Paychex

Paychex does the payroll for about 800,000 small and mid-sized American businesses — roughly 840,000 customers in total, supporting some 2.6 million worksite employees. It calculates the pay, withholds the taxes, files the returns with the IRS and fifty states, and administers the benefits. It also runs over 130,000 retirement plans holding $66 billion of participants' money.

The shape of the revenue is the appeal. Every employee, every pay period, forever. It is not a discretionary purchase — a business with staff must run payroll, and it must get it right. Paychex is paid a fee per client, per employee, per pay run. That is about as close to a toll booth as a services business gets, and it is why this company has earned operating margins above 40% for years.

But this report has to open with the fact that made us want to write it, and it comes from Paychex's own annual report. Companies are required to publish a five-year chart of what $100 invested in their shares would be worth today, alongside the market and their chosen peers. Here is Paychex's, for the five years to 31 May 2026:

Paychex lost about 36% of its value in a single fiscal year — a year in which it grew revenue 17%, grew adjusted earnings 11%, and announced no bad news whatsoever. Five years of holding it produced a total return of roughly 2.2% a year against the S&P 500's 14%.

★ And now look at the third row, because it is the whole story. The peer group — ADP, Intuit, Fiserv, Broadridge, Jack Henry, Equifax, Moody's, Verisk and others like them — fell 38% in the same year. Worse than Paychex. This was not a Paychex problem. The entire category of "boring compounder with recurring revenue and a compliance moat" was repriced downward at once, and the obvious candidate explanation is that the market has decided artificial intelligence will hollow out exactly this kind of rules-based, headcount-linked information processing.

The market has decided this business is structurally impaired. Paychex's own numbers do not yet show impairment. Either the market is early, or the market is wrong — and that tension is the entire investment case, in either direction.

The complication is timing. Those shares fell to $85.45 at their worst. They are now $121.11 — a rally of roughly 40% — and at that price they sit above the highest price target published by any analyst covering the company. The moment to act on "the market is wrong" was several months ago.

Here is the mechanic that makes Paychex genuinely interesting to anyone who has read Buffett, and it is not widely understood.

When Paychex runs a client's payroll, it debits that client's bank account for the entire cost at once — the net pay, the taxes withheld from employees, and the employer's own payroll taxes. It then pays that money out on the statutory calendar: wages to employees on payday, and taxes to the IRS and to fifty states and thousands of localities on their remittance dates, which for many small employers are monthly or quarterly.

In between, Paychex has the money. It invests it. It keeps every cent of the interest. In FY2026 the average balance held was $5.77 billion and the interest earned was $210.9 million.

This is float in the precise Buffett sense — money the company holds but does not own, invested for its own account. It is the mechanism on which the whole of Berkshire Hathaway was built.

Paychex is paid to hold other people's money. An insurer, on average and across the cycle, pays for the privilege. That asymmetry is real and it is the strongest Buffett hook available on this company.

But the last row is the honest counterweight, and it matters. Paychex has the better float on cost and certainty and the far worse float on scale and on what it is permitted to do. It cannot buy a business with client payroll taxes. Float quality without float optionality.

This is the question we set out to answer, and the answer corrected our own assumption. We expected the float to be the engine. It is not.

★ Two things about this float that almost nobody writes about, and both cut against the Buffett parallel.

First, it exists only because the government says so. The gap between collection and remittance is a feature of the US tax calendar, not of physics. Paychex's own risk factors name this explicitly: a change in regulations "either decreasing the amount of taxes to be withheld or allowing less time to remit taxes to applicable tax or regulatory agencies could adversely impact our interest income." Real-time payroll tax remittance is technically feasible, and faster payment rails push in the same direction over the long run. Berkshire's float cannot be legislated away. Paychex's can. No such change is currently proposed — but the two floats are not the same quality of asset, and the honest version says so.

Second, and more amusingly: a tax cut shrinks the float. The balance Paychex holds is a function of how much tax is withheld. This company is structurally long higher taxes.

★ And here is the fact that ends the section. In FY2026 Paychex earned $210.9 million of interest on its clients' money — and paid $269.5 million of interest on its own debt. The float company became a net borrower. That is what buying Paycor did, and it is the subject of Part IV.

The case for the moat is genuinely good and it rests on three things.

