Cleo is the world’s first AI financial assistant, with chat sitting at the centre of it. I led the redesign of our onboarding, leaning heavily on AI tools for prototyping, and ultimately ending with a new, fresh approach: onboarding via conversation.
Years of small optimisations had turned onboarding into a patchwork: lots of forms, inconsistent UI, and overly complex logic. Many utility screens sat there only because they always had, and removing them had never been a priority. The old UI also carried tech debt that was starting to block other work.
The flow had been tuned for so long that further tweaks were producing little gain. It also looked nothing like the product waiting on the other side. Onboarding and the app felt like two different products, and that gap didn’t meet our bar for quality.
Rethinking the most sensitive part of the app needed sponsorship. Since chat is the main way users interact with Cleo, it made sense to try a similar pattern for onboarding, so I built a rough conversational prototype that stakeholders could play with on their own phones. The point was to show the type of experience that was possible, not to focus on visual polish. It explored faster ways to enter information, a new priming step before bank connection, and the removal of screens that didn’t need to be there, all in one vertical conversational flow. The feedback was overwhelmingly positive, which was all I needed to continue.
Almost all the design work happened in code, not Figma. I shared a new build every couple of days; the team used it, gave feedback, and that shaped the next one. Feedback came much faster than with static screens, and people experienced the flow the way customers would. We also ran several usability sessions, watching participants go through the prototype. They helped us validate details that internally we weren’t sure about, such as how to show progress in a vertical flow and whether our idea for how users could go back was confusing.
Once we built a version we were comfortable with as an MVP, we launched as an A/B test against the old flow. Early results were mixed: fewer people connected a bank account, but a higher share of those who did went on to subscribe.
Analytics and recordings of real sessions pointed to some of the drop coming from a few usability problems, such as address input. These issues were not spotted in usability testing, but only surfaced once we launched to thousands of customers. Once we fixed them, bank connection and subscriptions improved.
Replacing a flow that had been optimised for years and coming out ahead with the updated experience gave Cleo a new baseline. Optimisation efforts started again, this time on a flow that met our quality bar.
This is a short version. Full case study available on request.