01 · Two surfaces
Two users, two surfaces, one system underneath
Product Gears started from a conversation my co-founder and I kept having: customers send feedback, product teams decide what to build, and the customer never finds out what happened to their idea. We saw that silence as a retention problem, not just an internal alignment one.
The product had two users with opposite jobs. Product managers needed to sort, merge and prioritize requests, and customers needed to suggest, vote and follow what happened next. Several competitors put both in one interface, and both sides ended up confused. I decided from the start to design two separate surfaces, an admin side and a customer-facing side, on top of one shared data layer.
02 · Interviews
The interviews changed two priorities
I interviewed product managers and designers recruited from WhatsApp and Telegram communities, alongside desk research and a heuristic review of competitors. Two patterns came up that we had not expected.
PMs dreaded duplicates. Several participants raised it without being asked, so merging duplicate suggestions moved from a later release into the MVP. Silence broke trust: customers who never heard back stopped sending feedback, so notifications to customers moved up as well, to close the loop the product existed to close. Integrations such as Zapier, GitHub and Apple login waited, because the core use case worked without them.
Usability testing with two parallel scripts found no critical navigation failures in the core loop. The MVP launched, and Agenda Edu, a Brazilian edtech company, used it both with its internal teams and with its customers. The product is no longer live.
03 · What I'd change
What I'd do differently
I would test the revenue model much earlier. The interviews confirmed the problem but never established whether small and mid-sized product teams would pay for a standalone tool. I also over-scoped the design system for an MVP: a lightweight token set would have done the job, and a full component library slowed us down. After launch, we watched session recordings without hypotheses to test, which produced data without direction. Next time I would decide what each number was supposed to answer before collecting it.