Nearly three in five large organisations exceeded their artificial intelligence (AI) budgets in the past 12 months – and for Canadian insurance brokers, the consequences are already arriving in two places: through carrier partners facing margin pressure from untracked deployment costs, and through underwriters asking harder questions about how commercial clients are governing their own AI tools. The data comes from a November 2026 research snapshot by the Capgemini Research Institute, based on a September 2026 survey of 500 C-suite executives from organisations with annual revenues above US$1 billion, across 10 countries and 13 industries, including insurance.
Among organisations affected, costs ran approximately 24% above expectations on average. One in five overshot by more than 30%, and two-thirds reported negative financial consequences – including lower free cash flow or reductions in discretionary spending. Only 56% of executives said they were fully confident in forecasting how AI costs would evolve as adoption scales.
Concern is rising: 53% of executives said they were worried about the overall financial cost of AI at the time of the survey, up from 45% a year earlier.
The Capgemini report identifies a consistent failure to track the full cost of AI deployment. Only a third of executives – and just one in four chief financial officers (CFOs) – said they were fully confident estimating indirect costs, including human oversight, rework, and governance. Six in 10 organisations said they lacked a complete enterprise-wide view of AI costs.
The least-tracked components were ongoing operations, retries, failures, and rework (measured by 48% of organisations) and human oversight and workflow change (tracked by 51%). Cloud and infrastructure costs were measured by 81% – a gap that shows how traditional IT cost disciplines have not yet extended to the full AI delivery chain.
Half of respondents said AI spending was approved without adequate visibility into the full cost.
Volker Darius, chief executive of Capgemini Invent, said the problem goes beyond procurement: “As AI moves deeper into business workflows, organisations need to stop treating its cost as a technology or procurement issue and start managing AI as a portfolio of economic outcomes. Cost discipline alone will not create a profitable AI estate, just as value creation without cost control will not be sustainable.”
That gap carries regulatory weight in Canada. The Office of the Superintendent of Financial Institutions (OSFI) published a bulletin on generative and agentic AI in July 2026, specifically requiring federally regulated insurers to maintain governance over AI systems that can act with limited human oversight and to manage third-party AI dependencies – areas that the Capgemini data shows most large organisations are not yet measuring. OSFI’s Guideline E-23 on model risk management takes effect May 1, 2027, raising the compliance bar for Canadian carriers further. For brokers, this regulatory timeline means the gap between what carriers are deploying and what they can demonstrate to OSFI is narrowing – and the AI governance guidance OSFI is imposing on federally regulated Canadian insurers affects the carrier partners brokers depend on.
Read more: OSFI maps out AI risk controls for federally regulated insurers
Agentic AI – systems that complete multi-step tasks autonomously rather than responding to a single prompt – is a growing source of cost pressure across insurance underwriting, claims, and operations. According to data from OpenRouter, an AI model gateway service, cited in the Capgemini report, agentic token usage grew 14 times in a six-month period to August 2026, compared to a 2.8 times increase for human-initiated usage.
A research paper published in April 2026 by researchers at Stanford University, the University of Michigan, and others – How Do AI Agents Spend Your Money? Analyzing and Predicting Token Consumption in Agentic Coding Tasks – found that agentic coding tasks use approximately 1,000 times more tokens than code chat, with up to 30 times cost variance on the same task depending on agent design.
Bernd Meurer, chief technology officer (CTO) and field CTO Europe at BT Group, was quoted in the Capgemini report: “Token costs have fallen drastically. But with more usage, and depending on how you design the agent, costs can rise substantially.”
Melissa Carruthers, partner and insurance consulting leader at Deloitte Canada, told Insurance Business in January 2026 that most carriers are still approaching AI as a technology project rather than a business transformation: “The biggest risk is treating AI like another technology rollout. Insurers still stumble when they bolt AI onto old processes or ignore workforce implications.” Her firm’s September 2026 analysis of the domestic market, The Orchestration Gap, found that organisations committing to agentic AI without designing for token-based cost scaling tend to discover the financial consequences only after they are locked into an architecture that is expensive to change.
The Capgemini report found that only one in five organisations calculates “AI Yield” – defined in the report as the verified value of an AI-assisted outcome after subtracting its full safe-delivery cost, including model usage, human oversight, governance, and rework. Among organisations with established spend-tracking capabilities, only 45% of expenditure could be linked to verified business value, and 20% produced no business value at all.
The top benefits attributed to AI were productivity gains (cited by 59% of respondents), operating cost reduction (56%), faster processes (50%), and improved customer experience (42%). Only 32% cited revenue growth, including from new products or services.
Demonstrating sufficient value to secure funding was identified as the biggest obstacle by 57% of respondents. The broker-rated technology providers delivering measurable impact across Canadian underwriting, claims, and distribution in 2026 offer a useful point of comparison – the platforms earning broker confidence are those that link spend to measurable workflow improvements, not activity metrics.
Read next: Global 5-Star Technology and Software Providers
The Moody’s Ratings October 2026 survey of 58 North American insurers and brokers found that most carriers and brokers now have AI in production but cannot consistently track returns – the same measurement problem the Capgemini data describes at the enterprise level.
For brokers, this plays out on two fronts. On the carrier side, insurers running over AI budget – without mechanisms to link spend to verified outcomes – face operating margin pressure that can influence pricing timelines and the pace of improvements in broker-facing platforms.
On the client side, commercial clients deploying their own AI tools face the same hidden cost problem: approval without visibility, oversight costs that go unmeasured, and spending that cannot be traced to business value. Underwriters are already asking brokers about their commercial clients’ AI governance frameworks, training protocols, and contingency planning – questions that the AI governance gap documented across Canadian businesses by Gallagher’s risk experts shows most businesses cannot yet answer fully. Brokers who can help clients document and tighten those controls before renewal are adding tangible value at the placement stage.