How Sleepiest lifted subscriber ARPU by 20% with Botsi.
A single static paywall was leaving revenue on the table. Botsi matched the right offer to each user and proved the lift against a live holdout.
Sleepiest Consumer subscription app · Paywall & price optimization · 4 min readOne 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.
Botsi launched a causal, Bayesian uplift model in learning mode and went live in about three weeks, matching the right paywall and offer to each user.
Measured against a live holdout: +20% subscriber ARPU and an estimated +$185K incremental revenue.
Sleepiest is a sleep app with a library of over 2,990 sleep sounds, stories and guided meditations to help people fall asleep and rest better. Named an Apple “App of the Day,” it has grown to more than 7 million downloads and a 4.59-star rating across 53,800+ reviews. Subscriptions are the core of the business, with roughly 80% of subscribers on annual plans and pricing localized by country.
| Before Botsi | With Botsi |
|---|---|
| One paywall, same price for every user worldwide | Package & offer matched to each user across 4 dynamic variants |
| Price testing was "one and done", last real test a year earlier | Offers continually optimized by a causal, Bayesian model |
| Highest price always won, so pricing was capped by gut feel | Higher willingness to pay captured without losing price-sensitive users |
| No clean read on what a pricing change actually earned | +20% subscriber ARPU vs. a live holdout |
Simple A/B testing led to higher prices, but fewer subscribers and conversions.
Like most apps, Sleepiest priced by intuition and infrequent A/B tests. The tests were slow and hard to repeat, so the team would run one, pick a winner, and leave it for months. The results kept pointing the same way: the top price always won, yet they could only push so far before it felt uncomfortable to charge more.
But raising the price for everyone has a hidden cost. It lifts revenue per user while quietly pricing out the price-sensitive ones, and a smaller subscriber base means fewer renewals and less organic growth over time. A single static paywall couldn’t tell a high-intent user from a price-sensitive one, so Sleepiest was stuck choosing between revenue and reach.
An AI/ML model that understands the Sleepiest user base.
Botsi didn’t just test prices. It learned which offer actually causes each user to convert, and kept exploring intelligently whenever it wasn’t sure.
- 01
Causal uplift from day one
The model predicts who buys because of a given offer, not just who buys. It targets incremental revenue, not sales that would have happened anyway.
- 02
Bayesian confidence
A Bayesian uplift model estimates the probability that each prediction is right, so every decision carries a built-in confidence level.
- 03
Smart sampling when unsure
When confidence is low, the model switches to Thompson sampling. It explores offers intelligently instead of guessing, then learns from the result.
- 04
Four paywalls, many offers
Sleepiest launched with four paywalls, each with distinct packages and offers, giving the model a real menu to match to each user.
Live in weeks, smarter every week after.
Botsi launched in learning mode, went live on a lean feature set for speed, then retrained on richer signals for sharper predictions.
Learning mode
Runs in learning mode, gathering data across real traffic.
Model goes live
Launched on Phase 1 device signals: OS, device, language, country.
Driving uplift
Serves dynamic offers, measured against the holdout.
Retrain, richer
Retrain on custom in-app signals for more accurate predictions.
A holdout group lets Sleepiest be confident in the true incremental revenue uplift.
A 10% holdout of traffic always sees Sleepiest’s original baseline paywall, while the other 90% goes to Botsi. Comparing the two gives a true incremental read on the revenue Botsi drives.
+20% subscriber ARPU, and +$185K in incremental revenue.
Against the live holdout, users on Botsi delivered 20% higher ARPU than users on the original paywall: an estimated $185K in incremental revenuethat wouldn’t have existed otherwise.
“The Botsi team has been a pleasure to work with and we’re very excited for the future of where Botsi can take us.”
Show the right price & paywall to each user.
Apps like Sleepiest use Botsi AI/ML models to show the right offer to every user to grow revenue.