AI automation that does the work software should be doing — for Insurance

Insurance carriers and brokerages run on legacy systems — most core policy admin, claims, and underwriting platforms in DACH and the broader EU were architected 15–30 years ago and have grown layers of integrations on top. The result is a sector where data lives in 8–15 systems per business unit, where analytics runs through Excel exports and weekly reports, and where every new product launch waits 6–12 months for IT bandwidth. The decision-makers are typically risk-averse for good regulatory reasons (BaFin, EIOPA, Solvency II) but increasingly under board-level pressure to modernise. The buying pattern: cautious vendor selection, long procurement cycles, but real budget once trust is established and a small win has been shipped.

• ▸Claims triage models must be explainable to claims handlers and regulators — SHAP values per decision, not just a confidence score.
• ▸Document-extraction (OCR + LLM) for medical reports, expert opinions, accident reports — the biggest near-term ROI lever in P&C claims.
• ▸Human-in-the-loop everywhere — every AI suggestion routes through an adjuster, with an audit trail of what was changed and why.
• ▸Integration with legacy claims systems via batch or message queue, not always live API — many policy admin systems can't support real-time AI calls.
• ▸Bias and fairness testing as a regulatory artefact — DORA and the EU AI Act both require demonstrable equal-treatment evidence.
• ▸Start narrow — one line of business (e.g., kfz/auto), one workflow (first notice of loss → triage), measurable in 6 weeks.

We helped a DACH multi-line insurer cut first-notice-of-loss triage time from 38 minutes to under 4 minutes by combining LLM-based document extraction with a rule-based routing engine — without changing the underlying claims platform.