Intelligent Automation for End-to-End Claims Processing
Client profile: Regional general insurer
9-day claims cycle, handlers spending 70 percent of time on data entry, competitor launched 48-hour promise. We built an end-to-end IDP + LLM + RPA pipeline. Cycle time to 36 hours, handler productivity 2.4x.
Headline metrics
- Cycle time reduction
- 9d to 36h
- Handler productivity gain
- 2.4x
- CSAT lift
- +14 points
- Audit defensibility
- 100%
Summary
Challenge
A regional general insurer in Southeast Asia was processing motor and home claims in seven to twelve days, with claims handlers spending 70 percent of their time on document review and structured data entry. Customer satisfaction was declining and a competitor had launched a 48-hour claims promise.
Approach
Magnis AI designed an end-to-end intelligent automation pipeline combining intelligent document processing, LLM-powered claims summarisation, RPA for system updates, and human-in-the-loop checkpoints for high-value claims. All within the existing core insurance system and aligned with regulatory record-keeping obligations.
Outcome
Average claims cycle time reduced from nine days to 36 hours. Claims handler capacity released to handle 2.4x the previous volume per FTE. Customer satisfaction score on claims experience lifted from 72 to 86. Full audit trail maintained for every automated decision.
Engagement narrative
The starting position
Motor and home claims took 7 to 12 days from FNOL (first notification of loss) to settlement. Handlers spent the majority of their day reading PDFs, scanning forms, copying data into the core insurance system, and chasing missing documents. A regional competitor had launched a 48-hour claims promise and was winning new policies on the back of it.
Discovery and design (weeks 1 to 5)
Process mining on 90 days of historical claims showed three bottlenecks: document collection (3 days median), document review and data extraction (2 days median), and back-system updates (1 day median). End-to-end process redesign before automation: replaced sequential collection with parallel collection, replaced manual review with IDP plus LLM verification, replaced manual system updates with RPA bots and API calls where available.
Build phase (weeks 6 to 16)
Intelligent document processing pipeline trained on the insurer's own claim documents (forms, police reports, medical records, repair quotes). LLM-powered summary and extraction layer with confidence scoring and human escalation. RPA bots for the back-system updates the core insurance platform did not expose via API. Workflow orchestration with human-in-the-loop checkpoints on claims above a value threshold and on any output below the confidence threshold.
Pilot and rollout (weeks 17 to 24)
Pilot on 20 percent of motor claim volume in one region. Cycle time cut to 48 hours within the first month. Handler escalation rate stabilised at 15 percent, exactly inside the original design target. Rolled out to all motor and home claims across the region in weeks 21 to 24.
Outcomes after 6 months in production
Average cycle time 36 hours across motor and home claims. Handler productivity per FTE 2.4x baseline. CSAT score lifted from 72 to 86. Audit trail captured every automated extraction and decision, allowing the insurer to defend automated decisions to its regulator. Three handlers retrained into customer-facing roles given the released capacity.
Client testimonial
“Our competitor's 48-hour claims promise forced us to act. Magnis AI did not just automate the existing process, they redesigned it so automation actually worked. Six months in, we are faster than the competitor, our handlers are happier, and customers tell us they notice.”
Head of Claims, Southeast Asian general insurer