Streaks began with a hypothesis about milestones: if we celebrated key days (3, 7, 10), users would be more likely to push toward the next one, driving our core metric of week-over-week food logging. For the first test, I worked with our business intelligence team to source social proof claims like "You're in the top 50% of users!". Users came back to log 3% more food and 1% more days in week 2.

Frequency | Learning: More celebrations = Good

That result opened a bigger question: how far could frequency go? We kept testing more celebration days until we were celebrating every single day from day 2–10. That experiment reached stat-sig +1.5% average app open days and +2% average food logging days compared to control.

Timing | Learning: Drive the Action → Reward cycle

Alongside frequency, we tested timing: celebrating a user the instant they logged food and incremented their streak, versus waiting until they returned to the Today screen. The in-the-moment celebration won, even though it interrupts the user mid-task. The suppressed celebration was stat-sig negative on average app open days. The principle this surfaced: celebrations are welcome distractions as long as they feel meaningful, and reinforce a powerful Action → Reward cycle.

Copy | Learning: Lean on social proof and users’ identity

With frequency and timing solved, I turned to copy. My hypothesis was that novel, varied copy each day would outperform repetition. But, it didn't. leaderboard-style social proof still won with a small neutral-to-negative result on the variant. The takeaway reframed how I thought about streak copy going forward: users respond less to novelty than to feeling part of something larger than themselves.

That insight opened a new question: What about Day 1, before leaderboards become relevant? I tested a day-1 streak for returning users leaning on identity: "Mary is back!" It won with a stat-sig +2% average app open days, confirming that identity-based framing can be as powerful as social proof.