AI / Solutions
Three-to-five-day claim cycles. Rework rates in the 20s. Senior adjusters buried in clean cases while the complex ones wait. Indonesian insurers and brokers are running claims ops designed for a volume that no longer exists. We build AI claims workflows that triage, extract, and pre-adjudicate in hours, with OJK-ready audit trails and your adjusters freed for the cases that actually need them.
A claims team's day is mostly reading. Medical reports. Police reports. KTP scans. Hand-filled forms. Mobile-photo receipts. Each file takes minutes a human can't give back. We build AI that does the first read, extracting fields, checking policy coverage, flagging fraud signals, and scoring the claim for auto-approval, human review, or investigation. Your adjusters stop clearing the queue and start handling the 15% of cases that need expertise. Every decision is logged, every override traceable, every step defensible in an OJK audit.
A four-phase path from your worst claim queue to a system your adjusters trust.
We shadow the queue. Which claim types cost the most hours per file? Where is rework happening? Which decisions would your adjusters happily hand off, and which must stay human? We map your document types, your policy logic, your routing rules, and your regulator footprint (OJK, UU PDP, BI for linked payment flows) before any model gets built.
A six-week pilot on one claim type end-to-end, motor, health, life, or travel, with a real sample of your mixed-format Indonesian claim documents. We build extraction + policy cross-check + initial adjudication scoring. Target: 80%+ straight-through-processing rate on clean cases with calibrated human review for the rest.
Where most claims AI fails, and where OJK audits land. We engineer the evaluation harness, bias testing, fraud signal tuning, and human-in-the-loop review queues. Every score gets a confidence band. Every denial gets an explanation. Every override gets a reason and a reviewer.
Handover to your claims ops and IT. Your team gets the runbook, the retraining cadence, the dashboards, the exception escalation paths. New claim types and new lines of business added as modules, not rebuilds.
Four capabilities that together replace the "someone has to read this first" bottleneck in the claims queue.
Every incoming claim routed within minutes, auto-approval track, standard review, senior review, or fraud investigation. Scoring uses your policy logic, your SLA tiers, and your historical fraud signals. Your team stops sorting the queue and starts working the cases.
Medical reports in Bahasa. Police reports with stamps and handwriting. KTP copies, mobile-photo receipts, faxed discharge summaries. We extract the fields that drive adjudication, dates, diagnoses, amounts, policy numbers, NIK, NPWP, with confidence scores and source-region citations for every field.
Once a claim is extracted, the system checks it against policy coverage, eligibility, exclusions, and claims history. Clean matches route to auto-approval. Edge cases go to human review with the discrepancy already flagged. Adjusters stop doing the policy lookup and start doing the judgment.
Every decision carries a trail, source document, extracted fields, policy rules applied, fraud signals considered, final disposition, and any human overrides. Exportable in formats OJK, UU PDP data-subject requests, and your internal audit team actually accept.
Claims automation is the single largest AI spend wave hitting Southeast Asian insurance right now. Here's the market shape we're building toward.
The global claims automation market is on track from $6.54B in 2026 to $17.09B by 2034, and 35% of global insurers are expected to deploy AI agents across three or more claims functions by end-2026. The shape of the wave: extraction, pre-adjudication, fraud signalling, auto-approval tracks.
PasarPolis alone issued over one billion policies between 2019 and 2021, evidence that Indonesian claims infrastructure already supports full-stack automation at scale. With 275 million BPJS members underpinning the market, the question is no longer whether claims AI can work here.
The April 2025 OJK AI Governance Guidance brings claims adjudication under a three-pillar framework: reliability, accountability, and human oversight. Bias testing, model validation, continuous drift monitoring, and human-in-the-loop escalation on high-risk decisions, all documented.
The April 2025 AI Governance Guidance applied to claims ops, bias testing, model validation, audit trails, and the three things adjusters still can't hand to a model.
The math of claims automation and why the most valuable pilot is the one that proves your human-in-the-loop design actually works. Straight-through ratios, exception economics, and how to pick the right bar.
A 60-person broker's path from manual triage to AI-assisted ops, without a head of data or an internal ML engineer. What to outsource, what to own, and when to bring it in-house.
Tell us the claim flow that costs the most hours, motor, health, life, travel, group, or a product you're launching into. We'll scope a six-week pilot on your real data, with clear pass/fail criteria and an OJK-ready audit trail from the first extraction.
Start a project