AIG · 2026 to present
Designing AI-assisted claims at AIG
I lead design for AI-assisted claims, from the tools adjusters use today to a longer-term view of how claims could work.
- Company
- AIG
- My role
- GenAI Design Strategist
- Scope
- Lead a small UX team
- Years
- 2026 to present
- Reading time
- 4 minutes
I lead design strategy for Claims Assist at AIG, an AI-enabled claims product for commercial claims professionals. The AI can read documents, spot duplicates and contact matches, summarize a loss, and connect a claim to the right policies and endorsements. That is useful, but an adjuster still has to know what the system did, check the parts that matter, fix it when it is wrong, and make the call. I treat that as a design problem first.
My work runs on two horizons. One is making today's AI-assisted claims experience work better. The other is helping shape a more connected view of how claims could work across the organization.
The problem
We were being asked to ship AI faster than we could design it well. Work came in with no shared way to frame it. Decisions got reopened late, and design kept running into development. AI capabilities were also arriving one at a time, with no single view of how they fit into one claims experience.
We needed two things: one consistent process, and a way to evaluate AI work specifically.
Making today's product work
1. Give the team one path from request to delivery
I gave the team a single process: Intake → Discovery → Design → Validation → Handoff. Each stage answers one question, so everyone knows what "ready to move on" means.
- Intake: What problem are we solving?
- Discovery: What do people need to get done, and what limits them?
- Design: What should the experience be?
- Validation: Does it work for the people who will use it?
- Handoff: Is enough resolved to build it as intended?
2. Add a trust review built for AI
Usability checks alone don't catch the risks in AI work, so I added a review every AI flow has to pass. It asks six questions:
- Traceability: Can I see where this information came from?
- Uncertainty: Do I know how confident the system is?
- Explainability: Can I understand what the AI did?
- Human control: Can I correct it?
- Recovery: What happens when the AI is wrong?
- Auditability: Can someone work out later what happened?
The principle behind it:
Let AI do more of the information work. Keep people in control of the decision.
The AI finds, sorts, and suggests. The adjuster checks, corrects, and decides.
3. Keep listening between releases
I set up an ongoing feedback and discovery approach so we hear how claims professionals use the product, not only at launch. Adjuster feedback, heuristic evaluations, and usability testing feed a prioritized list of improvements. The same problems kept coming up: workflow continuity, data validation, AI transparency, and handoffs between people.
Shaping what comes next
Alongside the product work, I helped turn leadership's broader claims vision into a future-state experience that business and technology partners could react to. I worked with design, product, business, and claims partners on journey maps, one-page experience stories, and capability models. We also organized the work by type of claim, so the approach can scale across lines of business instead of being designed one line at a time.
This was collaborative work, and the vision changed as partner feedback came in.
What changed
UX, Product, Engineering, and the business now share a language and know when decisions get made. Design-to-development rework dropped about 25%, mainly because we caught problems earlier instead of during build.
The journey stories and vision work gave partners something concrete to react to, and tied near-term product fixes to a longer-term direction.
What I'd do differently
AI has changed how fast design moves, and speed is now the baseline. Quality still depends on feedback from real users, and that is where we struggled. Adjusters are senior professionals making decisions on high-value claims, so their time is expensive and hard to get. Reaching them on a regular basis took a lot of evangelizing inside the company.
Next time I would set up access before I needed it:
- Get a sponsor in Claims leadership early. Someone who can protect adjuster time is worth more than any recruiting plan.
- Set a standing cadence. A fixed, recurring slot with a small group of adjusters, so I'm not renegotiating access every time.
- Ask for less and make it count. Short sessions with a working prototype and a few specific questions. AI prototyping lets us show something real in days.
- Use data between sessions. Usage data and feedback already coming in can answer the routine questions, so live time goes to what only adjusters can tell us.
- Close the loop. Show adjusters and leaders what changed because of their input, and tie it to results like less rework. The next request gets easier.
Disclaimer
Details are intentionally high level and generalized to protect confidential business information.