Chapter 3

AI and the Labor Model

Figures converted from Canadian dollars at historical FX rates — see data/company.json.fx_rates for the rate table. Ratios, margins, multiples, indexed series and growth rates are unitless and unchanged.

CGI sells roughly 94,000 people's time, so the question generative AI poses is direct: does it hollow out the revenue or lift the margin? The evidence points to margin. A managed-services majority priced on outcomes — where CGI has kept its productivity gains for decades — turns AI into a cost lever rather than a revenue hole, and headcount has stayed flat while revenue grew. The catch is that AI does nothing for the organic-growth stall, and the one lever that could turn it into growth — CGI's IP-revenue ambition — is moving backwards.

What AI puts at risk

CGI is a labor business. It employs approximately 94,000 consultants and professionals [1], and its revenue is, at root, billed expertise. The company says so itself in its risk disclosures: developments in AI — "including agentic AI, generative AI, as well as automation and machine learning" — could pressure "our revenue, net earnings and resulting cash flow from operations" if it fails to adapt [2]. AI also sits on the standing list of competitive risks in every recent filing [3].

The exposure is not evenly spread. About 45% of revenue is business and strategic IT consulting and systems integration (SI&C), and 55% is managed IT and business process services [4]. The SI&C half is the part a skeptic worries about first: consulting is largely time-and-material, and systems integration is the kind of implementation work AI is meant to automate. Management concedes the market shares that fear — one analyst noted that "the market apparently thinks there's going to be an impact from AI… automating a lot of the implementation processes that you're exposed to" [5].

The industry backdrop makes the concern concrete. The offshore-heavy players built on billable headcount are already shrinking: Infosys cut its workforce by about 8,000 sequentially in its most recent quarter [6], and Cognizant carries roughly 336,800 employees, the overwhelming majority in India [7]. For a pure labor-arbitrage model, fewer billable hours is fewer dollars. The question for CGI is whether its own model works the same way.

The buffer: outcome-based pricing

It mostly does not, and the reason is how CGI prices. Its default managed-services model is outcome-based — "we commit to cost predictability and delivering results, not just inputs" — and those engagements have, "for decades, included commitments to deliver ongoing productivity improvements" [8]. CGI has been here before: it absorbed the shift from onshore to offshore, then to cloud, each time promising the client savings and keeping a share of the efficiency as margin. Advanced AI, in management's framing, is the next lever in that same mechanism — one that "has contributed to improving profit and reinvesting in capability building" [9].

When work is priced on outcomes, a productivity gain does not shrink the invoice; it widens the spread. CFO François Boulanger put the arithmetic plainly: with AI in delivery, "I don't need necessarily the same head count number or same number of people to deliver the services," so revenue per employee should keep rising; and on fixed-price work, "every way of reducing the cost would go directly in our margin improvement" [10]. Even the SI&C side is more insulated than the "time-and-material" label suggests: management estimates 40% to 50% of the SI&C book is fixed price, where AI tools "increase the profit or the margin on our projects while hitting the right price tag" [11].

The early numbers fit the story. Over FY2023 to FY2025, revenue rose about 11% while headcount barely moved — and the FY2025 increase to 94,000 was itself acquisition-driven, so organic headcount is flatter still.

Source: revenue from reported financials FY2023–FY2025; headcount from annual reports (approximately 91,500, 90,250 and 94,000 consultants and professionals) [12]. Indexed to FY2023 = 100; ratios are unitless.

That gap is revenue per employee, and it is widening — up roughly 4% a year, to about $124,000 in FY2025. Management pegged the FY2025 lift near 5% and credited AI alongside offshore delivery [13]. The honest qualifier: that rise blends three forces, not one — AI in delivery, but also the acquisitions and a growing India-and-Poland offshore mix that Boulanger named in the same breath [14]. AI is a contributor to the decoupling of revenue from bodies, not yet a proven driver of it.

CGI is also putting operational substance behind the claim rather than only slideware. Its DigiOps suite runs in production for many clients with over 165 AI agents and 2,000 automation workflows, delivering up to 30% productivity gains and up to 40% faster resolution of IT requests; in software development, AI code generation — about a quarter of the development life cycle — is showing efficiency gains near 30% [15]. And CGI is buying the capability as well as building it: its FY2025 acquisition of Daugherty is described as a firm "specializing in artificial intelligence, data analytics, strategic IT consulting" [16].

The catch: deflation, and a stalling IP bet

The same mechanism that defends margin has a second edge. If AI lets CGI deliver a given outcome with fewer people, then on outcome and fixed-price contracts the value delivered — and eventually the contract's price at renewal — can compress even as margins hold. Boulanger himself expects the input-based billing model to "continue to reduce" in favour of fixed and outcome-based work [17]. That is good for margin and neutral-to-negative for the top line — which is precisely the organic-growth problem the report opened on (Build and Buy). AI helps CGI earn more on flat volume; it does not obviously help CGI sell more.

The intended answer to that is IP. CGI's stated "IP30" ambition is to earn 30% of revenue from its own intellectual property — software and platforms sold repeatedly, where AI would expand the model rather than deflate it, and where each sale is not a fresh block of billed hours [18]. It is the part of the strategy where the AI story turns from defense to offense. The problem is that the number is going the wrong way.

Source: quarterly earnings-call remarks, IP as a percentage of total revenue [19][20][21]; target from May 2026 Corporate Overview [22].

IP was 22.6% of revenue in late FY2023 and 22.9% a year later; through FY2025 it fell to 21.6%, then 21.5%, then 20.6% [23][24][25][26][27]. Nearly ten points below the target, and drifting away from it. Management's own explanation ties the two halves of the strategy together: the decline is due to "the dilutive impact of recent business acquisitions" — the labor-and-services-heavy consultancies like BJSS, Daugherty and Apside carry little IP, so buying them mechanically shrinks the IP share [28]. The Buy engine that sustains reported growth (Acquisition Math) is, on this axis, pulling CGI away from the IP-led model it says AI will reward.

What would change the read

Two markers would move this from defensive to genuinely offensive. First, the IP share breaking back above the low-20s and trending toward 30% — evidence that CGI can build and sell AI-era software faster than its acquisitions dilute the mix. Second, a sustained pickup in organic growth (stripping out both currency and acquired revenue) alongside flat or falling headcount — the signature of AI expanding the business rather than merely making the existing book cheaper to run. Absent those, the fair read is that AI protects CGI's economics without solving its growth problem, and CGI has not yet published data proving implementation work resists automation as well as it argues it does [29].