Data Analytics in Toronto, Canada

Insurance / P&C / Data Analytics / Toronto

A Canadian property and casualty insurer (CAD $3.8B in direct written premium, operating across all 10 provinces and 3 territories, 2.4 million policyholders, specialising in personal auto, home, and commercial lines, with 3,800 employees and a network of 4,200 independent broker partners) needed a claims intelligence platform to address a deteriorating loss ratio that was threatening the company's competitive position. The insurer's combined ratio had increased from 96.2% to 101.8% over three years — crossing the profitability threshold — driven primarily by claims performance. Core challenges: (1) Ontario auto insurance claims — Ontario representing 48% of the insurer's premium but 62% of claims costs. Ontario's auto insurance system (unique in Canada with its hybrid tort/no-fault structure, mandatory accident benefits, and regulated rate environment) creating complex claims that averaged 2.3x the cost of auto claims in other provinces. Fraud: Ontario auto insurance fraud estimated at CAD $1.6B industry-wide — the insurer estimating 8-12% of its Ontario auto claims had fraudulent or inflated elements, but its detection capability was catching only 3%. Fraud rings operating sophisticated schemes involving staged collisions, phantom passengers, and coordinated treatment mill referrals — patterns that individual adjusters couldn't detect because they only saw individual claims, not the network. Accident benefits: Ontario's Statutory Accident Benefits Schedule (SABS) creating complex claims where benefits (income replacement, medical/rehabilitation, attendant care) interacted with tort claims — analytics needed to manage both streams simultaneously and identify when benefit utilisation patterns suggested fraud or overtreatment. (2) Climate-driven catastrophe losses — the insurer's catastrophe losses increasing 340% over 10 years, reflecting Canada's accelerating climate impacts. Key events: Toronto urban flooding (the insurer's largest single loss event at CAD $120M), Alberta hailstorms (CAD $45M in a single event), BC atmospheric rivers (CAD $68M), and wildfire evacuation costs across Alberta and BC. Catastrophe claims required different analytics than attritional claims — surge staffing, vendor management, coverage determination at scale, and reserve accuracy when loss development was uncertain. The insurer's catastrophe response was reactive — scrambling after each event rather than using analytics to pre-position resources and predict claims volume by geography. (3) Claims leakage — the insurer estimating 6-8% claims leakage (overpayment through errors, missed subrogation, incorrect coverage application, and undetected fraud). On CAD $2.28B in annual claims costs (60% loss ratio on $3.8B premium), 7% leakage represented CAD $160M — a massive profit improvement opportunity. But leakage was invisible without analytics — individual claim overpayments of CAD $500-5,000 were undetectable by adjusters handling 120+ claims each. Only pattern analysis across thousands of claims could identify systematic leakage — specific repair shops consistently charging 15% above market, specific medical providers billing for treatments inconsistent with injury severity, or specific coverage interpretations that were more generous than policy wording required. (4) Provincial regulatory variation — the insurer operating under 13 different regulatory regimes. Auto insurance: regulated rates in Ontario, Alberta rate cap, BC public insurance (ICBC) meaning no private auto in BC, Saskatchewan Government Insurance (SGI) meaning limited private auto in SK, Manitoba Public Insurance (MPI) meaning no private auto in MB. Property: different building codes, different flood zone definitions, different reconstruction cost indices across provinces. Claims handling: different limitation periods, different dispute resolution mechanisms (Ontario's LAT, Alberta's ADR), and different mandatory reporting requirements. Each province requiring different analytics — Ontario loss ratios calculated differently than Alberta loss ratios, claims reserves held to different standards, and regulatory filings using different formats and schedules. (5) Broker relationship analytics — the insurer distributing exclusively through 4,200 independent broker partners. Broker quality varied enormously — some brokers writing profitable, well-underwritten business while others consistently placed high-risk, underpriced risks. But the insurer had no analytics on broker-level performance — loss ratios, fraud rates, customer retention, and complaint rates at the broker level were uncalculated. The insurer couldn't distinguish between a broker whose auto portfolio had a 95% loss ratio (excellent) and one with a 115% loss ratio (toxic) — both received the same commission rates and service levels.

