AI for Insurance in Canada: Claims, Underwriting, Fraud, and the OSFI AI Guidance
Canada's insurance industry is in the middle of a technology transition that will reshape how risks are assessed, how claims are handled, and how customers experience their coverage. AI is the primary driver — and the Canadian regulatory environment is evolving to address it.
The insurers that will lead this transition are not the ones with the most advanced AI technology. They're the ones that combine meaningful AI capability with strong governance, clear accountability, and the operational discipline to deploy AI responsibly at scale.
The Canadian Insurance AI Landscape
Canadian insurers began serious AI investment around 2018–2020, initially in fraud detection and basic chatbots. By 2025, AI adoption has matured considerably: Intact Financial, Aviva Canada, and Desjardins have each publicly discussed significant AI programs spanning claims, underwriting, and customer experience. The P&C sector has moved faster than life and health insurance, where regulatory sensitivity and data complexity have slowed adoption.
The current state: most large Canadian insurers have AI in production for at least one material use case. Mid-size and smaller insurers are increasingly deploying AI through SaaS platforms and insurtech partnerships. The competitive pressure from AI-enabled peers is becoming a meaningful factor in market positioning.
What Canadian insurers are learning: AI delivers value most reliably when implementation is disciplined, data quality is prioritized, and governance keeps pace with capability.
Claims Processing Automation: The Highest-Impact Near-Term Use Case
Claims processing is the largest operational cost center for most property and casualty insurers, and it's where AI creates the most immediate and measurable value.
First Notice of Loss (FNOL) Automation
The initial claim intake process — recording loss details, verifying coverage, creating the claim file — is highly repetitive and ripe for automation. AI-powered FNOL processing can:
- Accept claim reports through natural language interfaces (web forms, mobile apps, phone with speech-to-text)
- Automatically extract and structure key claim information (date of loss, description, policy number, claimant details)
- Verify coverage and policy status in real time against policy administration systems
- Route claims to appropriate adjusters based on claim type, complexity, and adjuster availability
- Generate acknowledgment communications automatically
For simple claims (minor auto damage, straightforward contents loss), automated FNOL can complete the entire intake process without human intervention in 3–5 minutes — compared to 20–40 minutes for manual intake.
Canadian implementation example: A major P&C insurer deployed AI FNOL processing for auto claims through its mobile app. Within 14 months, 67% of auto claims were fully automated through FNOL with no adjuster touch required at intake. Average time from claim report to adjuster assignment: reduced from 4.2 hours to 18 minutes.
Straight-Through Processing for Simple Claims
For a meaningful subset of claims — typically simple property claims under a dollar threshold with clear coverage and straightforward loss descriptions — AI can process the claim from FNOL through payment approval with minimal or no human intervention. This is "straight-through processing" (STP).
Canadian insurers implementing STP for home contents claims (stolen electronics, broken appliances) and minor auto claims (windshield replacement, parking lot dings) report that 15–30% of claims in these categories can be STP-processed with appropriate fraud controls in place.
The economic logic is compelling: a claim that costs $85 to handle through full adjuster workflow costs $12–18 to handle through STP. For an insurer processing 50,000 minor claims per year, STP represents significant operational savings.
Document and Evidence Processing
AI document processing tools can significantly accelerate the evidence review phase of claims handling:
- Extract relevant information from medical records for disability and accident benefit claims
- Process and categorize contractor estimates and repair invoices
- Identify inconsistencies between claimed losses and photographic evidence
- Extract data from police reports and third-party documentation
- Compare current claims documentation against historical claims for the same policyholder
This reduces the time adjusters spend on data extraction and comparison — freeing them for the judgment-intensive work of settlement negotiation and complex claims resolution.
Underwriting AI: Pricing Precision and Risk Selection
AI is improving underwriting accuracy across personal and commercial lines — but this is also where regulatory and ethical considerations are most acute.
Predictive Risk Scoring
Traditional actuarial pricing uses rating factors defined by regulatory-approved algorithms. AI enriches this with predictive models that incorporate a broader set of variables and identify non-linear risk relationships that traditional models miss.
For commercial property insurance, AI models incorporating satellite imagery, building characteristics data, proximity to fire stations, historical weather data, and telematics from building management systems can significantly improve risk differentiation — identifying higher-risk properties that traditional models would price similarly to lower-risk ones.
For personal auto, telematics-based pricing (Usage-Based Insurance or UBI) is already well-established in Canada. Intact's "My Driving Discount" and similar programs use actual driving behaviour data to set premiums — a form of AI-assisted individualized underwriting that regulators have generally approved.
Underwriting Workflow Automation
Routine commercial underwriting submissions involve significant data collection and initial analysis. AI can:
- Extract structured risk information from broker submissions in unstructured formats (emails, PDF applications)
- Auto-populate underwriting worksheets from extracted data
- Flag incomplete submissions and initiate automated follow-up requests
- Score initial risk quality to prioritize underwriter attention
- Apply pricing algorithms to pre-qualified risks for straight-through quote generation
This reduces underwriting unit costs for routine commercial risks, allowing underwriters to focus attention on complex or non-standard risks where judgment adds the most value.
