Nobody Underprices On Purpose

gps-tracker

By Ilia Grigorian, CPO at Findmykids – Findmykids is a family-safety and location-sharing app with around 7 million monthly users in 180 countries https://findmykids.org/

Every pricing discussion I have sat in has the same shape. Raise a price and lose customers and the decision has an author, a date and a dip in a chart. Leave a price 30% too low for three years, and it has none of those.

It looks like a flat line, and nobody is ever asked to explain a flat line. Most subscription prices are set by that asymmetry rather than by analysis.

At Findmykids, a family-safety subscription app sold in more than 170 countries, we used to run fewer than ten price tests a year. We assembled each one by hand, and every one risked a mismatch between the App Store and Google Play or a broken ratio between the monthly and annual plans.

I built a system to fix that. Price stopped being a one-shot decision per market, taken once and defended afterwards.

The rest of the industry behaves the same way. RevenueCat’s 2026 report covers more than 115,000 apps and $16bn in revenue, and finds median pricing “largely stable”, with weekly and monthly medians flat at $5.99 and $10, and only the yearly median ticking up from $31.60 to $34.80.

The most common price points ($5, $10, $30) are the report’s own “sticky psychological anchors” and among monthly plans only the top performers moved, from $20 to $22.70.

Adapty’s 2026 benchmark, across 16,000 apps, adds that nine in ten subscriptions sell at full price, and that price changes are the weakest experiment type for conversion. They improve it in 28.3% of tests, while still lifting lifetime value in 45.5%.

Those last two numbers are where teams get caught. A price test usually makes the conversion chart look worse and the revenue better, so the metric everyone watches punishes the move that pays, and founders learn to leave the number alone. But the two kinds of test do not work the same way.

A feature test is a hypothesis. It works, or it doesn’t, and failure closes it. A price test is a search, and if this number doesn’t fit, another one will. It also doesn’t compete with your other experiments. It runs as a second layer underneath them.

Four Rules Keep The Search Cheap

 

Work out what you can afford to lose before anyone argues about the number.

Revenue per new user is conversion times price, so every increase comes with a budget. At an unchanged renewal rate, you can lose 1 − 1/(1+Δp) of paying conversion and stay level.

 

Price move Conversion you can lose and stay level, renewals unchanged
+20% 16.7%
+30% 23.1%
+40% 28.6%
−20% none; you need conversion to rise by 25%

 

The budget holds only while the subscription lifetime does, and in our data a move of 20 to 30% changes who buys rather than how long they stay. In a +30% test in Israel, paying conversion did not fall at all. It moved up, short of significance, and ARPU finished 27.8% up. I had priced the test expecting to lose at least 5% of conversion, and in the end we lost none of it.

Build the grid first, then run pricing as a second stream on new users only – I built 15 pre-configured bands per currency-and-country pair, roughly 20% apart. Configuring a test takes under five minutes and no engineering time, so pricing only ever competes for cohorts, and a shared calendar keeps a price test and a product A/B off the same cohort. Plans move together, so no test re-packages your tiers by accident. We never put existing subscribers into a test.

Decide on a 13-month horizon, and let retention hold the veto – Results run to 13 months, so the first annual renewals sit inside the window. Each plan and tier has its own renewal average, and purchases after a user’s first month count only as cash charged. The veto I set at the beginning is a ratio. Unless ARPU rises at least 1.7 times faster than retention falls, we leave the price alone. We roll back automatically only when something breaks; the decision itself waits two to three weeks, because trials charge on days three to fourteen and early data lies.

Keep searching, because prices expire – Products improve, so an old no goes stale. The United States rejected an increase, with ARPU at −0.6% and retention softening. Then, 1.5 years later, I put it back, because our product had become better, and that run gave 20% ARPU growth. France took two rises in a row, 13% then another 6. It runs downhill too. In Indonesia a 20% cut lifted conversion enough to raise ARPU by 25%, with retention up 7.

Since 2023 the programme I established has run more than 200 controlled tests and won about 45. A 23% hit rate sounds like failure until you ask what we were searching for. Copying a competitor’s price takes twenty tests. Finding a price nobody charges takes two hundred. Cadence went from fewer than ten price tests a year to more than a hundred across key markets, and realised price levels moved by 20 to 30%, and by as much as 40 in some markets.

ARPU in the UK and Poland now sits 35 to 55% above the pre-programme baseline, measured against holdout cohorts that are still running. Most losses land in a dead band between −5% and +5%, and the worst cost 15 to 20% of ARPU, on the tested half of new users in one country, for a fortnight.

RevenueCat, FunnelFox and Qonversion all ship price testing now. In 2023 none of them offered it, which is why I built the grid in-house. What they sell is the split and the rollout, which is the cheap half. The pricing mechanics stay with you, and so does the calendar that keeps two tests off one cohort. You are the one who knows how your own tiers relate, and which of your tests must never run together.

The serious objection is that underpricing is deliberate: cheap entry buys users, word of mouth and share. It might be, but it is less dependable than it sounds. A low price does not automatically buy conversion, and it buys renewals less often still. Large cohorts react to price counter-intuitively, in both directions.

All of it is testable, and Indonesia is what it looks like when the test agrees with you. If you never run that test, you do not have a strategy; you have a habit nobody has questioned. The fairer objection is volume. Published guidance puts the detection floor at roughly 1,000 to 5,000 trial starts a month per market, and below that you should run fewer markets with bigger steps.

The bill for the fear arrives twice. The first instalment is revenue you never collected.

The second is worse, because margin is how a market confirms that a product deserves to keep being built, and margin is what pays for the speed to build it. A product that never asks for more never finds out what it was worth. I would rather find out.