We built a claims intelligence platform over 14 weeks — providing fraud detection, leakage identification, catastrophe analytics, and provincial intelligence across the insurer's Canadian operations. (1) Fraud detection network: graph-based analytics for Ontario auto fraud. Social network analysis: building a relationship graph connecting claims, claimants, vehicles, addresses, phone numbers, legal representatives, medical providers, and repair shops. When a new claim was filed, the system immediately identifying connections to known fraud indicators — "this claimant shares a phone number with 3 other claimants who all had accidents at the same intersection within 6 months, all represented by the same law firm, all treated at the same clinic." Pattern detection: machine learning models trained on 5 years of confirmed fraud cases — identifying patterns including: staged collision indicators (specific intersections, specific vehicle damage patterns, specific time-of-day patterns), treatment mill referral networks (claimants routed to specific providers regardless of injury type or geography), and phantom passenger indicators (accident benefits claimed by passengers whose involvement couldn't be independently verified). Provider analytics: medical and rehabilitation provider profiling — identifying providers whose billing patterns were statistical outliers: average treatment duration 3x peer average, billing frequency inconsistent with diagnosis, and treatment protocols not aligned with clinical guidelines. Provider scorecards distributed to adjusters, flagging high-risk providers for enhanced review. Fraud scoring: every new claim receiving an automated fraud score (0-100) within 15 minutes of first notice of loss — high-scoring claims routed to the Special Investigations Unit (SIU) immediately rather than after weeks of normal processing. Adjusters receiving fraud indicators with specific reasons: "fraud score 78 — claimant connected to 2 prior staged collision claims, treating at flagged provider, legal representation engaged before medical treatment." (2) Claims leakage detection: systematic overpayment identification. Repair cost analytics: comparing auto repair estimates against market benchmarks by geography, vehicle type, and damage type — identifying repair shops that consistently charged above market for specific repair categories. "This shop charges 22% above market average for structural repairs on Toyota Camrys in the GTA — review recommended." The system learning from millions of repair estimates to build granular market benchmarks that no individual adjuster could maintain. Medical cost analytics: accident benefits treatment cost analysis by diagnosis, geography, and provider — identifying claims where treatment costs were statistically inconsistent with injury severity. "This claimant has a minor soft tissue injury but treatment costs are tracking at 95th percentile for this injury type — peer review recommended." Coverage analytics: automated coverage determination cross-referencing claim circumstances with policy terms — identifying claims where coverage was more generously applied than policy wording warranted. Not overriding adjuster decisions but flagging: "this water damage claim was paid under sewer backup coverage, but the adjuster notes describe surface water — sewer backup coverage requires sewer system failure, which isn't documented." Subrogation identification: automated identification of subrogation opportunities — claims where a third party was at fault but subrogation hadn't been pursued. The system identifying CAD $38M in subrogation opportunities in Year 1 that would have otherwise been missed. (3) Catastrophe intelligence: pre-positioning and response analytics. Weather integration: real-time weather data (Environment and Climate Change Canada, private weather services) feeding predictive models — forecasting claims volume by postal code for approaching severe weather events. "Severe thunderstorm warning for GTA: estimated 2,400-3,200 claims within 72 hours, primarily auto hail damage in Brampton-Mississauga corridor based on storm track and vehicle exposure density." Surge resource planning: catastrophe analytics driving proactive resource deployment — independent adjusters pre-engaged, vendor networks activated, and call centre capacity increased before events hit, rather than after claims began arriving. Claims triage at scale: during catastrophe events, automated triage sorting thousands of claims by severity, coverage complexity, and fraud risk — ensuring the most complex and highest-value claims received immediate expert attention while straightforward claims (e.g., confirmed hail damage to newer vehicles with comprehensive coverage) moved through expedited processing. Reserve accuracy: catastrophe reserve models incorporating historical loss development patterns specific to each peril type (flooding reserves develop differently than hail reserves) and