Getting Paid Only When Users Lose Weight

Engineering under results-based pricing

NanoRhino is an AI nutrition coach that works entirely over text messages. No subscription, no upfront fee. The price is $10 for every pound lost, capped at $500. If users don't lose weight, we make nothing.

This started as a trust decision, not a business-model innovation. But pricing on results turned out to be the highest-leverage engineering decision we've made — because it quietly rewired what "important" means everywhere else in the system. This post is about what changes when your revenue is contingent on your users' outcomes.

Why price on results

Our core users are women in their 50s, 60s and 70s. By the time they text us, most have been through the full cycle at least once: the app with the streaks they stopped opening, the subscription that kept charging anyway, the program that was enthusiastic for exactly two weeks. The most common early message we get isn't about food. It's some version of "how do I know this isn't another scam."

A subscription asks the user to carry all of the risk: you pay whether or not it works. Results pricing moves the risk to us. For an audience whose default setting is distrust, it's the strongest signal we could send — and unlike copy on a landing page, it can't be faked, because it's enforced by our own invoice.

The billing constitution

"Pay per pound" immediately raises fair questions. What about water weight? What if I regain? Whose scale?

We answered by publishing our billing rules and treating them as a constitution: public, and not quietly changeable. Two rules do most of the work.

The two-week rule. A pound counts only after it has stayed off for two weeks. Daily weight is noise — water, sodium, sleep, hormones. Billing on noise would be theft with extra steps, and refund fights forever. So a lost pound sits in a provisional state until it survives two weeks of follow-up weigh-ins; only then is it billable.

The same pound is never billed twice. Lose a pound, regain it, lose it again — you pay once, ever. Weight is cyclical for most humans. A pricing model that profits from the cycle would put our incentives at war with our users' reality.

Internally, this means a pound is not a number — it's a small state machine (provisional → confirmed → billed), backed by a permanent ledger of every pound ever billed, so re-lost weight is recognized and excluded. The $500 cap (fifty pounds) closes the constitution: past that, coaching continues and billing stops.

What results pricing does to your engineering

1. Revenue becomes a measurement problem. The core of the company is not meal advice; it's the honest measurement of a noisy physical quantity through a low-bandwidth channel. Weigh-ins are the scarcest resource in the system — you can neither coach nor bill what you can't measure. A surprising share of our roadmap reduces to one question: how do we get more true weight data with less friction?

2. Retention engineering is revenue engineering, literally. Under subscriptions, churn is a slow leak you can paper over with acquisition. Under results pricing, a user who quits in week one is a pure write-off: acquisition paid, coaching delivered, nothing billed. Every "boring" ops metric acquires a dollar sign. Reply latency has a budget — nobody waits for a spinner inside a text thread. Reminder tone is load-bearing: one tone-deaf nudge at the wrong moment can end a relationship that would have produced months of results. We spend real engineering time deciding when not to send messages.

3. Every token is spent against uncertain revenue. Cost discipline stops being a finance concern and becomes a daily engineering habit. A transcript-handling bug once quietly burned about $400 of prompt-cache spend before we caught it; scheduled jobs for long-inactive users were burning compute while producing nothing. Neither incident is exotic. What changes under results pricing is that you audit for them constantly, because the margin they eat was never guaranteed in the first place.

4. Honesty must be designed, not enforced. Weights are self-reported. We considered requiring smart scales and rejected it — our users hate mandatory hardware, and surveillance is the wrong relationship with someone you're coaching. Instead, the pricing rules do quiet data-quality work: because a regained pound is never re-billed, there is no financial motive to hide a bad week from us. The rule that protects users also produces more honest data. That's my favorite property of the whole design.

5. The perverse incentive, said out loud. $10 per pound means we earn more when you lose more, faster. Unmanaged, that is a genuinely dangerous incentive for a coaching system. We manage it structurally: pace targets are capped around one percent of body weight per week, and faster plans are simply never generated; protein floors protect lean mass; a governance layer pauses coaching and points to a clinician when messages raise medical flags; we refuse under-18 users at the front door; and the $500 cap removes any incentive to chase extreme loss. Is the alignment now perfect? No — a results-priced coach will always have more appetite for aggressive plans than a subscription-priced one, and I'd rather admit that in public than pretend the incentive doesn't exist. Publishing the constitution is part of how we stay honest: the rules are checkable from outside.

What I'd tell other AI builders

If your product's pitch includes the word "outcomes," run the thought experiment: what breaks if you charge on them, even partially?

For us the answer was: nothing broke, but everything reordered. Measurement moved to the center. Retention became revenue. Cost hygiene became reflex. Safety guardrails became commercial necessities instead of compliance chores. And the pitch got shorter, because "we only get paid if it works" survives contact with a skeptical audience better than any benchmark.

There are real preconditions — an outcome you can measure honestly, a rule set you're willing to publish, and enough conviction in the product to eat the downside. But if you wouldn't dare price on your outcomes at all, that's worth sitting with too.

NanoRhino today is hundreds of real people texting meal photos to a phone number and, pound by confirmed pound, proving the model out. The billing rules are public at nanorhino.com/pricing; the coach lives at (915) 277-7888.

---

Written by Xiaomeng Li, Co-founder & CTO of NanoRhino. I'm an AI engineer, not a clinician — nothing here is medical advice. The deficit-loop coaching engine behind NanoRhino is U.S. patent-pending.

Built with care by Link Heart Limited in Houston, Texas.