Returns on Capital

Returns on Capital

Accenture's margins have proved the durable part of its economics. GAAP operating margin has held inside a ~14.7–15.2% band for six years while revenue growth swung from +26% to +1%, and the adjusted line has ticked up about 10 basis points a year to 15.6%. The returns on the capital behind those margins have not held: return on equity has fallen from 31% to 25% and return on capital employed from 32% to 24% since FY2022, because the balance sheet — chiefly $22.5B of never-impaired acquisition goodwill — has grown far faster than profit.

Margins are the resilient part of the story

Through a demand cycle that took local-currency growth from a boom to roughly flat and back, the profit ratio barely moved. GAAP operating margin was 14.7% in FY2025, 15.2% in FY2022, and 14.7% in FY2020 — a remarkably narrow range for a business whose top line more than doubled over the span [1]. Gross margin (revenue less cost of services) has sat near 32% throughout, easing to 31.9% in FY2025 as cost of services rose to 68.1% of revenue, offset lower in the stack by tighter sales, marketing and administrative spending [2].

Source: FY2025 10-K income statement and MD&A [3] [4]; FY2022 10-K [5]; FY2021 statutory accounts [6].

That stability matters directly to the report's central question. If generative AI were already compressing the price of Accenture's delivery hours faster than the company could reprice or take out cost, the first place it would show is the gross and operating line. Six years of flat-to-rising margins are the strongest single piece of evidence that AI-in-delivery has not yet reached the income statement — the same conclusion the recent-results detail reached from the revenue side (Anatomy of the Slowdown).

The expansion lives in the adjusted line

The "+10 basis points a year" that management highlights is an adjusted-margin figure, and the adjustment has become a standing feature rather than a one-off [7]. On a GAAP basis operating margin actually fell 10 basis points in FY2025, to 14.7% from 14.8%, because "business optimization costs" reduced it by 90 basis points; the adjusted 15.6% adds those costs back [8].

These charges did not exist before FY2023. There were none in FY2020, FY2021 or FY2022; then $1,063M appeared in FY2023, $438M in FY2024, and $615M in FY2025 — with a further ~$250M guided into the first quarter of FY2026, for roughly $865M across a six-month restructuring [9] [10].

Source: FY2025 10-K Consolidated Income Statements, business optimization costs line (FY2023–FY2025) [11]; FY2020–FY2022 income statements carried no such line [12].

The character of the FY2025 charge is the notable part. It funds a "three-pronged talent strategy" — upskilling, exiting people "in a compressed timeline where reskilling is not a viable path," and driving efficiency "including through AI" [13]. That is the cost of continuously rebalancing a 779,000-person workforce against shifting skill demand. As long as AI keeps reshaping which skills sell, the case that these are recurring costs of doing business — and that the adjusted margin flatters the through-cycle number by roughly 90 basis points — is a reasonable one for a skeptic to hold.

Returns on capital have eroded

Margins held; returns on the capital employed did not. Return on equity has fallen every year since FY2022, from 31.1% to 24.6%; return on capital employed from 31.5% to 24.1%; return on assets from 14.6% to 11.7% [14]. The 24.6% ROE that reads as elite in isolation — and that the report's opening chapter flagged as undiminished operating quality (Reinvention at Scale) — is in fact the low point of a steady four-year decline.

Source: derived from reported shareholders' equity and operating income, FY2022–FY2025 Form 10-Ks [15] [16].

The cause is the denominator, not the numerator. Between FY2022 and FY2025 revenue grew 13% and net income attributable to Accenture grew 12% (about 19% excluding the business-optimization charges), while shareholders' equity grew 41% and goodwill grew 72% [17] [18]. Capital grew several times faster than the profit it produced, and every return ratio fell as a direct result.

Goodwill is now most of the equity

The engine behind the growing capital base is the programmatic acquisition machine detailed earlier (Capital Allocation). Goodwill has climbed from $13.1B in FY2022 to $22.5B in FY2025 — never once impaired — and now equals about 72% of Accenture's shareholders' equity and 34% of total assets [19] [20].

Source: FY2025 10-K — MD&A operating margin [21], balance sheet shareholders' equity and goodwill note [22] [23].

Source: FY2025 10-K balance sheet and goodwill rollforward [24]; FY2022 10-K balance sheet [25].

As an investment, the acquisition program has earned a sub-average incremental return so far. The roughly $9.1B of book equity added since FY2022 sits against roughly $0.8B of additional GAAP earnings — near $1.3B adjusted — an incremental book return in the high single digits to low teens, well beneath the ~25% the existing business earns. The share count did the visible work of protecting per-share metrics: about $13.5B of buybacks over three years held basic shares down 1.2% even as equity swelled, which flatters ROE by shrinking its denominator. Absent the repurchases, the return ratios would read lower still. The acquisitions that pushed goodwill to ~72% of equity and cut return on equity from 31% to 25% are rewarded in the CEO's pay plan exactly like organic growth, because the plan's dominant metric is operating income with no measure of return on capital (Pay and Alignment).

The impairment record is clean, but not costless

Management reports that no goodwill impairment existed at August 31, 2025, with each segment's fair value "substantially" exceeding its carrying value [26]. That unblemished record deserves a caveat. In the FY2025 restructuring, Accenture wrote off about $271M in asset impairments "primarily related to the divestiture of two acquisitions in the Americas that are no longer aligned with our strategic priorities" — capital destroyed on deals that did not work, routed through the business-optimization line rather than the goodwill-impairment line [27]. The goodwill balance can stay pristine while individual acquisitions are unwound at a loss.

The forward pressure on returns is visible in what management is buying. It is deliberately skewing acquisitions toward higher-growth, non-FTE (software and platform) models and paying up to do it: on the OT-security platform it built from three assets, an analyst flagged "initial dilution" [28], and management itself acknowledged "higher valuations even for the services acquisitions" in AI-adjacent areas [29]. Higher entry multiples on a step-up in deal spend point to more goodwill per dollar of near-term profit, which extends the returns-erosion trend before it reverses.

What would change this read

The erosion is from an exceptional base, and two forces could arrest it. First, the non-FTE shift is meant to raise structural margins over time: recurring software and platform revenue can carry higher margins than time-and-materials work once the flagged integration dilution passes, and management frames the reskilling spend as an investment that lifts revenue per person rather than a permanent drag (Advanced AI Economics). Second, roughly $3.3B of annual investment in acquisitions, research and learning is deliberately front-loaded; if the recent deal wave (Ookla, the OT-security platform) compounds organic growth as intended, incremental returns rise and the ratios stabilize [30].

The read that fits today's evidence: the margin resilience is real and is the bull's strongest fact against AI deflation, but the steady decline in returns on capital is the cost of buying growth that the business is no longer generating organically — a decline that continues as long as goodwill grows faster than profit. The measurable turn would be incremental returns on the recent acquisition vintages beginning to match the legacy business; until reported ROE and ROCE stop falling, the capital base is compounding faster than the earnings it supports.