missed six. Two were genuine errors. The other four were a quirk in how Xero's import tool merges duplicate references.

A project like this used to take at least a few days. Maybe a whole week. So what happens to pricing when the work gets five to ten times faster?

The numbers don’t match the moment

CPA Trendlines reports firm pricing is up 4.2% year-over-year, reversing a 2.1% decline the year before. Tax prep and planning are up nearly 8%. Advisory is up over 6%. Audit, where AI adoption has been slower, is only up 2.3%.

Tax, one of the segments getting the most help from AI, is raising prices the fastest. The example I gave above is for bookkeeping, but we’re seeing similar productivity gains in tax. AI tools can now ingest an entire folder of client-provided documents and generate a return in minutes. This used to take hours to do by hand.

If we can do tax and bookkeeping work in minutes instead of hours, you'd expect that efficiency to show up as lower prices, just as it did when cloud accounting cut data entry time a decade ago.

Instead, firms are charging more for tax prep and advisory services than they were two years ago.

So either AI isn't saving as much time as people claim, or something else is going on with how clients value this work.

I think it's the second one.

Clients were never paying for data entry

We assume clients pay for the labor of preparing a return or reconciling a bank account. They don't. They pay for the answer at the end. They want to know the numbers are right and get advice on what to do next. They want someone they can call when they get an IRS letter.

Data entry was never the product. It was the cost of producing the product.

When AI chews through 2,200 transactions, it replaces the busywork that used to stand between me and the value I provide. The client doesn't care whether I spent 40 hours or four hours getting to a clean set of books. They care that the books are clean and that I can tell them what the numbers mean.

That's why tax planning and advisory pricing are outpacing audit. Clients pay for judgment, not labor. AI now handles the labor. The judgment is still scarce, and scarce things get more expensive, not less.

The panic is pointed at the wrong target

A lot of the conversation in our profession right now is fear dressed up as analysis. AI is coming for bookkeepers. AI is coming for staff accountants. AI will compress fees because the work takes no time.

I don't think that’s the whole story.

The work that's disappearing is the work nobody wanted to do in the first place. The work that's increasing in value is the work that requires a human brain that understands the client's business and can make a judgment call.

If your firm's pricing model has always been "hours times rate," AI is a real threat. If your value has always been advice, AI just freed up the time to deliver more of it.

What this means for your firm

If you're still pricing write-up and bookkeeping by the hour, it’s time to have a hard conversation with yourself about margin. The hours are shrinking. Either your price shrinks with them, or you find a way to charge for the outcome instead of the time.

If you're already positioned around advisory, the data is on your side. Clients are willing to pay more for tax planning and advice, even as we automate the mechanical work behind it. The market tells you what it actually values.

The fear that AI is going to gut accounting fees has it backward. AI will gut the fees for work clients never wanted to pay for anyway. The advice was always the product. Now we have time to deliver more of it.

Then it came back in 2026.

Shelly Weir, President & Chief Executive Officer at the Florida Institute of CPAs (FICPA), gave her first public interview about the fight on Episode 492 of The Accounting Podcast. Her story deserves the attention of every CPA in the country.

What the Bill Actually Said

This wasn't a small change. It was a comprehensive deregulation bill targeting 25 licensed professions in Florida, including architects, engineers, veterinarians, harbor pilots, realtors, and CPAs, all in one package.

The goal was to eliminate the regulatory boards for each of these licenses, remove all continuing education requirements, and establish a pathway to delete the education requirements for licensure altogether.

For CPAs, that would have meant the end of the Florida Board of Accountancy. It would have meant the end of CPE. And it would have created an opening toward awarding CPA licenses without a college degree.

Why Didn't You Hear About It?

FICPA deliberately chose to keep this out of the national accounting press.

They worried a fractured response from the profession would have been harder to manage politically. They needed unified messaging, targeted lobbying, and the ability to move fast without noise complicating the strategy.

So they worked the political trade press in Tallahassee, where legislators and their staff were paying attention. It was the right call for winning the immediate fight.

But it also means most CPAs across the country still don't know this happened. And if it happened in Florida, it can happen anywhere.

How They Beat It

FICPA deployed nine lobbyists, two public affairs firms, voter polling data, and a member activation campaign that had CPAs tracking down college roommates, sisters, and church friends of specific House and Senate members to make the case directly.

FICPA also looked in the mirror to see how it could get ahead of the deregulation effort by proposing sensible changes on it’s own.

It put together its own modernization legislation: a streamlined reciprocal licensure process, simplified mobility rules, and targeted CPE reforms.

They gave legislators a version of deregulation the profession could actually support. And they walked into committee hearings as the one profession in the bill that had already done the work to modernize.

The Real Risk

If the bill had passed, the immediate concern would have been mobility.

The CPA profession has something most other licensed professions don't: a functional interstate mobility system that lets CPAs practice across state lines without redundant licensure.

It's not perfect, but it works. Florida breaking from that system would have created new barriers.

The longer-term risk was bigger. As Shelly explained, there was no credible assurance that eliminating the board was the end of the road.

What This Means for the Rest of Us

Deregulation pressure is bipartisan. Republicans are pursuing it as a free-market and workforce initiative. Democrats are pursuing it on the grounds of access and equity. Both want to lower barriers to entry.

That's not going away.

And that’s why Florida could be a preview of what’s to come.

State societies that aren't already thinking about this are behind. The profession's best defense is showing up with credible, self-directed reform before someone else forces it.

What's your state society doing to modernize the license before someone else decides to do it for you?

Within weeks, employees figured out how to game it.

It's not a glitch. It's the natural byproduct of an accounting culture that rewards 'looking busy' over achieving results.

