Twelve weeks to rebuild a self-serve platform

The problem

Our logged-in web self-serve experience was a third-party, white-label platform that was no longer going to be maintained. It underperformed the app on the measures that matter: more customers contacting us, fewer completing what they came to do.

There were two obvious options. Take ownership of code we didn't build and an experience that wasn't working for customers. Or spend a year in discovery and delivery cycles building something new. Neither was good enough, and the deadline didn't allow the second.

The third option

We took the app experience customers already trust and rebuilt it for the web, on the same APIs the app uses. That gave us three things at once: we own the code and know exactly what it calls and when, customers get a genuinely better experience, and every problem a customer can solve themselves is a call they never have to make.

How we built it

The team was deliberately small: three designers and three engineers, with AI working as a counterpart to both disciplines. That only worked because of a pipeline we'd spent the previous nine months building: bidirectional Figma-to-code workflows, component generation founded on the design system, and automated design-versus-build fidelity checks.

Designers and engineers stopped passing work between tools. A designer could take an idea to a working, coded, on-brand front end, and an engineer could pick it up exactly where it sat and make it real. Around 60% of the code was production-ready at hand-off.

Speed without drift

Speed is easy to buy with AI. Consistency isn't. I built a design system skill that checks generated UI against our tokens and snaps anything off-system back into line, so moving at AI pace didn't mean the product drifting off-brand. Accessibility to WCAG 2.2 AA was built in from the start, not audited in at the end.

What it delivered

A six-week proof of concept to directors, then six more weeks to customer-facing production. The rebuilt platform is mobile-first for the 60% of traffic that arrives on a phone, has full feature parity with the app, and opened the rewards experience to web customers for the first time.

Every component built for the journey went back into the design system, so the next product starts from the same head start.

What I'd do differently

The design system team built much of this alongside their own roadmap, and some of the system's own work waited while they did. Next time I'd agree up front how much of the team a delivery like this borrows, and for how long.

This wasn't how we delivered one project. It's how we deliver now.