Accenture plc

Accenture sells expertise by the hour, the project and the year. A bank that wants to move its systems to the cloud, a carmaker redesigning its supply chain, a government department building a benefits website or a retailer running its customer-service centre hires Accenture to plan the work, write and connect the software, and often to run it afterwards. It employs about 814,000 people, most of them in India, the Philippines and the United States, and works for more than 9,000 clients, including three-quarters of the Fortune Global 500 and every G7 government. A ten-year-old's version: big companies hire Accenture's people to make their computers do new things, and then to keep them running.

Its revenue comes in two roughly equal halves. Consulting ($36.9bn in the year to August 2026) is project work — strategy, system design, building and installing software — which clients can postpone when budgets are tight. Managed services ($37.3bn) is multi-year contracts to run applications, infrastructure, security or business processes, which renew and grow steadily. In fiscal 2026 managed services overtook consulting for the first time, and grew twice as fast.

The economics are unusual. Accenture owns almost no physical assets: capital spending was $743m on $74.2bn of revenue — 1%. Its costs are people, its inventory is their time, and its margin (an adjusted 15.8% in fiscal 2026) depends on how many of them are working on paid projects (93%) and what clients will pay per hour or per outcome. That is why one sentence in its annual report, filed on 16 October, matters more than any other: "we experienced lower pricing in many areas of the business during the fourth quarter." At $236.93 the shares are 90% above their June low and 43% below their 2021 record.

Two lessons for an owner. The first is that Accenture's history is one of reinvention by necessity: mainframes, then enterprise software, then outsourcing, offshore delivery, the internet, cloud, and now AI. Each wave threatened the old work and created more new work, and each time Accenture bought and trained its way into the new skills faster than its rivals. That is a real competence — and, in Buffett's terms, a warning, because a business that must reinvent itself every few years is never safe. The second is that the price of the shares has done most of the damage to owners recently, not the business: earnings per share rose from $9.16 in fiscal 2021 to $13.56 in fiscal 2026, while the multiple fell from about 30 to 17. The five-year total return is −26%.

The pyramid. A consulting firm's profit comes from leverage: a small number of senior people win and supervise the work, and a large number of junior people — many in India — do it. Revenue is roughly the number of people billed, times the hours they work on paid projects, times the price per hour. Accenture's fiscal 2026 numbers are as good as this model gets: 93% utilisation (92% the year before), attrition down to 13%, and revenue per person, adjusted for currency, up 5% — the fastest growth in at least two years.

Why AI is both threat and opportunity. AI tools that write code, test it, document it and answer customer queries attack the bottom of the pyramid — the junior hours that make up much of the revenue in application maintenance, testing and business-process work. If a project that took 100 people now takes 60, and clients pay by the hour, revenue falls 40%. But AI also creates work that did not exist: preparing clients' data, rebuilding old systems so AI can use them, security, governance, and training. Accenture's answer is to sell more by outcome — a price for the result rather than the hours — and to keep the productivity gain itself. Whether clients let it is the business question of the decade.

The fourth quarter showed both forces. Revenue rose 7% in local currency, ahead of guidance, and large deals hit a record. But the 10-K says pricing was "lower in many areas of the business" in the quarter, after a year in which it was "relatively stable", and the company's risk factors now state that automation including AI "has led and may continue to lead to reduced demand for our services". Revenue per person rising while prices fall means people are doing more work each — consistent with AI raising productivity and Accenture, so far, keeping much of the benefit.

Buffett gave two tests in his 2007 letter that pull in opposite directions here. The first is the Mayo Clinic: "if a business requires a superstar to produce great results, the business itself cannot be deemed great." Accenture passes — it is an institution whose relationships and methods outlast any partner or CEO, and clients stay for decades. The second is the warning that "a moat that must be continuously rebuilt will eventually be no moat at all." Accenture fails that one in principle: its value lies in knowing the newest technology, so it must spend billions every year on training and acquisitions just to stand still, and each wave — cloud, now AI — resets part of the game.

In practice the moat has held for twenty-five years, because the cost of rebuilding it is lower for Accenture than for anyone else: it has the clients, so it learns where the work is first, and it has the cash to buy the skills. We score it 6 — wide relationships and scale, contested pricing, and a permanent reinvestment bill.

The fear. In February 2026 a run of AI launches — agent plug-ins for office work, then a claim that AI could modernise old COBOL mainframe systems — set off a sell-off across software and IT services. The logic was simple: if machines can write, test and migrate code, the world needs fewer people to do it, and companies that sell people by the hour will shrink. Accenture fell 9.6% on 3 February and 6.6% on 23 February. When its third-quarter bookings fell 3% in June and it trimmed its full-year forecast, the fear seemed confirmed: the shares fell 18% in a day and reached $124.44 by 30 June, nine times earnings.

What the numbers showed. By October the evidence pointed the other way, at least for now. Revenue grew 5% in local currency for the year and 7% in the fourth quarter. Revenue per person rose 5%, accelerating every quarter from zero in late 2024, while headcount rose by 35,000. Bookings were 1.1 times revenue, with a record number of very large contracts, and managed services — the multi-year work most exposed to automation in theory — grew fastest. The pattern so far is that AI is making each Accenture employee more productive and creating new work, rather than removing work.

