AI Mobile App Development in Melbourne, Australia

Insurance / Motor and Home

A Melbourne general insurer (A$1.5 billion GWP, 1.8 million customers, predominantly motor and home insurance) was processing 340,000 claims annually. The claims lodgement process was friction-heavy: customers called the contact centre (average 12-minute wait), described the damage verbally to an agent (15-20 minutes), then separately emailed or posted photographs and supporting documents. First notification of loss to claim assessment took an average of 4.2 business days. Customer satisfaction with the claims process (the moment of truth for insurance) scored 3.1/5 — the lowest-scoring aspect of the customer experience. Competitors had launched mobile claims lodgement but with basic photo upload and forms — not genuinely intelligent AI-powered claims experiences. The Head of Claims wanted to leapfrog competitors with an AI-powered claims app that: enabled customers to lodge claims in minutes (not days), used AI to assess damage from photos (providing immediate preliminary assessment), and automated the end-to-end process from notification to assessment assignment.

We built the AI claims features into the insurer's existing mobile app over 12 weeks. Phase 1 (weeks 1-4): AI damage assessment model. We trained a damage assessment model on 120,000 historical claims photos (motor and home): motor vehicle damage (trained on photos categorised by damage severity and repair type — scratch, dent, panel damage, structural damage, glass damage, total loss indicators), home property damage (water damage, storm damage, fire damage, impact damage, theft/break-in — each with severity classification), and damage location mapping (identifying which part of the vehicle or property is damaged — front bumper, driver door, roof, kitchen, bathroom — enabling automated routing to the appropriate assessor). The model runs as a hybrid: initial classification on-device using CoreML/TensorFlow Lite (instant feedback — "We can see damage to the front bumper. Can you take a photo from a different angle?"), with detailed assessment on the server (using the higher-resolution images for severity assessment and repair cost estimation). Phase 2 (weeks 3-8): claims lodgement flow. The AI claims flow: (1) Incident capture — customer taps "Lodge a Claim" and selects claim type (motor/home). For motor: the app uses GPS and time to pre-populate incident date, time, and location. (2) Damage photography — guided camera experience: the app shows the customer exactly what photos to take (for motor: front, rear, left, right, damage close-up, dashboard showing odometer; for home: wide room shot, damage close-up, source of damage if visible). AI provides real-time feedback: "Good photo — damage clearly visible" or "Please retake — photo is too blurry to assess." (3) AI damage assessment — on-device model provides immediate classification: "Front bumper damage — moderate severity. This appears repairable rather than replaceable." The customer sees that the AI has understood their claim — building confidence in the process. (4) Supporting information — the app collects: other party details (for motor — with OCR from the other driver's licence), police report number (if applicable), witness details, and a voice or text description of what happened. (5) Document AI — the customer photographs relevant documents (repair quotes, receipts, policy schedule) with AI extraction of key data (repairer name, quote amount, invoice details). (6) Claim submission — the complete claim (photos, AI assessment, supporting information, extracted documents) is submitted to the claims management system with an AI-generated claim summary. Phase 3 (weeks 6-10): automated triage and assignment. AI claim triage: the submitted claim is automatically classified by: severity (minor — likely under excess, moderate — standard claims process, severe — priority handling required, potential total loss — specialist assessment), complexity (straightforward — AI assessment sufficient for initial processing, moderate — requires desktop assessor review, complex — requires field assessor or specialist), and fraud risk (low, medium, high — based on claim characteristics, customer history, and fraud pattern detection). Automated assignment: straightforward, low-fraud-risk claims are assigned to the next available desktop assessor with the AI assessment pre-populated. Complex claims are routed to specialist teams. High-fraud-risk claims are flagged for the Special Investigations Unit. The assessor receives: the complete photo set, the AI damage assessment with severity and repair type classification, extracted document data, the customer's description, and a recommended next step (approve for repair, request additional information, arrange inspection). Phase 4 (weeks 8-12): integration, testing, and launch. Integration with the insurer's Guidewire ClaimCenter: claims created directly from the mobile submission with all data mapped to ClaimCenter fields. Testing: 2,000 claims processed through the AI flow in parallel with the existing manual process — comparing AI damage assessment accuracy against assessor decisions.

The AI claims feature launched as an update to the existing mobile app. Claims lodgement time: average 8 minutes on mobile (from 35+ minutes through the contact centre — including wait time). Customers could lodge claims at 2am after an incident, in the Uber home from an accident, or standing in their flooded kitchen — not waiting until the next business day to call during business hours. AI damage assessment accuracy: motor vehicle damage severity classification: 89% agreement with experienced assessor judgment. Home property damage classification: 84% agreement. The AI assessment is presented as preliminary — the assessor makes the final determination, but the AI pre-work reduces assessor time by 40%. Claims processing speed: first notification of loss to assessment assignment: reduced from 4.2 business days to 0.4 business days (90% reduction). Mobile-lodged claims are automatically triaged and assigned — the assessor receives a pre-assessed claim within hours rather than days. Adoption: 42% of new claims lodged through the mobile app within 4 months (from 0% — all claims previously lodged through the contact centre). Mobile claims lodgement was highest for motor claims (48% — often lodged at the scene or immediately after) and growing for home claims (35%). Contact centre impact: claims-related call volume reduced by 28% — the contact centre handled complex claims and customers preferring phone, while straightforward claims were increasingly mobile-lodged. Contact centre agents reported handling more interesting, complex cases rather than spending time collecting basic information over the phone. Customer satisfaction: claims experience satisfaction improved from 3.1/5 to 4.2/5 within 4 months. The primary drivers: immediate lodgement (no waiting on hold), AI feedback during photography (the customer felt the insurer was already assessing their claim as they photographed damage), and faster processing (assessment within hours rather than days). Fraud detection: the AI fraud risk scoring flagged 3.2% of mobile claims for investigation — of which 68% were confirmed as having fraud indicators. The visual analysis detected 12 cases of staged damage and 8 cases of pre-existing damage being claimed as new — damage patterns that phone-based claims descriptions would not have revealed. Annual savings: A$3.8 million (contact centre reduction, assessor efficiency, fraud prevention). Customer retention: claims experience improvement contributed to a 2.1 percentage point retention improvement at renewal — estimated A$6.3 million in retained premium. Implementation cost: A$340,000. First-year ROI: 11.2x on direct savings, with retention value providing additional return.