How to test a price increase before you roll it out

The margin is what is left of each euro of price after the cost of serving the customer. At an 80% margin, a 10% increase pays for itself until 11.1% of customers leave because of it. At a 50% margin, the same increase can lose 16.7%.

The line compares a one-off loss. If the higher price keeps more customers leaving every month, the loss grows past the test window. So the read-out separates the first wave of leavers from a lasting change in churn, and annual plans, which react at renewal, are read when they renew.

• Start with new customers. Show the new price to a random half of sign-ups and compare how many convert. Existing customers never see it.
• Move existing customers in waves. Raise the price for a random 10% first, with a stop rule agreed before the test starts, and widen only if the first wave stays well below the line.
• Read revenue as well as leavers. Customers who move to a cheaper plan cost revenue without leaving.

How large the groups must be depends on how close to the line you expect to land. A test only has to show that the extra loss sits clearly below the break-even, which takes far fewer customers than measuring the exact effect.

The formula, and the group sizes it gives for sign-ups, a first wave and a lasting change in churn, are in How many customers do you need to test a price change?

Who leaves after a price change is a churn question, so the same model that predicts churn (see churn prediction) shows which customers to keep out of a first wave, and the forecast (see revenue forecasting) carries the result into next year’s revenue.

Who it is for. Subscription and software companies with no data team, planning a price change and wanting to know what it will cost them in customers before everyone sees it.

The questions it answers. How many customers you can lose before the increase costs money. How to test it, on whom, and with how large a group. What the test showed, and whether the effect is still growing.

What you need to have. Billing history with each customer’s plan and price; trial and sign-up conversion, if you sell that way; churn history; and the cost of serving a customer, even a rough one.

The method, in plain words. The break-even first. Then the test design: who gets the new price, the group sizes, the stop rule, all agreed before anyone sees a price. Then the read-out: the extra loss with its range, against the line, and a check for a lasting rise in churn.

What you receive. A test-design memo before the test, and a read-out memo after it, each of up to two pages. A method note listing the checks that ran.

How it is checked. Agents build the analysis and draft the memos. Every deliverable gets number and claim checks run by scripts, not AI, and a review by an agent. On what you act on, such as a model going live or a decision memo, that review is a panel of three reviewer agents and a judge, and then the analytics director signs it; routine refreshes do not need a sign-off. The checks are described on how it works.

Your data. Where it goes, which model providers process it and what happens when the work ends: see data handling.

Price. Fixed per deliverable. A price test, designed and read out, is a sprint, from €12k, fixed in a one-page proposal before work starts. Repeat work runs on an async retainer, from €8k a month.

Not covered. Setting the price itself, packaging and positioning, and research into competitors’ prices. A price test rests on churn and revenue analysis: the same models the team builds for churn and customer value. Pricing is one of the money questions the same team works on.

A subscription software company has 2,400 customers on its Starter plan at €29 a month, an 80% margin, and 5% of them leaving a month. It wants to move Starter to €31.90, a 10% increase. Its break-even is 267 customers, 11.1% of the plan.

It raises the price for a random 1,200 of them and keeps 1,200 at €29. After three months, 16.9% of the new-price group has left, against 14.2% of the old-price group.

The 95% interval is a range built so that, in 95 tests out of 100, it holds the true extra loss. In a test, the line counts only the customers who would have stayed at the old price, 85.8% of the group over three months, so it sits at 9.5 points rather than 11.1. Even the interval’s upper end, 5.6 points, is below it, so the increase pays on the first wave. What the read-out still has to show is whether the new-price group keeps leaving faster after month three.

A first wave of 240 gives a wider range, and its upper end, 7.6 points, still sits below the line. The same test, read out in full, is the example decision memo; the price-test calculator works out the group sizes on your own rates.