A Deloitte Canada Insurance Point of View - 2026
The Orchestration Gap
Canadian insurers are investing heavily in AI, and many individual use cases are delivering value. The harder challenge is getting those investments to build on one another. We see the gap between AI investment and enterprise impact increasingly as an orchestration gap.
About this point of view
Over the past three years, Canadian insurers have moved quickly to test and deploy AI across the business. Many of those initiatives work and deliver real value within a function. What is less common is for the value to carry forward into the next use case, or for the enterprise to become meaningfully more capable as the portfolio grows.
That is why we describe this as an orchestration gap. The insurers beginning to pull ahead are not necessarily those that started first or spent the most. They are making clearer choices about what should be shared across use cases, what intelligence they need to control, and how people and AI agents will work together as adoption scales.
These choices are already being made, sometimes through enterprise strategy and sometimes one project at a time. The risk is that decisions optimized for an immediate use case create constraints that only become visible later. This first paper in the Compounding Intelligence series looks at the patterns we are seeing in the Canadian market and the economics behind them. The question is simple: as insurers move from individual AI solutions to agentic workloads at scale, what needs to be put in place now so each new capability builds on what already exists?
01 · The inflection point
Why the window is narrowing
Five pressures are converging across the Canadian market. None is new on its own, but together they are changing the operating environment for insurers at the same time AI investment is accelerating.
01
Evolving risk profiles across the market
The risk environment is shifting faster than traditional modelling cycles were designed to accommodate. In 2024, for the first time in Canadian history, insured damage from severe weather events surpassed $8 billion, nearly triple 2023 and 12 times the annual average from the previous decade, with a single five-week stretch producing approximately 250,000 claims.
The volatility is not confined to P&C. Health and benefits carriers are facing equally structural pressures: rising drug costs, complex chronic disease prevalence, and growing demand for mental health coverage. In life and longevity, demographic shifts and evolving mortality assumptions are compressing margins on products priced against very different long-term expectations.
$8.5B insured loss, 2024 · IBC / CatIQ
02
Shifting customer expectations
Since 2019, Canada has experienced a 115% increase in the number of claims for personal property damage and a 485% increase in the costs for repairing and replacing personal property. Canadians are engaging with their insurance more frequently and with higher expectations for responsiveness each time they do.
An aging population moving into retirement is generating growing demand for protection, long-term care, and financial security solutions. A next generation of insurance consumers is rethinking asset ownership, delaying traditional life events, and expecting coverage that adapts to how their lives actually change. Both groups are converging on the same expectation: an insurer that operates as a continuous partner rather than an intermittent administrator.
+485% personal property repair costs since 2019 · IBC
03
Financial performance pressure
Structural cost pressure across P&C, group benefits, and life is converging at exactly the point when AI infrastructure investment is most needed.
In P&C, combined ratios are manageable for well-run carriers, but the underlying cost structure tells a different story: expense ratios that include 15 to 20 percent in broker commissions leave limited room for operational efficiency gains without rethinking the distribution model entirely. In health and group benefits, drug cost inflation and rising utilization are outpacing premium growth.
Forrester projects that AI and automation will improve expense ratios at the top 50 insurers by two percentage points in 2026, but in our observation that improvement is flowing to carriers with the architecture to embed AI into core operations, not those running isolated point solutions at the margins.
+2pp expense ratio improvement projected, 2026 · Forrester
04
The regulatory environment
OSFI's E-23 model risk management guidelines, Quebec's Law 25, FSRA and AMF requirements, and incoming federal AI regulation are raising the governance bar for AI deployment. The direction of travel is clear: accountability for AI-driven decisions, explainability at the model level, and data governance are all moving from best practice to enforceable obligation.
As AI infrastructure increasingly runs on hyperscaler platforms headquartered outside Canada, insurers face a new question: do they genuinely control their own AI environment, or merely rent access to one? This distinction is attracting regulatory attention it has not previously faced.
Data sovereignty: new regulatoryfrontier
05
Agentic AI reaching production readiness
Interest in agentic AI is moving quickly from experimentation to active planning. While 82% of carriers are planning adoption within three years, only 21% report a mature model for agent governance. For many insurers, the constraint is less the capability of the technology than the readiness of the data, controls and architecture around it.
A global survey of 3,700 senior leaders found that 62% of organizations have not moved their AI initiatives beyond the pilot stage, and three in five leaders report feeling more pressure than ever to prove ROI from AI investments that are growing at 33% annually. The question is not whether to deploy agentic AI. It is whether the foundation exists to deploy it in a way that compounds rather than fragments.
82% planning agentic AI within 3 years
None of these forces is waiting for the others to resolve. They are all in play right now, in every line of business, across every Canadian carrier. The window for making foundational choices ahead of the pressure is narrower than it looks.
Statistics in this section are drawn from the sources listed at the end of this paper.
