A Deloitte Canada Insurance Point of View - 2026

01 · The unrecognized decision

A choice being made by default

Whether intentional or not, an insurer routing its highest-volume workflows through a frontier model it does not own is doing three things. It is donating the learning from those workflows to the vendor. It is accepting governance constraints it did not set. And it is building a dependency that gets harder to exit every quarter it continues.

Most insurers are making this decision without realizing they are making it. Selecting a claims AI vendor, standardizing on a single frontier provider, signing a multi-year hyperscaler agreement for a GenAI platform build: each of these is an intelligence ownership decision, even if none of them are recognized as one. By the time the strategic dimension becomes visible, the dependency is already entrenched and the architectural consequences are compounding. The first step toward making these choices deliberately is recognizing them as choices at all.

The argument for owned intelligence extends across six dimensions. The first two, economics and performance, are table stakes. The remaining four are where the lasting differentiation lies.

02 · Six dimensions

The case for owned intelligence

Table stakes

01 Economics

At agentic scale, a single workflow can generate ten to twenty model calls. Routing high-volume insurance work through frontier models at that volume carries a compounding cost premium that pilot economics simply do not reveal. Organizations using purpose-built models for their highest-volume workflows are seeing meaningfully lower cost per interaction. The economics case is real and at agentic scale it compounds significantly, though it is the least strategically interesting reason to own your models. (Deloitte, Navigate the Economics of AI, 2026)

02 Performance

A model fine-tuned on ten years of claims data from a specific carrier, calibrated against that carrier's policy language and underwriting guidelines, and validated against known failure modes will consistently outperform a general-purpose frontier model on those same tasks. Breadth matters for open-ended reasoning, but for specialist, high-volume insurance work, purpose-built wins every time. The performance advantage is measurable, and in high-volume workflows it compounds with every transaction.

Where lasting differentiation lives

03 Ownership of accumulated intelligence

Every time a frontier model processes a claims document or evaluates an underwriting submission, the system learns from that interaction. When you rent the model, that learning flows to the vendor, not to you. Over time, the carrier that owns its models builds a proprietary asset that deepens with every transaction, while the carrier that rents keeps paying for capability it will never own.

04 Governance portability and policy alignment

An owned model stack means governance travels with the insurer. Responsible AI controls, bias testing, explainability requirements, and model validation standards can be enforced at the model level, not only through the application layer. This distinction matters enormously under OSFI's E-23 guidelines, which require insurers to demonstrate how AI-driven decisions are made and governed. An insurer relying on a third-party model can describe what the model produced. An insurer that owns its model can explain how and why, at every level of the stack. When regulators ask hard questions, that distinction separates a credible answer from a qualified one.

05 AI and data sovereignty

For carriers operating under strict data residency rules, or in markets where certain cloud providers are restricted, owned infrastructure is a hard requirement, not a strategic preference. IBM's January 2026 launch of Sovereign Core made the stakes explicit: digital sovereignty goes beyond data residency. It encompasses who operates the environment, how data is governed, where workloads execute, and under whose jurisdiction models run. (IBM Sovereign Core, January 2026)

The scale of this shift is significant. Analysis suggests 30 to 40 percent of global AI spending could be shaped by sovereignty requirements by 2030, and Gartner estimates more than 75 percent of enterprises will have a formal digital sovereignty strategy in place by that date. The vendor response has been to market sovereignty as a service: localized clouds, compliance wrappers, managed environments. But as the Tony Blair Institute documented, these offerings often simulate autonomy while deepening the dependency they claim to resolve. (Tony Blair Institute, Sovereignty in the Age of AI, January 2026)

06 Frontier model optionality

Owning the orchestration layer does not mean rejecting frontier models. It means preserving the ability to choose among them. The frontier model landscape will continue to evolve rapidly. The insurer that owns its orchestration layer can integrate whichever frontier capability adds the most value at any given time, swap providers as the market shifts, and avoid being locked into any single vendor's roadmap. The insurer that has built its architecture around a single frontier provider has, by contrast, made a decade-long bet on that provider's continued relevance, pricing discipline, and regulatory acceptability. That bet is significant, and largely unexamined.

03 · Making it real

The practical architecture

The practical architecture is built in tiers: purpose-built small language models for the high-volume, well-defined insurance work where proprietary training creates a genuine edge; frontier models called in selectively where breadth and reasoning matter. The insurer owns and governs the orchestration layer throughout, and every transaction builds the insurer's intelligence rather than someone else's.

Owned intelligence does not mean built entirely in-house. Most insurers will partner on execution, with technology providers, system integrators, and specialist vendors. Investment capacity, internal expertise, speed to market, and strategic priorities all shape what an organization builds versus what it sources. The question is not whether to partner, but what to keep control of when you do: the orchestration layer, the governance, and the intelligence that accumulates with every transaction. Partnering on how it gets built is entirely compatible with owning what it produces.

Done deliberately, this architecture compounds with every transaction: every pearl added to the string inherits the intelligence of everything before it, a sequencing discipline we set out fully in The String of Pearls, the fourth paper in this series. The system gets harder to replicate as it matures. That is the difference between renting capability and building an asset.

04 · Closing

Questions worth asking now

Q1 When your highest-volume workflows run today, who captures the learning: your organization, or the vendor whose model processed them?

Q2 Can the current platform operate in every geography the organization serves, including markets with data residency requirements or cloud provider restrictions?

Q3 If the economics of a key AI vendor relationship changed materially, would the organization have meaningful flexibility to respond, or is it effectively locked in?

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.