AI Workflow Automation in Stockholm, Sweden

Insurance (Sakförsäkring)

A Swedish P&C insurance company (2.1M policies — motor, home, and commercial property, SEK 18.5B in gross written premium, 1,800 employees) processed 340,000 claims annually through a team of 210 claims handlers (skadereglerare). The claims workflow: (1) Skadeanmälan (damage notification): customers reported claims through the company's app (38%), phone (35%), or website (27%). Phone and website notifications were transcribed into the claims system manually. App notifications captured structured data but still required handler review. (2) Document collection: each claim type required specific supporting documents — motor: photographs of damage, police report (for collisions and theft — polisanmälan), repair estimate from approved repairer. Home: photographs, repair estimates, receipts for damaged property, police report for theft/vandalism. Commercial property: as home, plus business interruption documentation (revenue loss calculations, temporary relocation costs). Average documents per claim: 5-8. (3) Coverage verification (försäkringsprövning): handlers verified coverage against policy terms — checking: that the event fell within covered perils (allriskförsäkring vs. named perils), that the policy was active on the date of loss, applicable exclusions (wear and tear, intentional damage, gross negligence), deductible amount (självrisk — ranging from SEK 1,500 to SEK 10,000 depending on policy and claim type), and any special conditions or endorsements. (4) Claim assessment (skadereglering): handlers assessed claim value — repair cost validation for motor (comparing repairer estimates against insurance industry reference data — DRG system for motor repair cost benchmarking), replacement value vs. repair for property damage, actual cash value vs. new-for-old depending on policy terms (nytt-för-gammalt coverage). (5) Liability determination (for motor claims): applying Swedish strict liability rules (trafikskadelagen — the owner of the vehicle that caused the damage is strictly liable regardless of fault, with exceptions for force majeure). For multi-vehicle accidents: apportioning liability based on contributory negligence assessment. (6) Payment and recovery: issuing claim payments, coordinating with repair shops for direct-settlement motor claims, and managing subrogation (regresskrav) against at-fault parties. The problems: (1) Processing time: average 14 business days from notification to settlement for motor claims (industry average: 10 days). Home claims: 18 business days (industry average: 14 days). Slow processing was the primary driver of customer dissatisfaction — NPS for claims experience was +12 (vs. +35 industry benchmark). (2) Document collection delays: 40% of claims were suspended pending additional documentation. Average suspension time: 11 days. Most suspensions were caused by missing photographs (the customer submitted insufficient or poor-quality images) or missing police reports (the customer had not yet filed a polisanmälan). (3) Coverage verification errors: 2.4% of claims were assessed under incorrect coverage terms — usually incorrect deductible application or failure to apply a relevant exclusion. Each error required remediation and in some cases customer complaint handling. (4) Handler capacity: 210 handlers processing 340,000 claims = 1,619 claims per handler annually. With each claim requiring 45-90 minutes of processing (notification review, document checking, coverage verification, assessment, communication), handlers were consistently at capacity. Overtime was common during peak claim periods (winter storms, summer travel damage spikes). (5) Subrogation recovery: for motor claims where the company's policyholder was not at fault, the company should recover claim costs from the at-fault party's insurer. Recovery rate: 62% of eligible claims — the remaining 38% were either not identified as subrogation-eligible or not pursued due to handler workload constraints.

We built a 4-component AI claims workflow over 10 weeks. (1) Claims Intake Engine: processing damage notifications across all channels. The engine: transcribing phone notifications using speech-to-text with Swedish insurance vocabulary (skada, självrisk, stöld, inbrott, vattenskada, trafikskada), extracting structured data from web and app notifications, automatically linking claims to policies (from 2.1M active policies — using personnummer, policy number, or property address for matching), performing initial damage assessment from submitted photographs (AI image analysis classifying damage type — stöt/buckla/repor for motor, vattenskada/brandskada/inbrott for home — and estimating severity), identifying obviously incomplete notifications and generating immediate follow-up requests (missing photos, missing police report number, insufficient damage description), and classifying claims by type and estimated complexity (simple — straightforward damage, clear coverage, likely under SEK 25,000 — vs. complex — multi-party, coverage questions, potential fraud indicators, high value). Intake processing: 3 minutes per claim (from 15-20 minutes manual). Immediate completeness check: 95% of missing documentation identified within 1 hour of notification (vs. 3-5 days manually). (2) Coverage and Assessment Engine: automating claim evaluation against policy terms. The engine: retrieving the correct policy version and applicable terms (matching policy inception date, endorsements, and renewal history), verifying coverage (peril covered, policy active, exclusions checked, deductible identified — with confidence score for each verification step), calculating claim value (motor: comparing submitted repair estimate against DRG benchmark data — flagging estimates significantly above benchmark for review; home: assessing repair costs against market rates, replacement values against depreciation schedules per policy terms), determining liability for motor claims (applying trafikskadelagen strict liability rules, apportioning contributory negligence for multi-vehicle incidents based on police report and damage evidence), and generating settlement recommendations with detailed reasoning. Coverage verification accuracy: 99.1% (from 97.6% manual). Assessment consistency: variation between similar claims reduced by 65% (eliminating handler-dependent assessment differences). (3) Subrogation Engine: automating recovery identification and pursuit. The engine: identifying subrogation-eligible claims (motor claims where the policyholder was not at fault — based on police report analysis, liability determination, and damage evidence), generating regresskrav (subrogation claims) to the at-fault party's insurer with all supporting documentation, tracking recovery progress and sending follow-up communications on schedule, and escalating disputed subrogation claims to specialist handlers. Subrogation recovery rate: increased from 62% to 88% of eligible claims — SEK 48M in additional annual recoveries. (4) Workflow Orchestration Engine: managing end-to-end claims processing. The engine: routing claims based on complexity and value (simple claims — clear coverage, under SEK 25,000, complete documentation — fast-tracked with automated assessment and direct settlement; complex claims — coverage questions, high value, fraud indicators — routed to specialist handlers with pre-assessed dossiers), generating customer communications (in Swedish — status updates, document requests, settlement notifications — maintaining a warm, professional tone aligned with the company's brand), coordinating with repair shops (for motor direct-settlement claims — sending authorisation and tracking repair progress), producing management reports (processing times, settlement ratios, handler productivity, customer satisfaction metrics), and generating Finansinspektionen regulatory reports.

Motor claims processing: 5 business days (from 14 days). Home claims: 8 business days (from 18 days). Both significantly better than industry averages. NPS for claims experience: improved from +12 to +41 (industry benchmark: +35 — the company moved from below average to above average). Document collection: missing documentation identified within 1 hour (from 3-5 days). Claims in suspension: reduced from 40% to 18% (faster document identification and automated follow-up). Coverage verification accuracy: 99.1% (from 97.6%). Coverage verification errors requiring remediation: reduced by 65%. Assessment consistency: 65% reduction in handler-dependent variation — same type of claim now receives consistent assessment regardless of which handler reviews it. Subrogation recovery: 88% of eligible claims (from 62%) — SEK 48M in additional annual recoveries. Straight-through processing: 32% of claims processed without human intervention (simple, well-documented claims with clear coverage). Claims team: reduced from 210 to 155 handlers (55 positions eliminated through attrition). Remaining handlers focus on complex claims, fraud investigation, and customer relationship management for high-value or sensitive claims. Annual value: SEK 68M (operations staff savings) + SEK 48M (improved subrogation recovery) + SEK 22M (faster processing — reduced complaint handling, improved retention) = SEK 138M (approximately €12M). Development cost: SEK 2.8M (approximately €242K). Payback: under 8 days.