Building a design system across global insurance products

Liberty Mutual and Safeco operated without a shared design foundation. The result was fragmented experiences that undermined brand consistency and eroded user trust at the exact touchpoints where an insurance company most needs to feel professional.

How fragmented? When I joined, even the true source of the Safeco logo and the correct brand font were unknown.

I partnered with one other designer to apply design thinking to our own colleagues' pain points, starting with the small set of base components everything else would rely on. The early insight that shaped everything: what we'd assumed were three separate systems, Agent, Consumer, and Employee, turned out to be one system waiting to be unified.

We built a site that brought the system to life: a sandbox, React components, documentation, accessibility guidelines, and ready-to-use code snippets.

Components weren't enough; the system needed rules for how it would grow and stay healthy. I aligned the team on a clear intake process and a shared definition of done, so every request was triaged consistently and every decision could be explained.

We documented two paths: an All Inclusive Path for complex components needing full design, research, and QA, and a Fast Lane for simple changes. A decision tree answered the question every design system team faces: is a new pattern actually necessary, or can we amend what exists? This governance layer is what let the system scale beyond the core team.

For 18 months, Safeco and Liberty Mutual ran parallel systems, meeting periodically but operating independently, until an audit revealed how much overlap existed and how much there was to gain by combining forces. We brought both teams into a workshop to weigh what to keep, retire, and rebuild. The team grew from two to five, and the result was a single unified system spanning agent-, employee-, and customer-facing products across U.S. and global markets.

In parallel, I joined a cross-functional team shaping updated brand guidelines for Safeco.com, pressure-testing concepts with key stakeholders before rolling the new brand into the merged system.

As the system matured, we brought in an accessibility expert to review components and built a simple checklist into handoff and QA. It kept the team educated and made "production-ready" mean something, not just a box checked at the end.

The system stopped being a project and became infrastructure other teams built on, and the results showed up in their numbers: a 97% NPS lift on Manage My Account and $161M in new business revenue on the Independent Agent Book Transfer tool. Quarterly surveys tracked the system's own health: component satisfaction, code adoption, and guidelines usage all climbed steadily.

The documentation site carries that same specificity all the way through: a welcome page and roadmap, full color and accessibility guidance, and component pages with usage rules, live code snippets, and a sign-off checklist before anything ships.

The system didn't just create consistency, it became infrastructure other teams built on, and the results showed up in their numbers.

Running parallel systems for 18 months was a good exercise. Both teams sharpened their thinking independently, and the workshop where we merged was stronger for it. But if I did it again, I'd push for consolidating the teams much earlier. The audit that triggered the merger didn't reveal new overlap, it revealed overlap that had existed the whole time, and we paid for it in duplicated components, parallel decisions, and patterns that later had to be reconciled.

The lesson I carry forward: when two teams are solving the same problem side by side, the question isn't whether to combine forces, it's how much it costs to wait.