David Leary and I broke down what happened on Episode 491 of The Accounting Podcast.

Token Maxing Is the New Billable Hour

KPMG rolled out an AI usage dashboard in late 2025. The target, according to reporting we covered on the show, was roughly 75% of the week. Basically, four out of five days a week, you need to be in an AI tool, doing something.

So what did KPMG's people do?

They had AI summarize emails they'd already read. They asked it to make drawings. Anything to show activity on the dashboard.

This includes employees who actually believe in AI. People who want to use these tools well are now wasting time performing usage for a metric.

There's a name for this now: token maxxing.

It's not just KPMG. Amazon ran an internal leaderboard tracking AI usage across its workforce. They shut it down for the same reason. People were burning tokens to climb the rankings rather than to solve problems.

This is the billable hour in a new disguise.

For decades, accounting firms have measured value by counting hours. Not outcomes.

You know what that produces. Staff who pad timesheets. Staff who work an extra fifteen hours a week and call it normal. Staff who burn out and quit, which is a big reason we don't have enough senior accountants today.

Now, firms are taking that same approach to AI. But it doesn’t work.

What Software Companies Already Figured Out

Picture a software company telling its engineers, "I want you writing code with GitHub Copilot for six hours a day. We're tracking it."

What happens?

Engineers accept every suggestion Copilot offers, even the bad ones. They generate code they don't need. They inflate their "AI-assisted" stats because that's what gets measured, and what gets measured gets rewarded.

Meanwhile, another engineer uses Copilot for fifteen focused minutes to solve a hard problem, then spends the rest of the day thinking, reviewing, and talking to the team.

They look lazy on paper. But that’s the person making a difference.

Jeff Seibert, CEO of Digits, said he hasn't written a line of code himself since December.

This is a guy who taught himself to code at age 12 and says his passion is coding.

Claude Opus writes almost all of it now. His job changed from doing the work to guiding the agents and reviewing the output.

Nobody tracks how many hours Jeff spends "in Claude." Nobody needs to. What matters is whether the software shipped and whether it works.

The metric is output, not activity.

Firms that measure activity will train their people to fake activity. Firms that measure outcomes will actually benefit from the tool.

Measure the Work, Not the Tool

If you're a firm leader who's already built (or is tempted to build) an AI usage dashboard, here's your first step.

Throw out the usage metric.

Don't ask "how many hours did this person spend in an AI tool this week?"

Ask "What got done that didn't get done before?" instead.

Otherwise, you'll get the appearance of progress and none of the substance.

When CBIZ acquired Marcum at the end of 2024, the stock was at $78. It climbed to $90 in early 2025, right around the peak of PE deal velocity in the profession.

By the end of March, it had fallen to just under $27. It's recovered somewhat, sitting around $35 as of this writing. But that's still less than half of what it was at the top.

To make matters worse, CBIZ anticipates just 2% to 5% revenue growth for 2026.

Factor in inflation, and that's basically flat.

When David Leary and I broke down these figures on Episode 490 of The Accounting Podcast, we reached the exact same conclusion: the traditional private equity consolidation playbook is running out of runway.

What Is the Market Saying?

Gary Shamis, CEO of Winding River Consulting, wrote a piece in Accounting Today framing this as the market pricing in execution risk, including integration challenges from the Marcum deal, investor concerns about leverage, and limited organic growth.

He's right about all of that. But investors are starting to register a deeper problem that no amount of better integration can fix.

The PE playbook for accounting goes like this: consolidate firms, create scale, squeeze out inefficiencies, and flip it to the next buyer.

It makes sense if the underlying business model holds up. But AI puts pressure on that model.

Billing for hours and running a headcount-heavy pyramid structure made sense when scale required bodies. That assumption is breaking down fast.

The Competitive Threat

Mid-size firms like CBIZ aren’t doing Fortune 500 audit work. That's still the Big Four's territory. So they can't go up-market.

But they can absolutely lose work to smaller, faster firms.

AI is breaking the link between headcount and capability.

A 10-person firm using AI can handle work that once required a team of 100. And I’ve written before about solo firms that do serious volume with almost no staff.

Mid-size firms sell clients on the breadth of their services and on having specialists in every area.

But small firms can now access deep expertise across disciplines through AI. That competitive advantage is shrinking fast.

Suddenly, thousands of smaller, leaner competitors can go after work that used to flow only to the CBIZes of the world.

That's a big wave of new competition hitting exactly the market segment where mid-size consolidators live.

The Billing Model Problem

There's another wrinkle.

Even if a mid-size firm rolls out AI successfully and its people become dramatically more productive, how do they capture that value under a billable-hour model?

They don't. They get paid less.

If it takes half as long to do the work and you're billing by the hour, your revenue drops by half. The only way to benefit from AI productivity is to switch to fixed-fee or value-based pricing. That's a fundamental change to how a legacy firm prices and sells its services.

Large, PE-backed firms with investors expecting predictable quarterly returns are poorly positioned to make that transition.

Private equity is good at finding operational efficiencies within an existing model. It's not designed to reinvent the model from scratch.

What Happens Next

Massive firm size won’t be the winning hand over the next decade.

The future belongs to smaller firms that can move first on pricing, rethink their staffing, and use AI to do more with fewer people rather than just layering it on top of an unchanged operation.

Small and mid-size firms willing to pivot have a real shot at taking clients and talent from the large regional players.

That's a risk investors in PE-backed accounting firms are just starting to think through.

CBIZ's stock decline is the first visible data point. It won't be the last.

To hear more of our analysis, check out Episode 490 of The Accounting Podcast.