Switching costs are real and they are seasonal. Changing payroll provider mid-year means reconciling year-to-date wages, taxes withheld and benefit deductions across two systems, and then producing one clean set of tax forms at year end. The failure mode is not "the software is annoying" — it is "my employees were paid late" or "my tax filing was wrong." That is a very high bar, and it means unhappy clients typically wait months for the year boundary and then still hesitate.

Regulatory complexity is a fixed cost that only scale can carry. Payroll compliance spans federal rules, fifty states and thousands of localities, each with its own rates, forms, thresholds and calendars, all of which change. Building and maintaining that engine costs the same whether you have 800,000 clients or 200. That is a classic Buffett shape.

And the customer base is atomised to the point of powerlessness. Paychex's own annual report states it plainly: "No single customer is material in respect to total accounts receivable, service revenue, or results of operations." There is no annual renegotiation with a powerful buyer, because there is no powerful buyer. Pricing power over 800,000 small businesses that individually cannot negotiate is structurally strong — and Paychex named "price realization" as a growth driver again this year.

Payroll client retention was in the range of 82% to 83% — the same in FY2026 as in FY2025. Paychex loses roughly one client in six, every year.

To stand still, this company must sell about 135,000 new clients a year. That is not a subscription business with negligible churn; it is a business running an enormous sales machine to refill an enormous hole. Much of that churn is not a service failure at all — small businesses have a mortality rate, and Paychex serves small businesses. But the consequence is the same: Paychex's fortunes are levered to small-business formation, because that is its funnel.

Note how the company phrases it: retention of 82–83%, "and we have sustained high revenue retention." That second clause is the tell. Revenue retention is materially better than client retention, because the clients that disappear are the smallest ones and the survivors expand. Both facts are true. The bulls quote the second and the bears quote the first. We think you should hold both.

★ And the client count is flat. Roughly 800,000 payroll clients last year; roughly 800,000 this year. All of the growth came from price, from mix, and from buying Paycor. A moat that produces pricing power but no unit growth is a real moat — around a business that is no longer getting bigger the way it once did.

Here is the uncomfortable way to describe this transaction, and we think it is the accurate one.

Paychex spent $4.1 billion of borrowed money to move out of the market where it has a moat and into the market where it does not. It is a moat-dilutive acquisition.

The strategic logic is perfectly defensible: the small-business core is not growing in client numbers, so go where the growth is. But be clear about what was bought. The old Paychex was a narrow-moat business with real pricing power over 800,000 captive small employers and no debt. The new Paychex is that, plus a subscale mid-market challenger fighting Paycom and Paylocity on their own ground, financed with $4.6 billion of debt at over 5%.

★ The return on that $4.1 billion is not yet visible, and it is the question that matters more than any other for the next five years. No impairment has been taken. That is management's assertion that the deal is working — it is not independent evidence that it is.

One thing that deserves credit, though. Look at how the acquisition was paid for in practice. Buybacks were cut from $169m to $104.5m in the deal year to protect the balance sheet, then run at $611m the following year once it was digested — at an average price of $108.81, well below today's. The dividend was never touched. That is a revealed preference, and for an income-focused holder it is the right one: management treats the dividend as inviolable and the buyback as the flexible variable.

Payroll and HR administration is exactly the sort of rules-based, document-heavy, headcount-linked work that artificial intelligence is pointed at. The market clearly believes this: it took 36% off Paychex and 38% off its peer group in a single year. So let us apply the same test we used on weight-loss drugs and snacks at PepsiCo: is there MEASURABLE evidence, or is this narrative?

The verdict as of today: not yet measurable. Anyone claiming AI is already eroding this business is asserting something the disclosed data does not support.

But we are not going to let this piece become a rebuttal, because there are three serious counter-arguments.

If the seat count is the channel that matters, it can be measured — not from Paychex, but from the government. What we found is more interesting than the AI debate.

The threat to Paychex was never that a machine would do the payroll. It is that there are fewer people on the payroll to do — and that the pipeline of new employers has shrunk by a fifth while nobody was looking.

⚠️ One caveat about the index Paychex publishes, since we have used it. Paychex produces the most-quoted monthly gauge of American small-business employment — and it is measuring its own end market, using its own surviving clients. Businesses that fail or leave drop out of the panel, which biases a same-store measure upward. In 2025 the Paychex index never printed a figure that would have shown a reader what the government's data captured: that small firms were destroying jobs. It is a genuinely useful high-frequency read. It is not an independent one.

