From 0.2%
to 10% of MTUs.
I bet that recurring investing wasn’t a feature problem, it was a narrative problem. Education-first, incentive-second, ROI-positive by construction.
Of monthly transacting users on a recurring buy, 60 days post-launch. Up from 0.2% baseline. 50× lift in adoption rate.
From a 600-user baseline. 55× in absolute users.
The North-Star at Gemini was Monthly Transacting Users, and the team had been pushing on the top of the funnel for a year. The numbers crept up. The shape of the line stayed the same.
I’d been watching a different number. Of all monthly transacting users, only 0.2% had a recurring buy active. The product existed: well-built, well-priced, and almost no one was using it. That was the gap.
It’s a habit, and habits don’t respond to ads.”
I made the call to treat Recurring Buy not as a feature but as a multi-week growth experience. Education first, incentive second.
The hypothesis, written down: replace the conversion-first surface with an education-first one, back it with an ROI-positive incentive, and we move from 0.2% to 10% of MTUs within two quarters. A 50× bet.
0.2% adoption isn’t one problem, it’s three. Before designing the experience, I asked which gap was binding: did users know the feature existed (awareness)? Could they find it when they wanted it (discoverability)? Could they actually commit (friction)? All three were broken. We attacked all three.
Awareness
Problem. Most users didn’t know recurring buy existed. The feature lived inside the trade flow.
Fix. Dedicated in-app surfaces and an education email arc explaining what recurring buy was and how to use it. Framed around outcomes: what compounding looks like, not what the feature does.
Discoverability
Problem. The entry point was buried two taps deep, behind a default state that always landed on one-time buys.
Fix. Promoted recurring buy to a dedicated card on the home screen, the first surface every user sees on app open. No more hunting.
Friction
Problem. Setup demanded an amount and asset with no preview of outcomes. Most users didn’t see the value.
Fix. A calculator built into the commit flow: pick an amount and asset, see what the same plan would have returned over 5 years on real historical prices. Education at the moment of decision, easy commit on the other side.
Education
Slider-driven simulator. Pick an amount, pick an asset, see what the same plan would have returned over 5 years on real historical prices. CTA pressure deliberately undersold.
Reward
If user hadn’t committed, they entered the reward track: payouts at 3-, 6-, and 12-month milestones. Capped at $150. Tuned with finance until provably ROI+.
The simulator was the soul of the thing. We tuned its physics and copy across 11 internal builds before it shipped. The slider had to feel like a tool you’d use to think, not a step in a funnel.
The reward design was the discipline of the thing. Every payout band was modeled against retention curves and per-user revenue share. We refused to ship anything that wasn’t ROI-positive on its own.
Two wins, distinct on attribution. Fully attributable: recurring-buy users moved from 600 to 33,000, a 55× lift in absolute users, and adoption rate moved from 0.2% to 10% of MTUs, a 50× lift in rate. Both are direct outputs of the experience and the experimentation that shaped it.
Contributed to: the platform’s MTU base grew from ~300K to ~330K over the same window, and recurring buy was one retention lever among several pulling on it. The compounding nature of the behavior, every adopter transacting automatically thereafter, made it a structural contributor rather than a one-time spike.
Appendix A · ROI Modeling and Outcomes →
The band-level NPV math, the payout structure, the conversion lifts, and the one thing we got wrong.