Most pricing decisions in SaaS get made once, by gut feel, early on, and then never revisited with anything resembling a real test. That's understandable — pricing feels riskier to experiment with than a button color, because getting it wrong touches revenue directly and getting it wrong publicly touches trust. But it is possible to test pricing rigorously without either of those things happening, provided you're careful about who sees what and what you're actually measuring.
Test on new signups only
The single most important rule in pricing experimentation is that you test on new customers, never on existing ones. Showing two existing customers different prices for the identical product they're already using is the fastest way to turn a pricing question into a trust question — and if it ever surfaces publicly (screenshots get shared, support tickets get compared), the damage isn't limited to the customers directly affected. New signups haven't formed an expectation yet, so a price test just looks like the price — there's no before-and-after to feel unfair about.
In practice this means routing new traffic into price variants at signup, tracking which variant each account landed in, and never changing an existing customer's price outside of the deliberate, announced increase process (a different problem with a different playbook). The cohort boundary has to be clean: no existing customer should ever see a different number than the one they signed up under.
Depending on jurisdiction and how the test is structured, showing different prices to different customers for the same product can raise pricing-discrimination questions, particularly in consumer contexts. B2B SaaS has more latitude than consumer retail, but it's still worth a conversation with counsel before running a live price test, not after someone asks about it.
The metrics that actually matter
Conversion rate at each price point is the obvious metric and the one every pricing test dashboard leads with — but it's an incomplete answer on its own. A higher price that converts fewer trials but produces customers who stick around, expand, and rarely open support tickets can easily beat a lower price that converts more people who churn at the first renewal. The metric that actually settles the question is downstream: activation rate, retention through the first renewal cycle, and expansion revenue from that cohort, not just the initial signup number.
This is the part that's easy to skip because it's slow. Conversion rate you can read in a week. Whether a cohort actually sticks, you can't know until they've been through at least one renewal — which for an annual plan means waiting a year before you have a clean read. Anyone promising a definitive pricing answer in two weeks is measuring the wrong thing.
Willingness-to-pay research before a live test
Before committing engineering time and cohort exposure to a live price test, a Van Westendorp price sensitivity survey is a cheap way to narrow the range worth testing. It asks existing users or a target segment four questions — at what price does this feel too cheap to trust, cheap, expensive, and too expensive to consider — and the overlap between those answers gives you a band of acceptable prices rather than a single guess. It won't tell you the optimal number, and people's stated willingness to pay is reliably more generous than their actual behavior at checkout, but it will stop you from live-testing a price that was always going to fail, or missing a higher price nobody thought to try.
Treat willingness-to-pay research as a filter, not an answer. It tells you which two or three price points are worth the cost of a real cohort test — it doesn't replace the test. Skipping straight to a survey-derived number and shipping it to everyone is how you end up "confirming" a price nobody actually behaves consistently with.
Patience is the actual constraint
The hardest part of pricing experimentation isn't the statistics, it's the calendar. Trials run two to four weeks. First renewal for a monthly plan is a month out; for an annual plan, a year. A test that looks decisive after three weeks of signups is usually just measuring who was willing to start a trial, which correlates only loosely with who renews. Teams that call a pricing test early, based on top-of-funnel conversion alone, tend to lock in a price that looks great in the launch announcement and quietly underperforms two quarters later once the churn from that cohort shows up.
The practical approach is to run the test long enough to see at least one full renewal cycle for the plan length you're testing, and to keep the sample size honest — a test on a trickle of new signups over two months isn't a result, it's an anecdote with a chart around it.
Wrapping up
Pricing experimentation works when it respects three constraints: it never shows existing customers a different price than the one they agreed to, it measures retention and expansion alongside conversion rather than conversion alone, and it accepts that the real signal takes months, not weeks, to arrive. Willingness-to-pay surveys are useful for narrowing the range before you commit cohort exposure to a live test — they're a filter, not a substitute for one. Skip any of these and you'll get an answer quickly; it just won't be the right one.
Independent software engineer in Nairobi specialising in Acumatica customisations, Laravel backends, and tax fiscalisation integrations across East and Southern Africa.