Worth noting too what small employers themselves say. In the National Federation of Independent Business survey, hiring plans jumped in July to their highest since 2022 — while actual employment change has been negative in 22 of the last 24 months, and hit its weakest reading of 2026 in the same survey. Small employers have spent two years saying they intend to hire while quietly shedding staff.

Payroll's difficulty was never the arithmetic. Calculating gross-to-net pay has been solved software since the 1970s. What is actually hard is three things:

Somebody must be liable. A party of record has to sign and file, and be answerable to the IRS and fifty states when it is wrong. Penalties for late or misfiled payroll taxes are automatic and land on the business owner personally. Somebody must move the money — billions of dollars at settlement-grade reliability, with bank relationships and fraud controls. And the tolerance for error is zero. A model that is 99% accurate at filing payroll taxes is an unusable product. Nobody wants an AI that probably filed the taxes.

An AI agent can draft the handbook and answer the HR question today. It cannot assume the legal liability or hold the settlement account. So the right framing is not "Paychex gets replaced." It is this: the advisory and administrative labour that Paychex sells around the compliance core gets commoditised, and Paychex is left with a thinner, more utility-like business. That is a margin and pricing story rather than an extinction story — and it is a far better description of the risk than either the bull or the bear cartoon.

For what it is worth, Paychex is pointing the technology at its own cost base as much as at clients. It launched an engine it calls WISE this year, deployed — in the chief executive's words — "across our HCM platforms and internal operations, enabling more proactive, autonomous execution." That second half is the financially interesting one: this is a labour-heavy service business, and if AI works on its own service desk, margins expand. ⚠️ No revenue figure for WISE has been disclosed. Treat it as a strategic claim, not a result.

Our house rule is that any dividend-paying company gets its payout examined properly — coverage on free cash flow rather than earnings, the trend over time, and what would have to happen for a cut. Paychex yields 3.9% on its current quarterly rate of $1.19, so this section matters.

First, a correction, because this one is widely misstated. ⚠️ Paychex is not a Dividend Aristocrat and does not have an unbroken multi-decade increase record. It has paid a dividend since the late 1980s and raised it in most years — but it held the payout flat through the financial-crisis period, which breaks the consecutive-increase test. If you have seen this company described as a decades-long unbroken raiser, that description is wrong.

The two payout measures diverge, and the divergence is the story. On reported earnings the payout has climbed relentlessly — 72%, 75%, 78%, 87%, and now roughly 90%, which the company states in its own filing. On free cash flow it improved last year to 68.5%, with coverage back to 1.46 times.

Why the gap? Because Paycor's non-cash amortisation — $242 million a year — crushes reported earnings without touching cash. On adjusted earnings of $5.51 against dividends of $4.43, the payout is about 80%, which is probably the fairest single number.

The honest verdict: the dividend is covered, and the cushion is thinner than it was. FY2025 was the pinch — coverage of just 1.18 times, in the year the Paycor debt landed. It has recovered, but a ~90% reported payout alongside $4.6 billion of debt and a permanent $270 million interest bill does not leave much room.

★ And the growth rate has decelerated through the acquisition: +17.5%, then +11.9%, then +10.1%, then +9.7%. Not dramatically, and never into a cut — but the double-digit-teens dividend growth of the pre-Paycor era is gone. With FY2027 adjusted earnings guided up 7–9%, expect dividend growth of roughly 6–9%, not 12–17%.

★ A note on that 45% return on equity, because it is the most quoted statistic about this company and it deserves decomposing. It comes from four sources and they are not equally admirable. Genuine asset-light economics — capital expenditure is under 4% of revenue and clients pay at or before delivery. That part is real and excellent. The float — $5.8 billion earning a return with no equity behind it. That is legitimate, and it is leverage. A ~90% payout ratio, which keeps the equity base small and therefore the ratio high by arithmetic rather than by merit. And, since April 2025, actual debt. A leveraged 45% is a lower-quality 45% than the unleveraged version Paychex used to earn.

★ And a technical warning that catches almost every screen. Paychex's balance sheet carries $4,832 million of "funds held for clients" as an asset and $4,885 million of "client fund obligations" as a liability. That is the same money, shown twice. It is not Paychex's asset and not Paychex's debt. Screens that sweep the client funds into "cash" understate enterprise value by nearly $5 billion; screens that treat the obligations as debt overstate it by the same amount; and any return-on-assets figure for this company is meaningless unless both are stripped out. If you look Paychex up on a stock screener, assume the enterprise value and the asset ratios are wrong until you check.