Peak Experiments

How do I create my first experiment?

Describe in one sentence what you want to improve. Peak turns it into a draft with a hypothesis, primary metric and guardrails and names what's still open. Then you choose “Start simply” or “Plan carefully”. Nothing is saved until you save, and nothing starts until you start it.

1. 01Open Peak Experiments and choose “+ Experiment”.
2. 02Write what you want to improve and why you believe a change will help.
3. 03Check the draft and answer the open questions.
4. 04Enter the baseline, volume per week and success threshold.
5. 05Start as soon as the critique no longer shows anything critical.

How do I enter the result of an experiment?

After the run, with “End and evaluate”. You enter either both groups (units and successes, or mean) or just the change in the primary metric in percent. Without group sizes, evidence is at most C.

1. 01Open the experiment and choose “End and evaluate”.
2. 02Choose “Enter groups” or “Enter change only”.
3. 03Enter the values for control and variant, plus the guardrails in %.
4. 04Optional, under “Segments, data quality and confounders”: segments, data problems and what happened at the same time.
5. 05“Save as interim result” or “Save final result”. Then the decision is due.

How do I decide after an experiment?

After the final result, you choose a decision: Roll out, Iterate, Retest, Roll back, Wait, Don't pursue or No clear result. Peak marks one as “recommended”. You write briefly why. The decision then also appears in Peak Log.

1. 01Save the final result first.
2. 02Choose the decision and write the rationale under “Why?”.
3. 03For “Roll out”, you can optionally set an owner, “Roll out by” and a task for Peak Work. Leave it empty if no task is needed.
4. 04Save.

How do I record what a rolled-out experiment actually delivered?

After the “Roll out” decision, you enter the measured change after rollout in percent on the experiment. That way you see later whether the effect holds in day-to-day operations. Before “Roll out”, this field doesn't exist.

1. 01Open the rolled-out experiment.
2. 02Under “Change after rollout (%)”, enter the measured value, and optionally the “Value per month (€, optional)”.
3. 03Save.

How do I collect ideas and turn the best one into an experiment?

Under “Ideas”, you capture an idea in one sentence. Each idea gets a PEAK score from 0 to 100 based on potential, evidence, alignment and ease. You turn the best ones into experiments. Peak also suggests ideas from open questions in your learnings.

1. 01Open “Ideas” and write a sentence under “New idea”, for example “What if we made shipping free from €50?”.
2. 02Click “Capture idea”.
3. 03Use “Rate” to adjust the four values.
4. 04“Make it an experiment” creates a draft. “Discard” removes the idea.

What does evidence level A to E mean?

It says how robust a result is methodologically, not how confident you feel. The design sets the upper limit: randomized at most A, quasi-experimental B, before/after C, qualitative D. Every limitation lowers it by one level: sample too small, stopped early, data problems, confounders, accepted overlaps, or changes to the hypothesis, metric or threshold after the start. Under “Advanced”, every reason is listed.

How does Peak calculate sample size and duration?

From the baseline, success threshold and volume per week, with 5% significance and 80% power, two-sided. For averages without a known spread, Peak assumes a standard deviation equal to the mean and says so. If a classic test needs more than eight weeks, low-traffic mode suggests alternatives: a bigger change, sequential testing, switchback, a comparison period or interviews.

Where do learnings go, and what does “contradicted” mean?

Every learning is in the library, with its scope, effect, evidence and the experiments behind it. You can link learnings: supports, contradicts or supersedes. Contradicted and superseded learnings stay visible, and after a year or on the review date a learning is set to “Review”. Peak doesn't make up contradictions.

Can I pause or cancel a running experiment?

Yes. Depending on its state, the experiment offers “Pause”, “Resume”, “End and evaluate”, “Mark as blocked”, “Mark as invalid”, “Back to draft”, “Cancel” and “Archive”. At the start, Peak freezes hypothesis, metric, threshold, guardrails and design, so the result stays reliable later. If a step doesn't fit the current state, Peak says so.

What do the Insights in Peak Experiments show?

Velocity, results, evidence and impact across all experiments: what's running, what needs deciding, the distribution of results, what's been realized and what's still untapped, each per month. Plus a weekly review (“Five minutes for the week”). Every statement names what it's based on. Insights appear once the first experiments have run and been decided.

Why can't I start my experiment?

The review still shows critical issues. As long as a critical issue is open, “Start” stays locked. If your experiment overlaps with another one, you can deliberately accept the overlap; that lowers the evidence by one level.

1. 01Read the list above the start button. At the bottom it says how many issues are blocking the start.
2. 02Fix the issues in the plan, such as a missing metric or threshold.
3. 03For an overlap: enter a reason under “Deliberately accept the overlap because …” and click “Accept”.
4. 04When it says “No open issues. The plan is ready to start.”, click “Start”. If the start day is in the future, the button is called “Schedule”.