Information Architecture Case Study

17 Rounds of Information Architecture Testing to Redesign GEICO’s Customer Portal Navigation

GEICO was preparing to redesign the navigation of its customer portal, where policyholders log in to pay bills, print ID cards and change their coverage. Marketade, a UX research firm, ran 3 rounds of card sorting, 10 rounds of tree testing and 4 rounds of prototype testing with about 120 participants, redesigning the navigation between rounds and cutting the main menu from 10 tabs to 6.

Person holding a payment card while using a laptop

Last updated: November 20, 2026

Goals

GEICO is one of the largest auto insurers in the US and sells directly to consumers, so its website carries much of its customer service. Policyholders log in to its customer portal to make payments, print insurance ID cards, update their address, add a car or a driver and change their coverage.

GEICO’s digital service team was preparing to restructure and redesign the portal’s navigation. The main menu had grown to 10 tabs, and some links appeared under several of them: the ID card link sat under 3. The team wanted a new, human-centered structure that was tested with customers before it was designed and built.

The research set out to:

  • Learn how customers think about the core service tasks and ways to organize them
  • Generate structures and labels that lead people to the right place for their most common tasks
  • Find the best structure through rapid testing and iterative design
  • Test the new navigation within a full page design, alongside other competing links

‍

Approach

We ran the project as information architecture (IA) research, the part of UX research that tests how a site’s content is organized and labeled. The work ran in 3 phases, each with its own method, for a total of 17 rounds of testing and about 120 participants:

PhaseRoundsEst. participantsMain goal
Card sorting330Learn how customers group service tasks and what they call each group
Tree testing1060Test and refine the structure and labels on their own, before any design
Prototype testing430Test the navigation on a full page, alongside other competing links
Total17120

The work followed 5 steps: discovery and analytics, then the 3 phases, then final analysis and reporting.

Unlike most UX research projects, this one had us designing as well as testing. Before each round we recruited new participants, and after it we analyzed the results and revised the navigation ourselves for the next round.

1. Discovery and analytics

Discovery started with GEICO’s site map of the customer portal, focused on its top 50 destination pages, and a list of the top customer tasks. We paired these with GEICO’s analytics on how customers actually navigated: which pages drew the most visits, which links on the portal homepage people clicked and how often customers did each task. We also reviewed how 3 other large insurers organized their customer portal navigation.

The analytics shaped the research from the start. The most-visited pages became the cards for the card sort, and the most common tasks became the tree test tasks, weighted by how often customers did them. Between rounds, the same data settled close calls, such as whether a page with little traffic earned a place in a menu, or whether a common task deserved a shortcut on the homepage.

2. Card sorting in 3 rounds

Card sorting is a UX research method in which people group a site’s content or functionality in whatever way makes sense to them, then name each group. It shows how customers think about the content and helps UX teams generate IA structures and labels that align with customers’ mental models.

We ran 2 trial rounds of about 5 participants each, then a main round with about 20 participants, using Optimal’s online card sorting tool. Each participant sorted about 50 randomized cards, one per destination page, and thought aloud as they grouped the cards and labeled their groups.

For each round we recruited US adults with car insurance, ages 30 to 65 and older, across a range of incomes.

Before our first round, we wrote the card labels following card sorting best practices. Many page names shared words like “policy”, “payment” or “coverage”, and people tend to group cards by matching words rather than by meaning. Using synonyms instead made participants sort by what each page was for. After each trial round, we reviewed every session and fixed unclear cards and instructions.

An example of a card sort finding: people used the word “Account” more broadly than the site’s designers, grouping a wide range of pages under it. Together with the analytics and discussions with GEICO’s team, the card sort helped us produce the first version of a new structure.

3. Tree testing in 10 rounds

Tree testing is the fastest, most effective way to assess a site’s information architecture and pinpoint its problems. We ask representative users to find information or start tasks using a simple, clickable “tree” of the site’s navigation, with no visual design, and we record each click. The results show which tasks are easy for customers to find and where they struggle, and which category structures and labels are likely to cause confusion.

We ran 10 rounds of tree tests with more than 60 participants, 5 to 10 per round, using Optimal’s online tree testing tool. One round tested GEICO’s existing navigation as a control, so every new version had a baseline to beat.

Each round gave participants 8 to 15 tasks drawn from customers’ most common tasks, such as checking a claim’s status or changing how often they’re billed. As with the cards, the task wording avoided the words used in the menus. Participants thought aloud, so we could see where they hesitated as well as where they clicked.

After each round, we compared success rates with what we observed in the sessions, logged each finding with a priority and a recommended change, and revised the tree. Every change went into the next version with a written hypothesis, such as a clearer label drawing more correct clicks, and the next round marked it validated, invalidated or inconclusive. Over 2 rounds of relabeling, the success rate on one key payment task rose from 22% to 80%. Some rounds tested 2 or 3 variations of one version side by side, and we tuned the tests as well, rewording any task that pointed people to a label. By the final round, we recommended 5 main tabs, down from 10.

