CII Conference Looks at AI Outcomes for the Insurance Sector
AI is a tool. Like a policy document or a pricing model, it helps us decide, analyse, and assess risk. The responsibility sits with insurance to make AI useful to customers and society, not just balance sheets or headcount.
At the CII's Future of Insurance conference, Professor Richard Susskind challenged the room to think beyond current delivery models. Short term: automation. Next decade: clients and organisations empowered by AI to get outcomes without traditional intermediaries.
Key points from Richard Susskind
- Automation first, empowerment next: The near-term gains are workflow automation. The bigger shift is clients using AI to get results directly.
- Your future competitor may be your client: If customers can generate advice, quotes, and claims decisions with AI, they will.
- Build vs. compete: Do you compete with these systems, or build them? Someone will. That's a product decision, not a press release.
- Sell outcomes, not process: Make the result faster, cheaper, less intrusive, more convenient. That's the bar for genuine innovation.
Message from CII leadership
Matthew Hill urged the market to treat technology as a tool, not a threat. The goal isn't how fast you adopt AI, but how wisely and profitably you use it-and how much you invest in people alongside platforms.
What this means for insurers, brokers, MGAs, and claims leaders
- Underwriting: Use AI for triage, appetite checks, and pre-bind data enrichment; keep human judgment for edge cases and capacity calls.
- Claims: Straight-through processing for low complexity; human-in-the-loop for injury, fraud, and coverage disputes.
- Distribution: Quote-and-bind experiences move closer to the client. Embed advisory tools in portals and partner ecosystems.
- Risk and pricing: Scenario testing with synthetic data, continuous monitoring, and clear escalation paths when confidence drops.
- Customer outcomes: Measure time-to-resolution, transparency, and fairness-not just cost-to-serve.
A practical 90-day plan
- Inventory outcomes: List your top customer outcomes (quote accuracy, claim resolution time, fraud reduction). Pick one to improve with AI.
- Select use cases: Choose 3 high-volume, low-variance tasks for pilots (e.g., FNOL summarisation, document extraction, broker email routing).
- Define guardrails: Write clear policies for data handling, human oversight, and model boundaries. Keep an approval matrix simple and visible.
- Measure what matters: Set target SLAs, error thresholds, and escalation triggers. Track customer effort and complaints.
- Upskill teams: Train adjusters, underwriters, and ops on prompt quality, verification, and exception handling.
Governance that stands up to scrutiny
- Explainability: Log inputs, outputs, and rationale for every automated decision. Make it auditable.
- Fairness checks: Test for proxy bias across protected groups. Document remediations.
- Human-in-the-loop: Require review for high-impact decisions and low-confidence outputs.
- Model lifecycle: Version control, drift monitoring, rollback plans, and periodic revalidation.
- Vendor diligence: Demand data lineage, security controls, and incident response commitments.
Build vs. buy: choose with intent
- Build when the task is proprietary, high volume, and a differentiator (e.g., underwriting workbench logic, fraud signals).
- Buy for commodity layers (OCR, summarisation, translation) to speed time-to-value and reduce maintenance overhead.
- Integrate with existing core systems to avoid shadow IT and orphaned pilots.
Skills and capability
Tools change fast. Capability compounds. Invest in frontline skills-prompting, verification, and exception handling-alongside data and model stewardship.
If you're building a training plan by job function, explore a curated set of practical courses here: AI courses by job.
On-demand access
The conference will be available on demand from 7 November. Check the CII events page for details: CII events.
Your membership also unlocks:
