Business audit Experience
How do you ask a business user to complete a complex insurance audit with as little friction as possible? By running the research to find out exactly what you can stop asking — then redesigning around that answer.
Business users were completing insurance audits through a process that felt overwhelming and opaque — answering questions that weren't always necessary and uploading documents without understanding why. The goal wasn't just to simplify the UI. It was to determine the absolute minimum information and documentation needed for a valid audit, then design a flow that collected exactly that — no more.
I led end-to-end UX research and design — running click tests to identify navigation failures, designing and executing an A/B test on our data grid component, and conducting moderated user studies with two distinct audiences: internal auditors who administer the process and external customers who experience it. I introduced a new RAS (Recommended Adoption Score) tracking method to categorize and prioritize the 17 issues surfaced. Findings were synthesized into a stakeholder readout with formal recommendations.
Two users. One process. Completely different mental models.
Internal auditors understand exactly what documentation they need and why — but have no visibility into how their requests land on the other side. External customers experience the audit as an opaque, high-stakes checklist with no clear rationale. Designing for both required running two parallel research tracks simultaneously: one to define what was truly required, and one to understand where external users were confused, hesitating, or providing incomplete information.
Click Testing
Identified 2 specific points of navigation confusion where users were being directed off the correct path. Updated language and removed options that were derailing users before they reached the audit questions.
A/B Testing
Tested a live prototype migrating data tables from AG Grid to our in-house design system. Results were comparable with a slightly improved pass rate — providing the data-backed green light to proceed with the full component conversion.
Moderated User Studies
7 sessions across 3 internal auditors and 4 external customers via moderated think-aloud and Figma prototype walk-throughs on Userzoom. Preliminary Questions scored 6.3/7 and Document Upload 6.1/7 on Ease of Use, with the prototype rated 6.0/7 against the current process. Sessions surfaced 17 distinct issues, severity-categorized using a new RAS (Recommended Adoption Score) tracking method introduced for this project.
17 issues. 6 that couldn't wait.
Rather than presenting a flat list of problems, I introduced the RAS — Recommended Adoption Score — as a new severity-categorization framework for tracking research findings at Travelers. The 17 issues identified were classified by impact: all critical and high-severity items (6 total) were flagged for immediate resolution. The RAS gives product and engineering teams a clear, prioritized action list rather than an undifferentiated pile of feedback.
Ask less. Verify more.
The redesigned flow eliminated 3 questions entirely and reworded 4 others for clarity. Three UX patterns were introduced: conditional logic so questions only appear based on previous answers, a multi-step format breaking the audit into manageable stages, and inline validation giving users immediate feedback on their inputs. Three new features were added to reduce manual effort: a printable checklist, automatic pre-population of last year's data, and AI-powered document scraping — which reads uploaded payroll and tax documents, extracts the relevant data, and verifies it automatically so users don't have to enter it twice.
Prototype Beat the Status Quo
6.0/7 vs. current process · 6.3 prelim · 6.1 upload.
Across 7 moderated sessions, the prototype scored 6.0/7 against the current workflow, with Preliminary Questions at 6.3/7 and Document Upload at 6.1/7 on Ease of Use — clear evidence the redesign improved on what auditors and customers use today.
Critical Findings, Triaged Immediately
State-specific docs, password-protected files, downloadable summary.
Two CRITICAL issues (state-specific document and pay requests for WC audits) and four HIGH issues — including password-protected payroll files, missing FL signature surfacing, and the lack of a downloadable audit summary — were routed straight into the build via the RAS framework.
AI-Powered Verification Added
Bulk uploads, automatically categorized.
Introduced AI document scraping and categorization as a new audit feature — automatically organizing 'mass dump' uploads from experienced insureds and verifying data from payroll and tax documents, eliminating a major manual burden for both auditors and customers.
- A severity framework turns research into a roadmap.
Presenting 17 issues without prioritization would have paralyzed the team. Introducing the RAS gave stakeholders a clear action order — critical and high items moved immediately into the build, while lower-priority items became a structured backlog rather than forgotten feedback.
- The people who design the process and the people who suffer through it need separate research tracks.
Internal auditors had strong intuitions about what they needed — but no insight into how those requests felt on the other side. Running both tracks simultaneously meant the redesign served both audiences rather than optimizing for one at the expense of the other.
- Some recommendations require more than design approval.
The communications features — a significant part of the redesign — were sidelined as Day 2 items due to legal review and additional channel dependencies. Learning to present a phased roadmap, and to separate what can ship now from what needs more stakeholder alignment, is as important as the design itself.
- AI features belong in UX research workflows too.
Introducing AI document scraping wasn't just a technical feature — it was a UX decision that eliminated an entire category of user burden. The best AI features are invisible: users upload a document and the work is done. That's a design outcome, not just an engineering one.
/ Click any screen to expand