Our reading: AI is a price cut on Accenture's work, not yet a volume cut. The June price assumed volume would collapse. It has not; Accenture is doing more work with more people. What is happening instead is slower and more subtle: clients expect part of the productivity gain, so the price per unit of work falls, and Accenture must sell more units — more projects, more AI-related work, more outcome-based contracts — to grow. So far it has. The risk is that the price cuts accelerate faster than volume grows, as happened to the Indian outsourcers' margins in earlier waves. The decision to stop reporting AI bookings in the headline metrics is the one thing we dislike: a company confident in its AI business would show it. We judge the June panic overdone and today's price a fair reflection of a business growing mid-single digits with a moat under pressure.

The pattern. Accenture sits in the middle of the market: more expensive than the Indian firms, cheaper and more hands-on than the strategy houses, and broader than anyone. AI squeezes the bottom of that market hardest — the routine maintenance and testing work where Indian rivals are strongest — and creates work at the top, where clients need help deciding what to do and connecting models to messy old systems. That is why Accenture has outgrown its Indian peers through the AI scare. The newer threat is the model companies themselves: if OpenAI and Anthropic deploy their own engineers into clients, they take the most valuable part of the work. For now both are Accenture's partners — Anthropic is paying Accenture to help evaluate its models — and the competition is more theoretical than visible in the numbers. The comparison on our board is ServiceNow, a software platform facing the same AI question from the other side.

Integrity: no veto, two flags. Accenture's reporting is clear and conservative, and it has not hidden bad news — it cut guidance in June rather than stretching. The flags are the open Justice Department investigation of its federal unit (Part XII) and the decision to drop AI bookings from its headline metrics. Capital allocation: good in aggregate, poor on price. For a decade Accenture has returned roughly 70–100% of its free cash flow, which is right for a business that cannot reinvest much internally — Buffett noted in 2007 that great asset-light businesses "can't for any extended period reinvest a large portion of their earnings internally at high rates of return". But it bought shares at 25–35× earnings for five years, and only became price-sensitive this summer. His 2011 test — "what is smart at one price is dumb at another" — cuts both ways in this record. Owner mentality: weak. Executives are paid mostly in stock, but none bought any with their own money when the company itself said the shares were cheap. We score management and capital 6.

A steady compounder, slowing. Over ten years Accenture grew revenue 7.9% a year and adjusted earnings per share roughly 9–10% a year, helped by a shrinking share count and rising margins. Fiscal 2026 fits the pattern: 5% organic growth, 1.5 points from acquisitions, a margin up 10 basis points, adjusted EPS up 8%. The plan for fiscal 2027–28 is similar but slightly slower — mid-single-digit growth, with acquisitions adding 2–2.5 points — and fiscal 2027's guidance implies adjusted EPS growth of only 3–6%, because higher interest costs and a higher tax rate offset the operating gains.

The cash is better than the earnings. Accenture turned $1.40 of every $1 of net income into free cash flow in fiscal 2026, helped by working capital. Management guides 1.2–1.3× for fiscal 2027 and "at least 1.1×" afterwards. Share-based pay of $2.1bn a year is not a cash cost but is a real one for owners, which is why we deduct it in owner earnings (Part XI).

The verdict on the dividend: safe and growing slowly. Accenture's dividend is among the safest on our board: a payout of a third of free cash flow, net cash, and a business that has never cut it. The change in October is not a warning about the cash; it is a decision about where the cash goes. With the payout already near half of earnings, management has chosen to grow the dividend at 5% and put more into acquisitions — about $8bn in fiscal 2027 — on the view that buying AI, data and security skills earns more than paying it out. Income investors should expect the dividend to grow a little slower than earnings for the next three years. At 2.9% the yield is respectable, not compelling; at the June low it was 5.5%.

What is it worth? We discounted owner earnings of about $13.50 a share at 10% for ten years. If they grow 6% a year for five years and 5% for the next five — roughly the Investor Day plan plus a shrinking share count — and the business is then valued at 15 times, Accenture is worth about $245, almost exactly today's price. If AI deflation wins — earnings flat for five years, then falling 2% a year, ending at 11 times — about $125, which is where the stock traded in June. If AI turns out to be another cloud wave and earnings grow 9% and then 7%, ending at 19 times, about $350.

At $236.93 the price is our central value. The range is wide because the question in Part VI is genuinely open, and the June price — the pessimistic case — was the opportunity: anyone who bought then with the company has nearly doubled their money in four months. Our playbook does not demand a bargain for an exceptional business at a fair price, but Accenture scores 6 on its moat, not 8 or more: a moat that must be rebuilt every year, facing a technology that could cut the price of its work. For that we want a margin of safety. Our advice: hold if you own it; buy below about $195 — 20% under our central value, about 13 times fiscal 2027 earnings and a 3.5% yield. The stock closed below that level for most of the period from late February to late September 2026; the AI debate is likely to offer it again.