Powerverse
Self‑serve onboarding and an AI charging assistant for UK EV owners.
Product designer in a 3–5 designer team; sole owner of the flexibility feature, from idea to release.
Smart EV charging · UK · B2B2C · 2024–now
Under 5 min to onboard, down from about a day
About 40% of users act on AI recommendations
35% lower charging cost and 17% less CO₂ per user
The problem
Powerverse joined the UK grid flexibility network: EV owners can earn by charging when the grid needs it. But every new user was registered by the support team, and the white‑label partner portal had no view of flexibility at all. The business needed users to join on their own and partners to see what flexibility was doing for their customers.
Key decisions
1. Self‑serve onboarding to the aha moment
I took a one‑line idea – “add flexibility to the app and the partner workspace” – and expanded it: interviewed the customer ops team, mapped the problems, defined the jobs users hire flexibility for. I explored 3–5 onboarding directions, then tested the strongest as a live coded prototype with real customer ops users before engineering started.
2. An AI assistant that advises, never decides
The assistant suggests cheaper and greener charging through recommendation cards and notifications, and can plan charging sessions – but the user always makes the final call. Each recommendation explains why it should work and shows a forecast. When a recommendation turns out wrong or data is missing, the assistant says so, pauses to re‑analyse and comes back with a better answer instead of guessing.
3. Flexibility dashboards for partners
Customer ops teams at white‑label partners track flexibility performance and key asset metrics across 20,000+ chargers. Instead of scoring every user, the dashboard highlights the ones who don’t follow recommendations, so support knows whom to contact and what to suggest.
Where AI helped. Where I overruled it
Claude got me up to speed on the UK energy flexibility market in 2 days instead of a week and turned ops interviews into JTBD. Figma Make for exploring directions, Claude Code for coded prototypes, Claude Code + Figma MCP to restructure the design system into semantic tokens and docs.
Where I overruled it. AI proposed a 5‑point score for every user’s charging and flexibility behaviour. It gave the customer ops team nothing to act on, so I cut it – the dashboard now flags only users who don’t follow recommendations, so support can reach out with a next step.
Result
- Time to onboard: about 1 day → under 5 minutes
- Onboarding support requests: down about 90% (fewer than 10% of users now need support, per customer ops)
- Partner dashboards covering 20,000+ chargers
- About 40% of users act on the assistant’s recommendations
- 35% lower charging cost and 17% less CO₂ per user
- About 25% higher retention among assistant users than non‑users
What I’d do next
Study the 60% of users who don’t act on recommendations – that gap shows where trust is still missing.