4. Prototype testing in 4 rounds

Prototype testing put the new navigation on the full page. GEICO’s team wanted to know whether customers could still find what they needed with every other link on the portal homepage competing for attention: shortcuts, a resource center and the page header. GEICO’s designers built a clickable version of the homepage, and we built further variations as the rounds went on. GEICO’s team also chose to add a 6th main tab, and the prototypes tested it alongside the other 5.

We ran 4 rounds of remote usability testing on these prototypes, with 5 to 10 participants per round and about 30 in all, each thinking aloud as they worked. GEICO ranked about 40 customer tasks by priority, and we wrote each one to avoid the labels it was testing.

GEICO’s team watched the sessions themselves and took part in each decision, so findings reached the next prototype within days. Across the rounds the team tested at least 5 prototype versions.

5. Final analysis and reporting

Analysis ran through the whole project in a shared findings log. Each entry recorded the round, how many participants ran into the problem, its priority, the recommended change and whether the change made it into the next version. The log grew to about 60 findings.

After the tree tests, we wrote a summary report covering the card sort and tree test findings, the recommended navigation and the reasoning behind each tab. For prototype testing, the findings log served as the report, since GEICO’s team had followed every round.

After the 4th round of prototype testing, we recommended stopping. The issues that would be expensive to fix after launch had been caught. The rest were minor, and each new tweak risked creating new problems. We advised testing those once the navigation was live, when the team could see a customer’s whole path rather than a single click. Any team running iterative tests can use the same rule: stop when the remaining problems are cheaper to fix after launch than to chase in another round.

‍

Outputs

By the end of the project, GEICO’s team had:

  • A main navigation of 6 tabs, down from 10, with the links under each tab and a set of header links
  • About 60 findings logged across 17 rounds, each with a priority and a recommended change
  • Clickable prototypes of the portal homepage, refined over 4 rounds of testing
  • A summary report of the card sort and tree test findings
  • Guidance for future changes, such as adding simple “pathway” pages rather than more menu links
  • A short list of minor issues to test after launch

The findings log kept the work focused. Each round’s findings were ranked, so the team fixed the problems that blocked common tasks first and left minor ones for later. Three examples show the kind of insight the rounds produced.

More tabs didn’t make tasks easier. We tested giving some task areas their own tabs. Success rates didn’t clearly improve, and the sessions showed participants hesitating over the extra choice.

Shortcuts competed with the main menu. On the full page, many participants clicked a box of quick links before trying the main navigation, even when the menu was the right place to go. We trimmed the shortcuts to the most common tasks.

A label promised more than its page delivered. A heavily used link sounded like it led to the customer’s own policy details, but the page held general tips. Participants clicked it for coverage questions and didn’t find what they wanted. We recommended a label that described the tips, accepting an extra click on a less common task to stop wrong clicks on common ones.

All 17 rounds
RoundPhaseEst. participantsWhat we testedNotable learning or change
1Card sorting (trial)5The first set of about 50 cardsReworded cards and simplified the instructions
2Card sorting (trial)5Revised cards and instructionsSmall findings and tweaks
3Card sorting20The final card setCustomers’ own groupings shaped the first new structure
4Tree testing (trial)5 to 10Tasks and instructionsReworded tasks that confused people
5Tree testing5 to 10GEICO’s existing navigation, as a controlA baseline success rate for every task
6Tree testing5 to 10The first new structureFound the labels that drew the most wrong clicks
7Tree testing5 to 10Revised labelsThe success rate on a key payment task rose from 22% to 50%
8Tree testing5 to 10Labels revised againThe success rate on the same task rose from 50% to 80%
9Tree testing5 to 10A variation of round 8’s structureSmall findings and tweaks
10Tree testing5 to 10A second variation of round 8’s structureSmall findings and tweaks
11Tree testing5 to 10A structure with one more main tabNo clear change in success rates
12Tree testing5 to 10A revised structureSmall findings and tweaks
13Tree testing5 to 10The final treeWe recommended 5 main tabs, down from 10
14Prototype testing10GEICO’s clickable homepage, in 2 versionsLinks elsewhere on the page competed with the main menu
15Prototype testing10A revised prototypeSeveral changes from round 14 helped
16Prototype testing5A revised prototype, with focused tasksThe remaining issues were minor
17Prototype testing10Follow-up tasks on the final prototypeSmall findings and tweaks
What 17 rounds showed

Iterative testing settled questions that a single test couldn’t. Some labels needed several rounds of rewording before participants reliably told them apart, and some structural ideas were tested more than once before the team settled on a simpler menu. That is the case for iterative testing. Each round was small and quick, so the team could try an idea, learn from customers and adjust within days. Problems that would have been expensive to fix after launch were found and fixed while the navigation was still easy to change.

‍

About the Project

  • Industry: Auto insurance
  • Platform: Customer portal on the web
  • Audience: Auto insurance policyholders
  • Methods: Card sorting, iterative tree testing and prototype navigation testing
  • Length: About 3 months
  • Stakeholder teams: Digital customer service, design and analytics
  • Company size: More than 25,000 employees

‍

More Case Studies

Marketade led card sorting and first-click testing to develop and refine the IA for a large UN agency serving 2,000+ employees.

Marketade conducted tree testing to overhaul a B2B machinery company’s site structure and increase self-service, web sales, and sales team productivity.