Situation. A SaaS business considering a price increase. The board wanted the revenue. Management was afraid of the churn, and the fear was not unreasonable — nobody could say what raising prices would cost in lost conversion, so the discussion had run for months without resolving.
The number nobody had computed. Expected initial revenue per visitor at the higher price. Not the price, and not the conversion rate, but the two combined against a constant traffic base — because a price increase that costs you conversion can still leave the same revenue coming out of the same traffic, or more. Per visitor is the right denominator: measured per customer, the conversion loss disappears from view entirely. And initial rather than lifetime, because a price test has to be readable in weeks, not after a full cohort has run its course.
Until that number exists, the discussion is between two people guessing, and the more cautious guess wins by default.
The test and how we made it readable. We split-tested the higher price against the existing one on new customers only, leaving the existing base untouched. That mattered for two reasons. It removed the risk that made management nervous, and it produced a clean read: two populations arriving through the same channels at the same time, differing only in price.
Result. The higher price produced the same expected initial revenue per visitor as the control — the conversion loss was real, and it was exactly offset by the higher price. Read carelessly, that is a tie. Read correctly, it was the answer: the increase cost nothing on new business, which meant it could be extended to the entire existing base — and there, applied to revenue already flowing, it was worth a great deal.
We paired the rollout with a pre-committed retention policy: any existing customer who objected received a credit covering the increase for a full year, granted on request, no negotiation. It intercepted nearly all of the at-risk churn at a small fraction of the revenue gained.
What transfers. Price tests read on expected initial revenue per visitor. Any denominator further down the funnel has already absorbed the conversion loss you are trying to measure.
A test run on new customers can license a decision about existing ones, at a fraction of the risk — the new cohort is a proxy population, and it answers the question without exposing the base. And a result showing no difference is not a failed test: when the variable is price, parity on revenue per visitor is a green light, because the increase carries through to every customer you already have.