geography — providing CFO with accurate reserve estimates within 48 hours of an event rather than the 4-6 weeks previously required. (4) Provincial intelligence: regulatory-aware analytics across Canada. Provincial dashboards: separate analytical views for each province, reflecting different regulatory regimes, different coverage structures, and different market dynamics. Ontario auto analytics completely different from Alberta auto analytics — the platform understanding and applying provincial context automatically. Regulatory reporting: automated provincial regulatory filings (FSRA for Ontario, AIRB for Alberta, BCFSA for BC) — each province requiring different formats, different metrics, and different schedules. The platform generating compliant filings automatically, replacing manual quarterly processes. Rate filing support: analytics supporting rate filings with provincial regulators — particularly Ontario's FSRA, which requires actuarial justification for rate changes supported by credible data analysis. The platform providing filing-ready analytics that accelerated rate approval processes. (5) Broker intelligence: distribution performance analytics. Broker scorecards: comprehensive broker performance analytics — premium volume, loss ratio (reported and developed), policy retention rate, fraud claim rate, customer complaint rate, and growth trajectory. Updated monthly and shared with broker management teams. Segmentation: brokers classified into performance tiers — top performers receiving enhanced commission, increased binding authority, and priority service; underperformers receiving targeted improvement plans with specific analytics showing where their portfolio was underperforming. Not punitive but informational: "your auto portfolio in Ottawa has a 118% loss ratio driven by a concentration of high-risk drivers in the 18-25 segment — here are underwriting guidelines that your peers use to manage this segment profitably." Broker meetings: automated quarterly business review packs generated for each broker — complete with performance trends, peer benchmarking, and specific improvement recommendations. Broker managers attending meetings with data-driven talking points rather than anecdotal impressions.

Results over 18 months. Fraud detection: Ontario auto fraud detection rate from 3% to 14% of suspected fraudulent claims — CAD $67M in fraud savings in Year 1. SIU referral precision improving 4x — fewer false positives meant SIU investigators could focus on genuine fraud. The social network analysis identifying 4 previously unknown fraud rings involving 89 connected claims totalling CAD $12.8M. Medical provider analytics flagging 23 providers for enhanced review — resulting in 6 regulatory referrals to FSRA for suspected billing fraud. Leakage reduction: total claims leakage from estimated 7% to 3.2% — saving CAD $86.6M annually on the $2.28B claims base. Repair cost analytics identifying 180 repair shops with above-market pricing — renegotiated agreements reducing average repair costs 8%. Subrogation recovery: CAD $38M in identified opportunities, CAD $26M recovered in Year 1. Coverage analytics identifying CAD $14M in overpayments from incorrect coverage application — corrective training reducing future errors. Catastrophe response: the 2025 Toronto derecho generating 4,200 claims in 48 hours — catastrophe intelligence enabling pre-deployment of 45 independent adjusters and 3 vendor teams before the storm hit. Claims triage at scale: 78% of straightforward claims processed within 5 business days (versus 15-day average in previous catastrophe events). Reserve accuracy: CFO receiving credible reserve estimates within 48 hours (versus 4-6 weeks) — improving quarterly financial reporting accuracy. Combined ratio: from 101.8% to 95.4% over 18 months — a 6.4-percentage-point improvement worth CAD $243M annually on the $3.8B premium base. The claims intelligence platform directly responsible for approximately 60% of this improvement (3.8 percentage points, CAD $146M). Provincial compliance: automated regulatory filings across all provincial regulators — FSRA, AIRB, BCFSA quarterly filings produced in 2 days (from 3 weeks manual). Rate filing turnaround improved 40% — faster regulatory approval enabling more responsive pricing. Broker management: bottom quartile broker loss ratios improving 12 percentage points within 12 months of receiving performance analytics. Three broker relationships terminated based on persistent underperformance and refusal to implement underwriting guidelines — removing CAD $28M in toxic premium that had a 134% loss ratio. Top performer recognition driving 8% premium growth from highest-quality brokers. Financial impact: fraud detection CAD $67M; leakage reduction CAD $86.6M; subrogation CAD $26M; catastrophe efficiency CAD $18M; broker management CAD $22M. Total annual impact: CAD $219.6M. Development: CAD $840K. Annual platform: CAD $220K. ROI: 261x first year.