Heat

Heat: A year of solving the 'cold start' issue

Context

Heat is a real-time social discovery app I co-founded and designed. The idea is simple: know the vibe of a place before you go, who's out, what the energy is, instead of guessing or scrolling old reviews. My co-founder Abdi built it, I led design and product.

The problem

Heat launched into the hardest version of cold start there is. No users, no check-ins, no reason for the map to feel alive. Day one had 40 opens. Day two dropped to 13. A few hundred downloads total in the first week, and almost nobody came back twice.

This case study covers a year of testing, watching where users dropped off, and rebuilding what didn't work. It's a summary of that process, not a single fix.

What I learned

The biggest shift was learning to trace a symptom back to its real cause before touching any UI. An empty chat thread looks like a dead app, but it's a seeding problem. A quiet "who's on" screen looks like low adoption, but it's a time window problem. I also got better at killing my own ideas fast once feedback proved them wrong, the original taxonomy, the real-time-only presence model, the untargeted push strategy all got scrapped and rebuilt.

Tools

Lovable for prototyping. Claude for brainstorming. Figma for the design system and final screens. Manual, in-person seeding in Gòtic doubled as research, watching what actually made people open the app twice. Moderated user testing.

Key problems and how I solved them

Cold start An empty map with no check-ins looks dead, and nobody opens a dead app twice. I wrote predictive, editorial signal instead of relying on live data alone. "Your crowd is usually here on Friday evenings" is honest and useful even with zero check-ins behind it.

The "who's on" screen only showed today's activity, which read as a ghost town at our size. I extended the window to 7 days so it reflects a real community.

Users told us directly that checking in for other people's benefit wasn't enough. I built a punch card system, three visits to the same venue unlocks a reward, so there's a personal reason to show up.

An empty chat thread signals abandonment more than no chat at all. We seed the first message myself so a thread is never completely dead.

Safety and visibility

Being shown on a map felt like a location risk, which made people reluctant to appear at all. I made presence contextual. You show up when you're actually out, and messaging only unlocks with mutual signal.

Retention

People opened the app once and had no reason to return the next day. Punch card progress gives an added reason to check-in to a specific venue when no one else is there. In app onboarding to explain the product on first open ships next release. We needed to promote and incentivize check-ins because they produce massive engagement.

Content and discovery

A handful of mediocre venues with guessed tags isn't worth recommending, and marketing before that's fixed wastes every first impression. I personally curate the top spots being shown on the map.

Tags like "Chill" and "Social" told users nothing Google couldn't already tell them. I rebuilt the taxonomy into blunt, votable, single-word tags split into Atmosphere and Crowd, surfacing what's specific to a venue instead of traits every venue shares.

Matchmaking and personalization

Onboarding explained the product but told me nothing about the user. I rebuilt it around eight taste taps that feed match percentage, notification targeting, and eventually a Discover Weekly for going out.

Predicted outcomes

Most of this is in build or just shipped. Onboarding ships next release, punch card is two weeks out. My 30-day targets: 10 venue partnerships live, 50+ punch cards redeemed, at least 3 venues independently reporting Heat-driven customers, and a measurable lift in return visits.

Want me to apply this same shape (context up front, short learnings, problem/solution list, honest predicted outcomes) to another case study next, maybe HeatChat since it's the closest sibling to this one?