Christine YimDigital Visual Designer

New York

A laptop showing the Berkley agent dashboard, with a welcome panel, a "Request a Quote" section and a list of resources.

Berkley Technology Services

I ran usability tests on the quote flow in Berkley’s new agent portal before it was built.

  • UX case study
  • 2023—2024
  • Berkley Technology Services
Role
UX research: prototyping, note-taking and synthesis
Duration
2 months
Tools
Figma, Microsoft Suite, Miro

The quote flow in Berkley’s new agent portal had been designed by an outside agency, but nobody had tested it.

Berkley Beyond is an online portal that puts policy servicing and payments in one place for brokers and policyholders. The quote flow is where an agent enters a policyholder’s details and gets an estimated premium. Before developers built it, the lead UX designer and I spent two months testing it and working out what to fix first.

Deliverables

  • Clickable prototype
  • Usability tests
  • Note-taking template
  • Affinity map
  • Prioritized findings

Results

A ranked list of what to fix, ready for the next design round.

4
Subject-matter experts tested the flow
60—90
Minutes per session
13
Groups of findings, one for each part of the flow

The problem

Designed, but never tested.

The screens for General Liability and Commercial Property quotes existed, but no one had watched a person use them. We wanted first impressions, the pain points, and a ranked list of what to fix before release.

  • Prototyping limits

    Figma couldn’t reproduce every interaction the built product would have.

  • Recruiting

    Agents were hard to reach, so subject-matter experts stood in for them.

  • Learning curve

    I had to learn how both lines of business are quoted before I could test them.

The prototype

I built the whole quote flow as a clickable prototype.

First I learned how agents quote, and went through every screen with the product managers and the lead UX designer. Then I built the prototype in Figma with realistic dummy data.

Building it also showed me design discrepancies between screens, which I logged to come back to.

Two screens from the prototype. Full screens are available on request.

The prototype had limits, so I prototyped components to ease the confusion.

In the clickable prototype, some things only looked like they worked. That confused the subject-matter experts we tested with. To ease it, I learned to prototype at the component level. This dropdown is one example.

  • Key takeaway

    Even so, a prototype with limits isn’t low-quality work. The goal is to find the right amount of prototyping for people to finish the main tasks.

Dropdown componentTry it

Protection Class

One of the component-level prototypes, rebuilt here so you can try it.

Behind the scenes

The Figma file for the component prototypes: each dropdown laid out as a row of states, joined by a web of purple prototype connections.

How it was built in Figma. Each purple line links one state of a dropdown to the next.

Testing

My notes showed exactly where people struggled.

Agents were hard to recruit, so we tested with four subject-matter experts who work closely with them. The lead UX designer ran each session over a recorded Teams call, following a script we wrote together. I took the notes.

  • Every pause on record

    I built an Excel template with timestamps. The elapsed-time column showed where people slowed down or got stuck.

  • Coded as I wrote

    Each note got a code from a dropdown: process, interruption, quote, issue, positive finding, recommendation or error.

Spreadsheet of session notes with columns for note, code, time and elapsed time; the note text is blurred.

My note-taking template. The content is blurred.

Findings

The affinity map showed the team what to fix first.

I put every note into Miro as one long list per session, then grouped them by screen with the lead UX designer. The biggest groups were the screens that confused people most, and the map showed inconsistencies between the two quote flows for the product managers to settle.

  • Optional fields raised questions

    People weren’t sure why some fields were optional, and expected them to be hidden when they didn’t apply.

  • The address questions felt out of order

    One tester expected the mailing address first.

  • An edit icon was unclear

    A tester couldn’t tell what it would change.

  • What worked

    The side navigation, autosave, and clearly marked optional fields.

Before

A Miro board with hundreds of sticky notes in long columns, one column per topic, too many to read.

Every note, in long lists.

After

The same notes, sorted into 13 groups. Click to enlarge and read them.

Outcome

The team used the findings to redesign the flow.

After the research, the project was put on hold. The team picked it back up after I left, and my supervisor later told me they used the findings to improve and redesign the flow.

  • More data than expected

    The sessions gave the team enough to start a design round.

  • Good feedback

    Product managers and subject-matter experts praised the depth of the research.

  • A method shared

    I presented component-level prototyping to the UX team and recommended adopting it.

Looking back

What I’d do differently.

  • Use AI for synthesis

    Try AI tools to make the data synthesis faster.

  • Recruit working agents

    Hear directly from the people who will use the quote flow.

  • Write it up

    Turn the Miro board into a structured document that’s easier to read.