Scaling GenAI Design Operations at AIG

Building a human-centered UX operating model for AI-assisted Insurance Claims workflow

Overview

Claims Assist helps commercial insurance adjusters review information and make decisions using GenAI.

When I joined, the platform had strong technical momentum, but UX practices were informal. Work often moved from request to development without enough discovery, validation, or shared checkpoints—creating rework and making it harder to build trustworthy AI experiences at scale.

My role was to bring structure to the process, strengthen human-AI interaction patterns, and expand the product’s value beyond Claims.

The Challenge

Three issues stood out:

No consistent UX process
Requests came through informal channels and moved into build before teams aligned on the problem, evidence, or intended outcome.

Unproven human-AI interactions
Adjusters were expected to review, correct, and act on AI outputs, but the experience had not been evaluated against how they actually work.

A narrow view of the opportunity
Verified claims data had potential value for Underwriting, Actuarial, and Pricing, but those needs had not been explored.

My Role

As UX and Service Design lead, I:

  • Built the UX operating model across intake, discovery, design, validation, and handoff
  • Led an AI-focused heuristic evaluation and translated findings into design priorities
  • Currently working on developing design principles for AI-assisted workflows
  • Currently leading research across Claims, Underwriting, Actuarial, and Pricing
  • Led Team Reset and Work Mapping exercises, assigned work, coached the team, and reviewed design and research outputs

Early Impact

Work is ongoing. But, here are some early results.

  • 25% reduction in design rework after implementing the structured operating model, by catching misalignment during the discovery process
  • Faster design-to-dev iteration, giving engineering earlier and more stable design specs to build against
  • Expanded research across four business functionsA repeatable UX model for GenAI delivery

Approach

1. Built a lightweight operating model

I mapped the existing workflow, identified broken handoffs, and introduced clear checkpoints across:

Intake → Discovery → Design → Validation → Handoff → Measurement

The goal was simple: catch misalignment earlier without slowing delivery.

2. Evaluated AI through the lens of trust

I focused on the moments where adjusters had to understand, verify, correct, approve, or override AI-generated outputs.

This surfaced gaps in:

  • Status visibility
  • Explainability
  • Source traceability
  • Confidence signaling
  • Error recovery
  • Auditability

The core principle: AI generates. People verify, correct, and decide.

3. Expanded discovery beyond Claims

I launched research across Claims, Underwriting, Actuarial, and Pricing to understand how verified claims data could support broader enterprise decisions. The work engaged stakeholders from frontline adjusters to global leaders and translated their needs into roadmap opportunities.

What This Demonstrates

  • Systems-level design thinking: building a repeatable operating model, not just shipping individual screens
  • Rigor in AI-specific UX evaluation: applying heuristics tailored to human-AI trust and collaboration, not generic usability checklists
  • Cross-functional influence: leading discovery that required credibility with stakeholders from adjusters to Global Leaders, across four different business functions
  • Team leadership: running a small, partially offshore team with clear OKRs and lightweight, sustainable measurement

Key Takeaways

The outcome was not just a better interface.

It was a more reliable way to design AI-assisted decisions—giving users the evidence, control, and confidence to act.

Details have been generalized to protect confidential business information.