02 · Industry patterns
What we are observing across the industry
Across many insurers, AI adoption has followed a predictable path. Claims builds a solution around a high-value use case. Underwriting and distribution do the same. Each can deliver real value, but as the number of initiatives grows, so do the platforms, integrations, data pipelines and controls required to support them.
The issue is not that every capability needs to run on the same stack. It is that too many organizations are rebuilding common foundations for each use case. Data is prepared again, integrations are recreated, governance varies by team and costs become harder to see. The enterprise gets more AI, but not necessarily more reusable capability.
Data readiness is often where the problem becomes most visible. Enterprise agentic workflows need information that is accessible, reliable and governed across multiple systems. Many insurers are finding that the data exists, but is siloed across legacy platforms, structured inconsistently or difficult to access at the speed required. As a result, initiatives can stall even when the underlying AI performs well.
What works at pilot stage does not multiply into enterprise capability. It multiplies into complexity.
The lesson is not to stop experimenting. It is that the model used to launch the first ten use cases is unlikely to be the model that scales to fifty or a hundred. At that point, shared data, reusable services, common controls and clear ownership start to matter more than the speed of any individual pilot.
03 · The economics
Record investment, elusive returns
Global AI investment continues to grow rapidly. Worldwide corporate AI investment reached $252.3 billion in 2024, up 26% year on year. (Stanford HAI AI Index Report, 2025) Insurance shows the same pattern of rising investment, but the return profile remains uneven. A 2026 survey of 250 UK and US insurance managers found that 82% believe AI will dominate the industry's future, while only 14% had fully integrated AI into their financial operations. (Insurance Operations and Financial Transformation 2026) The gap between experimentation and scaled adoption remains significant.
The economics compound the problem. AI is now the fastest-growing expense in corporate technology budgets, with cloud computing bills rising 19% in 2025 as generative AI became central to operations. (Deloitte, Navigate the Economics of AI, 2026)
What makes this dynamic particularly acute for insurers considering agentic AI is how costs scale. In a single-step AI application, one user action generates one model call. In an agentic workflow, that same action can generate ten to twenty model calls as agents reason, retrieve, validate, and coordinate. Token-based economics make consumption the primary cost driver as multi-agent systems scale, and organizations that have not designed for this from the outset are discovering the cost implications only once they are committed to an architecture that is expensive to change. (Deloitte, Executive Decisions Shaping Agentic AI Value, 2026)
88% of AI proofs of concept never reach widescale deployment. For every 33 pilots launched, four make it to production.
IDC / Lenovo AI Survey, 2025
The shared foundation becomes more important as use cases move into production. A pilot can absorb bespoke integration, manual controls and one-off data preparation. An enterprise portfolio of agentic use cases cannot. These are business and operating model choices as much as technology decisions.
The more those costs and dependencies are embedded in individual solutions, the harder they become to unwind. This is why the path from pilot to production needs to address architecture, data and governance early rather than treating them as issues to solve after a use case proves itself.
04 · The stakes
The cost of waiting
The cost of waiting tends to accumulate gradually rather than arrive as a single event. As more capabilities are added, architecture becomes harder to simplify, governance more difficult to apply consistently and vendor choices more embedded in the operating model.
That does not mean every insurer needs one platform or should build everything itself. Specialized solutions will remain valuable. The issue is whether those choices are being made within a clear enterprise design, with explicit decisions about what should be shared, what can remain specialized and where portability matters.
The advantage of making those choices now is flexibility. Insurers can shape the foundation while the portfolio is still manageable, rather than retrofitting it once agentic workloads are more deeply embedded. The remaining papers in this series explore the destination, the choices around ownership and sequencing, and what it takes to execute.
05 · Closing
Questions worth asking now
|
|
|
|---|---|
Q1 |
Is the current AI architecture designed to be reused across capabilities, or does each new initiative rebuild shared foundations from scratch? |
Q2 |
Is there visibility into the total cost of AI at current scale, and a realistic model for what that cost trajectory looks like as agentic workloads grow? |
Q3 |
If nothing changes, what will another four quarters on the current path cost in architecture debt, governance exposure, and hardened vendor dependencies? |
The Compounding Intelligence series
Five standalone points of view on scaling agentic AI in insurance. Each paper is written to be read on its own; together they form a single argument.
|
|
|
|---|---|
1 |
This paper The Orchestration Gap Why record AI investment is not compounding, and why the window for foundational choices is narrower than it looks. |
2 |
The destination: how an agentic enterprise runs, and how four core insurance roles change. |
3 |
The strategic case for controlling what your AI learns, across six dimensions, and the tiered architecture that makes it practical. |
4 |
Reimagining the value chain from business outcomes, and sequencing capabilities so each inherits the intelligence of the last. |
5 |
What enterprise orchestration requires in practice, and six questions that surface the decisions worth making now. |
Melissa Carruthers
