Hollard Insurance pilots AI to speed claims - and sharpen accuracy
Hollard Insurance Australia is piloting AI inside its claims division to compress multi-page claim notes into a clear, one-paragraph summary. The goal is simple: help consultants give customers accurate updates faster, with fewer handoffs and less re-work. The pilot was revealed at a Guidewire event in Sydney.
According to Hollard's claims value owner, Daniel Dearsley, the system "synthesises an entire claim into an easily digestible paragraph with meaningful information." What usually takes a consultant minutes of scanning - sometimes dozens of pages - now appears in seconds.
Measured gains where it hurts most
Hollard reports a 70% reduction in time to review complex claims. In some cases, consultants are saving 25-35 minutes per claim file. The more notes and touchpoints, the bigger the delta.
Previously, consultants could spend up to 15 minutes piecing together the story or relying on the last note left in the file. With a clean summary on tap, they can move straight to action and communicate with confidence.
Catastrophe surge: the immediate upside
Dearsley sees the biggest payoff during disasters, when claim volumes spike and backlogs build. Even in its current form, he says the tool would deliver "tremendous upside" during catastrophe periods by accelerating triage and customer updates at scale.
For context on event frequency and pressure on claims operations, see the Insurance Council of Australia's catastrophe declarations.
Accuracy, leakage, and early guardrails
In testing, Hollard hasn't seen hallucinated content. In fact, the AI has flagged leakage cases that humans missed - for example, where an excess was applied when it shouldn't have been, or not applied when it should.
One gap remains: sentiment analysis. The model can be factually accurate yet miss signs of vulnerability in customer correspondence. That feedback loop is now a focus as Hollard refines the system and explores extending summarisation to inbound and outbound communications.
Practical takeaways for insurance leaders
- Start with high-friction claims: large loss, complex liability, high note volume. Time saved there moves the whole KPI stack.
- Build an "explainable summary" pattern: include source note anchors so consultants can click through and verify in seconds.
- Instrument the workflow: track review time, first-contact resolution, leakage detections, and customer update latency.
- Use catastrophe events as structured pilots: predefine surge playbooks that route complex files to AI-assisted review first.
- Add controls for leakage checks: codify excess and coverage rules for automatic cross-checks and alerts.
- Close the sentiment gap: fine-tune on claims email/chat data, and escalate flagged vulnerability cues to specialists.
- Keep humans in the loop: require quick validation on high-impact decisions until accuracy and trust thresholds are met.
What's next at Hollard
Hollard plans to extend the same summarisation approach to customer correspondence. That would reduce context switching for consultants and speed up accurate, empathetic replies - once sentiment detection improves.
The direction is clear: keep shaving minutes off complex reviews, protect customers from leakage errors, and scale calmly during the worst weeks of the year.
Learn more
For implementation patterns, training, and tools specific to this space, see AI for Insurance. For teams working on correspondence and frontline updates, explore AI for Customer Support.
Your membership also unlocks:
