Pricing Experiment Designer

Most pricing "experiments" are price changes with anxiety attached: no hypothesis, no control, no pre-agreed verdict rule — so whatever happens, someone argues it worked. This skill designs pricing tests that end in a decision: a falsifiable hypothesis, a treatment that is fair and lawful to run, guardrail metrics that catch collateral damage, and a call rule written down before launch so the result cannot be negotiated after the fact.

When to use this skill

Considering a price change and wanting evidence before committing the whole base to it
Testing packaging moves: a new tier, a feature moving up-market, annual-versus-monthly framing
Entering a new segment or region where current pricing is a guess wearing a spreadsheet
Settling a standing argument — "we are too cheap" versus "we would crater conversion" — with data

Workflow

Write the decision first. "If X, we roll the new price to all new customers; if Y, we keep current pricing; if ambiguous, we run Z next." An experiment that cannot change a decision is theater — skip it and spend the effort elsewhere.
State the hypothesis with direction and size. "Raising the mid tier 20 percent reduces new-customer conversion by less than 8 percent and lifts revenue per qualified visitor at least 10 percent." Vague hopes cannot be falsified; numbers can.
Choose a design that is fair by construction. Prefer new-cohort tests where only new prospects see the test price, geographic or segment splits, packaging tests (same price, different composition), or clean before-and-after cohorts. Showing different customers different prices for the identical public offer invites fairness blowback and legal exposure in some jurisdictions — flag any such design for human and legal review before launch.
Pick the primary metric that matches the decision: usually revenue per qualified visitor or per trial start, never conversion rate alone. A price increase that drops conversion 10 percent while lifting revenue 15 percent is a win that a conversion-only readout would kill.
Set guardrails with tripwires: refund and chargeback rate, support ticket volume and tone, trial-to-paid time, early churn within the test cohort, and sales-cycle length for sales-led motions. Define in advance the guardrail level that halts the test early.
Size the runtime honestly. Estimate weekly qualified traffic per arm and how long before the difference you care about separates from noise; on modest traffic, pricing tests routinely need six to twelve weeks. Commit to the window — peeking early and stopping on a good day is the classic way teams ship noise with confidence.
Decide the grandfathering policy before launch. Existing customers keep their price unless a migration is designed deliberately, separately, and with notice. Test-cohort customers keep whatever terms they signed up under; honoring the price people accepted is non-negotiable.
Run, log, call it. Freeze the design doc at launch. At the pre-committed end date, apply the decision rule as written, then record the result and the decision in a pricing log so the next debate starts from evidence instead of folklore.

Output format

A one-page experiment spec: decision statement; hypothesis with numbers; design and arms; population and exclusions; primary metric; guardrails with halt thresholds; runtime and traffic arithmetic; grandfathering policy; the decision rule; owner; launch and readout dates.

Guardrails

No test touches existing customers' current prices without an explicit, separately designed migration plan
Any design where identical buyers could see different prices goes to fairness and legal review first
Revenue per visitor over conversion rate, with the acceptable conversion tradeoff stated in advance
The decision rule is frozen at launch; post-hoc metric shopping voids the experiment
If traffic cannot resolve the question inside a quarter, choose a sequential cohort design or a reversible full rollout with a rollback threshold — and write down which, and why