How Fitness AI lifted revenue per user with Botsi.

“After years of A/B tests & various marketing campaigns, we knew different users would pay differently depending on their intent. Botsi processes all that information and prices each user individually, generating real & consistent ARPU uplift.”
Adi Racu, CTO · Fitness AI
Fitness AI Consumer subscription app  ·  Paywall & price optimization
Summary
Problem

One static paywall showed every user the same price. Testing was slow and one-and-done, and the highest price always won, so pricing crept up until it started costing subscribers.

Strategy

Botsi trained a causal uplift model in learning mode and went live within weeks. It matches each user to the right paywall, package, and price.

Result

Botsi beat the control group and lifted revenue per user, with the lift climbing as the model learned. At full rollout, it adds incremental revenue every year.

About Fitness AI
Fitness AI
Consumer subscription app

Fitness AI is a popular, highly rated iOS fitness app that creates AI-guided workouts and personalized training plans. It’s subscription-led, and the team had already optimized its price points and packaging through years of A/B testing.

01 The challenge

The highest price always won, but pricing stayed frozen.

Fitness AI had already optimized pricing through years of A/B testing, and the pattern always held: the highest price won, but only up to a point.

02 The solution

The right offer for every user that maximizes LTV.

Fitness AI’s paywalls ran several packages across monthly, quarterly, and annual terms. Botsi trained a causal uplift model to give each user the package worth the most over their lifetime, not just at checkout.

  1. 01

    The right plan for each user

    One price for everyone hits a ceiling, so each user gets the plan worth the most over their lifetime.

  2. 02

    Causal uplift, not correlation

    It counts only the users an offer actually converts, so the revenue it drives is genuinely new.

  3. 03

    Learns from real revenue

    It trains on renewals and cancellations, not just the sale, so it sharpens with every cohort.

Exploit to earn now, explore to keep improving
every outcome retrains and sharpens the model, continuouslymostsomeCausal uplift modelreads each user's signalsEXPLOIT · MOSTServe the best offerthe highest-uplift price, earns nowEXPLORE · SOMETry other pricesrandomized, gathers fresh signalOutcomesconversions, revenue, churn
Most users see the best offer, so the model earns now (exploit); a smaller slice sees other prices to keep learning (explore), and every outcome retrains it.
Adi Racu
“Botsi was one of the easiest tools we’ve ever had to implement, and the underlying ML infrastructure is rock solid.”Adi Racu, CTO · Fitness AI
03 Results

The lift kept climbing as the model learned.

ARPU lift over the baseline, improving as the model trains
Botsi ARPU lift vs baseline Training period
TRAINING PERIODARPU LIFTBaselineas the model trains on more data →
The model trains first, then serves dynamic offers. As it learns from more data and new signals, the lift over the baseline keeps improving. Directional trajectory, not exact weekly values.

Measured against an evenly split live control, Botsi lifted ARPU well above the baseline, and the lift kept climbing as the model trained on more data and new signals came online. Because a matched group always saw the baseline, that gap is genuinely incremental, and at full rollout it adds to revenue every year.

Fast to launch, sharper every week.

Botsi trained a model specific to Fitness AI, not a generic one, and launched it in learning mode. It went live on a lean, device-level feature set, so it needed only a small volume of conversions for its first reliable read. Custom signals layer in from there.

Phase 1

Learning mode

Collects signal across live traffic before making a single pricing call.

Go-live

Model goes live

Launches on device-level signals: OS, device, country, and timing.

Measured

Driving uplift

Serves the best offer per user, measured against the live control.

Phase 2

Retrain, deeper

Retrains on onboarding and attribution signals for more accurate calls.

04 The signals

The model reads each user in rich detail.

A price is only as good as what the model knows about the user. Botsi reads a rich set of signals on each one, from how they reach the paywall to their device, their market, the fitness goals they set, and how they found the app. It is a fuller picture than any single price test could see.

  1. 01

    Timing

    Time of dayDay of weekPaywall moment
  2. 02

    Device & platform

    Device modelOperating systemDevice tier
  3. 03

    Market economics

    CountryGDPMobile adoptionPurchasing powermore
  4. 04

    Fitness profile

    GoalsMotivationFitness levelTarget weightEquipmentmore
  5. 05

    Attribution & acquisition

    Attribution sourceAcquisition channelPaid vs organic

No personal data, by design. The model runs on anonymous signals and never captures PII, so every user stays fully anonymous for safety and compliance.

Adi Racu
“Individual pricing is just the start. We are continuously working with the Botsi team on using the same underlying technology across multiple areas of the business, not just paywalls.”
Adi Racu, CTO · Fitness AI

Meet every user at their price.

See what per-user pricing can add to your app’s ARPU, proven against a live control before you scale it.