Every insurer today is exploring AI. But moving from GenAI “experiments” to production-grade agentic AI requires a different mindset and a different playbook.
In our recent webinar, Dufrain’s Owen Greenwood and Fraser Black, along with Microsoft’s Jamie Taylor, shared practical, grounded advice for insurers who want to start strong.
Missed the session – here’s a summary and you can watch the full 30 minute discussion and demo here.
1. Start with one high-value use case
Owen and Fraser stressed this repeatedly: don’t generalise the use case.
Pick something:
- Tangible
- Measurable
- Repeatable
- High impact
Underwriting submission triage was given as a perfect example.
2. Define clear success criteria upfront
Agentic systems can grow quickly and become too general-purpose.
Fraser emphasised the need to “ring fence” the use case and define:
- What success looks like
- What data is required
- What actions the agent is allowed to perform
3. Build the right skills early
Agentic AI is not “plug and play”.
Insurers need:
- Engineering expertise
- Data pipeline capability
- AI Foundry familiarity
- Product thinking
As Fraser noted, ChatGPT has unintentionally “oversimplified” expectations. Real agentic AI requires engineering discipline.
4. Prepare your data for AI, especially policy admin data
Agentic AI can absolutely integrate with policy admin systems, but insurers must modernise data into AI-ready formats.
Fraser’s checklist included:
- A modern platform (Fabric, Databricks or Azure)
- Accessible structured and unstructured data
- Vector-ready storage
- Pipelines that make data consumable by agents
5. Build, don’t buy if you want long-term value
Off-the-shelf tools have a place, but they ultimately limit differentiation.
Jamie highlighted that insurers gain the strongest competitive advantage when they build solutions grounded in their own data and IP.
6. Use Copilot as your “UI for AI” for adoption
Fraser showed how integrating agent workflows into Copilot removes friction and drives business adoption, the biggest barrier to production success.
7. Move fast, the technology is ready today
Azure AI Foundry, Copilot, orchestration workflows and multi-agent capabilities are all already live. The winning insurers will be the ones who stop exploring hypotheticals and start building.
Don’t miss the full 30 minute discussion and demo – watch it here.
If you’d like to here more information about Agentic AI and the webinar, we have a suite of blogs from the webinar. Blog 1 and Blog 2.