ASO course · lesson 42 of 60

ASO experiment reporting and rollout decisions

Produce a result memo another practitioner can challenge, reproduce and act on.

By Gabriel Machuret · Editorial methodology

Before you start: A completed experiment or a clearly labeled hypothetical result set.

Module 7: Experiment design and interpretation

Plan about 55 minutes for reading, practice and review

You will produce: Module project: a preregistered brief, results memo and owned rollout decision.

Understand the decision

Reconstruct what actually happened

Start with the original question, control, treatment, population, metric and planned rule. Then record actual dates, sample context, interruptions and deviations. Do not rewrite the hypothesis to match the result. If setup differed from the plan, explain the consequence. The report should show whether the result answers the original decision or a narrower question.

Show evidence and competing explanations

Present counts, rates, effect and method-specific uncertainty together. Note relevant source mix, release events and operational issues. A randomized design addresses some confounding, but does not excuse broken implementation or unsupported generalization to another market. A nonrandom comparison needs stronger qualifications. Include neutral or negative findings rather than selecting only favorable charts.

Turn the result into controlled action

State adopt, retain control, investigate or inconclusive, with a reason tied to the original rule. Assign the rollout owner, affected surfaces, verification and follow-up. Preserve the losing concept and learning; a treatment may fail because its promise is wrong or because the execution was unclear. The next hypothesis should follow evidence, not an obligation to make every test sound successful.

Apply the method in practice

Reconstruct the experiment before interpreting it

Start with the original question, control, treatment, audience, dates and metric definitions. Confirm that the intended configuration actually ran. List interruptions and concurrent changes. If the treatment differed from the approved brief, explain the difference before showing the result. This protects the reader from interpreting an attractive chart as evidence about a design that was never tested. Keep the original hypothesis even when the outcome is disappointing.

Report the whole decision-relevant result

Show counts where available, rates, effect estimates and the native uncertainty measure. State the statistical method or link the platform guidance. Include relevant guardrails and their measurement limits. Segment analysis can be useful, but unexpected subgroup patterns should usually be treated as exploratory unless planned and supported. Avoid selecting only the segment where the preferred treatment looks strongest. A readout should make an informed reviewer able to disagree with the recommendation using the same evidence.

Turn the result into a bounded action

Choose adopt, reject, investigate or repeat with a specific reason. State where the conclusion applies and what remains uncertain. Before rollout, recheck product accuracy, locale suitability and operational readiness. Record the person responsible for applying the treatment and verifying the live result. Close the loop by adding the learning to the hypothesis ledger. A failed or inconclusive experiment can still be valuable when it prevents repeated work or clarifies the next question.

Platform references for this work: Apple — Product page optimization results · Google Play — Run store listing experiments

Your step-by-step procedure

  1. Compare the original brief with the actual configuration and timeline.
  2. Report the primary result with counts and the appropriate uncertainty interpretation.
  3. List limitations, guardrails and plausible explanations not resolved by the design.
  4. Make a rollout or follow-up decision with an owner and verification step.
1. Compare the original brief with the actual configuration and timeline. 2. Report the primary result with counts and the appropriate uncertainty interpretation. 3. List limitations, guardrails and plausible explanations not resolved by the design. 4. Make a rollout or follow-up decision with an owner and verification step.
Graphic 1. The procedure. Select the graphic to view or save the full-size version.

Worked example and interpretation

Trail Notes and numerical research scenarios are fictional teaching examples. Platform limits, where shown, come from the linked official references.

Trail Notes’ journal-led concept produces a promising but inconclusive native result. The readout keeps the existing control, records that the concept passed comprehension review and proposes a stronger demonstration of the journal workflow for a future test. It does not use the positive point estimate as a public success statistic. The report still delivers value by narrowing the next question and preventing an unjustified rollout claim.

Readout section: Setup; Content: Actual treatment and audience; Decision value: Establishes what was tested. Readout section: Evidence; Content: Counts, effect and uncertainty; Decision value: Shows strength and magnitude. Readout section: Context; Content: Deviations and guardrails; Decision value: Reveals limits and tradeoffs. Readout section: Action; Content: Owned rollout or next investigation; Decision value: Makes the learning usable
Graphic 2. The worked example. Select the graphic to view or save the full-size version.
Example details
Readout sectionContentDecision value
SetupActual treatment and audienceEstablishes what was tested
EvidenceCounts, effect and uncertaintyShows strength and magnitude
ContextDeviations and guardrailsReveals limits and tradeoffs
ActionOwned rollout or next investigationMakes the learning usable

A result summary without setup or uncertainty is incomplete. A long report without an action is also incomplete. The structure above keeps the analysis connected to the original decision and the next operational step. If evidence is inconclusive, say why another test would be different before recommending repetition. Otherwise, the team may spend another cycle producing the same uncertainty.

Build a handover someone can use

Original brief: Attach the hypothesis, configuration and exact asset versions that actually ran. Result evidence: Preserve native reports, counts, effect estimates and uncertainty with definitions. Validity review: Explain deviations, audience changes and limits on subgroup interpretation. Recommendation: State the bounded decision and why the evidence supports it. Execution and learning: Assign rollout verification or further research and update the experiment history.
Graphic 3. The handover. Select the graphic to view or save the full-size version.

Original brief

Attach the hypothesis, configuration and exact asset versions that actually ran.

Result evidence

Preserve native reports, counts, effect estimates and uncertainty with definitions.

Validity review

Explain deviations, audience changes and limits on subgroup interpretation.

Recommendation

State the bounded decision and why the evidence supports it.

Execution and learning

Assign rollout verification or further research and update the experiment history.

Your ASO assignment

Use your own app and evidence, or work through the teaching case. Keep your observations separate from assumptions and explain the reasoning behind your decisions.

  1. Write a one-page readout from a hypothetical or permissioned result.
  2. Include an explicit section on what the test does not establish.
  3. Prepare a rollout checklist or follow-up hypothesis based on the decision.

Scenario challenge

Write a five-sentence readout for an inconclusive test: question, setup, observed result, limitation and next action. Avoid using “failed” as a substitute for explaining what the team learned.

Assess your work

Common mistakes to catch

Knowledge check

A valid test wins in one locale. Does the result justify replacing assets globally?

Read the answer and reasoning

It supports a decision for the tested context. Assess language, audience, product and source differences before extending it. A staged adaptation with verification or further testing is more defensible than treating one locale’s result as a universal law.

Sources and further reading

Platform references reviewed 29 September 2026. These sources support platform capabilities and constraints; the teaching frameworks, assignments and illustrative cases are original course material.

Your ASO lesson notes

Use this space for your assignment, evidence and handover. Select Save to keep a draft in this browser, or download a copy to take with you.

Notes stay on this device and are not submitted to ASO Agency.

Complete your assignment with the ASO experiment planner.

Loading your progress…

Progress is saved only in this browser. It does not sync across devices. Mark a lesson complete after doing its exercise.