Fraud Detection: AI's Strongest Proven ROI
Insurance fraud costs Canadian insurers an estimated $3.2 billion annually (Insurance Bureau of Canada, 2024). AI fraud detection has the clearest and most consistent ROI of any insurance AI application.
How AI Fraud Detection Works
Rules-based fraud systems are effective but brittle: they catch known fraud patterns but adapt slowly to new schemes. AI fraud detection uses machine learning to identify:
Anomaly detection: Claims that deviate statistically from similar claims in ways that suggest manipulation (inflated repair estimates, unusual treatment patterns in accident benefits, contents claims inconsistent with property value).
Network analysis: Identifying fraud rings through relational patterns — multiple claims involving the same repair shops, healthcare providers, or legal representatives in combinations that suggest organized fraud schemes.
Temporal pattern recognition: Identifying suspicious claim timing patterns (claims filed very shortly after policy inception, multiple claims within a short period, unusual seasonality).
Cross-policy analysis: Identifying policyholders with suspicious histories across carriers (fraud detection improves significantly when insurers share anonymized fraud signals through consortium data networks like the Insurance Bureau of Canada's fraud database).
Canadian insurers using AI fraud detection report 25–40% improvements in fraud detection rates with significantly lower false positive rates than rules-based systems — reducing the cost and reputational damage of incorrectly flagging legitimate claims.
Application Fraud Detection
AI is increasingly applied to the front end of the insurance process — identifying fraudulent applications before policies are issued. This includes detecting misrepresentation of risk characteristics, identifying synthetic identity fraud in commercial applications, and flagging applications with patterns consistent with premium fraud schemes.
Customer Service: AI That Improves the Claims Experience
Insurance customers rate claims handling as their primary driver of loyalty and switching behavior. AI can meaningfully improve the claims experience:
AI-powered claims status communication: Proactive, automated updates at each stage of the claims process — claim received, under review, adjuster assigned, estimate received, settlement approved — reduce inbound status inquiry calls significantly. Insurers implementing automated claims communication report 25–40% reductions in status inquiry call volume.
Intelligent virtual agents for routine inquiries: AI chatbots and voice assistants handling policy questions, coverage inquiries, and simple claims status requests handle 40–60% of contact center volume at major Canadian insurers without live agent involvement.
Personalized claims guidance: AI tools that provide policyholders with personalized, relevant guidance during the claims process — "based on your claim, here are the next steps and what to expect" — improve satisfaction scores and reduce anxiety-driven call volume.
OSFI AI Guidance: What It Means for Your Program
OSFI's technology and operational risk guidance has evolved significantly. For federally regulated insurers, key expectations include:
Board and executive accountability: AI must be governed at senior levels. Boards should understand material AI risks and receive regular reporting on AI model performance, incidents, and emerging risks.
Model Risk Management: AI models in underwriting, pricing, and claims decisions are "models" under OSFI's model risk management expectations. They require validation before deployment, ongoing performance monitoring, and clear documentation of assumptions and limitations.
Third-party dependencies: OSFI expects insurers to manage AI risk in vendor relationships with the same rigor as other material third-party risks. This includes understanding the AI tools in use, validating their outputs, and maintaining contingency plans.
Explainability: OSFI expects insurers to be able to explain AI-driven decisions — particularly in adverse decisions (coverage denial, claim refusal, pricing for high-risk categories). "The model said so" is not sufficient explanation.
Fairness and non-discrimination: OSFI and provincial regulators (FSRA in Ontario, AMF in Quebec) are increasingly scrutinizing AI for discriminatory outcomes. Insurers should conduct regular bias audits and be prepared to demonstrate non-discriminatory outcomes.
The practical implication: AI programs need governance documentation that satisfies regulatory scrutiny. This is not optional — and it's achievable alongside meaningful AI innovation.
Where to Start: A Prioritization Framework
For Canadian insurers assessing AI investment, we recommend prioritizing in this order:
First: Claims intake automation and document processing. Clearest ROI, lowest regulatory complexity, immediate operational benefit.
Second: Fraud detection enhancement. Strong ROI, broadly applicable, OSFI-acceptable with appropriate human oversight.
Third: Customer service AI and claims status communication. High customer impact, manageable implementation complexity.
Fourth: Underwriting workflow automation (not pricing). Automating submissions processing and data entry — without changing pricing algorithms — is lower risk and still valuable.
Fifth: AI-assisted pricing and risk selection. Highest potential but requires the most robust governance, actuarial validation, and regulatory engagement.
Remolda works with Canadian insurers to design AI programs that deliver competitive advantage while meeting OSFI expectations and provincial regulatory requirements. Our insurance AI practice combines technical implementation capability with deep understanding of the Canadian regulatory environment.