Experiment sample-size honesty

How to use this guide

Use this guide to admit when you cannot detect a small effect with a rule you can inspect. Skip anything that requires a fake benchmark, a guaranteed inbox, or a statute this page does not claim to interpret.

Work section by section. Keep what matches your data, capacity, and qualified counsel. Discard anything that would require shipping a 'winner' from 200 sends.

What experiment sample-size honesty is for

Experiment sample-size honesty is easy to name and easy to misunderstand. In a retention program it is the operating practice that helps a team admit when you cannot detect a small effect. If the work does not change eligibility, message, timing, channel, offer, suppression, or measurement, it is decoration—even if the subject line is clever.

Retain Inc uses experiment sample-size honesty as a planning object inside measurement, not as a campaign theme. That means a written job, a source of truth, and an owner who can stop the work when it harms customers. We do not present this page as a client case study, and we will not invent a statistic to make the definition feel more 'benchmarked.'

Write the definition in language a new teammate can use. 'Experiment sample-size honesty means we admit when you cannot detect a small effect.' Add what it is not: it is not shipping a 'winner' from 200 sends. Keep the constraint visible: underpowered tests should not change policy. Those three sentences prevent a quarter of the implementation arguments that otherwise happen in Slack.

Platform features can help, but Klaviyo, HubSpot, Salesforce, or Shopify will not invent a definition you refused to write.

How to define the window and the population

Every useful tests artifact changes a decision. For experiment sample-size honesty, the decision is whether a person is eligible, what they should receive, when they should receive it, and who is accountable. If two teams can apply the idea and get opposite customer experiences, the decision is not specified yet.

Start with the smallest change that still helps you admit when you cannot detect a small effect. Then name the people who must agree: marketing, CRM, service, and whoever owns sample notes in the log. A decision that cannot survive a support ticket is not a retention decision.

Composite example: a team discusses experiment sample-size honesty in a workshop, then ships a calendar send that still shipping a 'winner' from 200 sends. Nothing in the CRM changed. The useful version of the meeting ends with a field, a rule, a suppression, or a retired journey—not with a headline.

A useful working session ends with a named owner for sample notes in the log and a date to look again.

What this metric cannot prove

Data for experiment sample-size honesty should be boring enough to trust. List the fields, events, and consent flags required to admit when you cannot detect a small effect. For each, record source, freshness, allowed values, owner, and what happens when the value is missing. Unreliable personalization is worse than a clear default.

Eligibility is where measurement becomes customer experience. Include who must be excluded: unsubscribed, deleted, do-not-contact, active complaints, in-flight returns, open high-severity tickets, employees, test profiles, and anyone outside the purpose of the capture. Underpowered tests should not change policy.

Consent is not a banner screenshot. Channel permission, disclosed purpose, timestamp, and source should travel with the record. If you cannot reconstruct why a person is receiving experiment sample-size honesty related mail, you are guessing. Guessing is how complaint rates and legal risk both rise. This guide is educational and is not legal advice.

A useful working session ends with a named owner for sample notes in the log and a date to look again.

How it connects to journeys and CRM stages

Operating experiment sample-size honesty means collisions, versioning, and a kill switch—not only copy. Map which live journeys can reach the same person in 48 hours. Give experiment sample-size honesty a priority. If a more important operational message is in flight, this work should wait or skip.

Document the happy path and the exits: purchase, booking, opt-out, bounce, complaint, reply, disqualification, and entry into a higher-priority journey. Duplicate events should not duplicate sends. If a webhook retries, the customer should not live the retry.

Quality assurance should include identity, merge-tag fallbacks, inventory or appointment truth, links, rendering, quiet hours, and a sample of excluded people who must not receive the message. Experiment sample-size honesty fails more often on data than on fonts. Keep a plain-language logic note so the practice survives vacation coverage.

Owners should be able to explain experiment sample-size honesty to a customer in one sentence that matches the permission they were shown at signup.

Reporting habits that keep it honest

The signature failure is shipping a 'winner' from 200 sends. It is attractive because it is fast and it looks like activity. It usually produces a short spike in a dashboard and a longer problem in sample notes in the log.

Adjacent failures include treating experiment sample-size honesty as a slogan in a kickoff deck, copying another brand's screenshots, and reporting platform-attributed revenue as incremental lift. None of those help you admit when you cannot detect a small effect. Composite example: a team 'launches experiment sample-size honesty' by renaming a blast, then wonders why unsubscribes moved while the customer relationship did not.

Build a refusal list. Refuse purchased lists, invented statistics, fake client names, guaranteed inbox placement, and any copy that operations cannot fulfill. Refuse to shipping a 'winner' from 200 sends. If a stakeholder asks for a number Retain Inc cannot defend, the answer is a method and a limitation—not a fictional benchmark.

A useful working session ends with a named owner for sample notes in the log and a date to look again.

What to change when the number moves

Measure experiment sample-size honesty against sample notes in the log. Delivery, clicks, and opens can diagnose friction, especially after privacy protections damaged open rates, but they are not the outcome. Tie the work to a customer behavior and, where you can see it, to contribution margin.

When possible, use a holdout or another comparison that estimates what would have happened anyway. When that is not practical, say so. Last-click attribution can still be a useful operational view if you label it as association. Do not brief a board on causality you do not have.

Create a review rhythm: weekly health (did we violate underpowered tests should not change policy?), monthly learning (did we admit when you cannot detect a small effect better than last month?), and a test log with hypothesis, dates, audience, result, limitations, and decision. If the number moved and nobody changed a rule, you are watching weather.

Owners should be able to explain experiment sample-size honesty to a customer in one sentence that matches the permission they were shown at signup.

Working decisions

Use this table in a live working session. Replace the examples with your actual fields and owners. The point is to make Experiment sample-size honesty operable.

Implementation checklist

Print or copy this list into the brief. If an item is missing, you are not ready to automate Experiment sample-size honesty.

• Job statement exists: we admit when you cannot detect a small effect.
• Failure mode is listed on the brief: do not shipping a 'winner' from 200 sends.
• Consent, suppression, and missing-data fallbacks are defined.
• Collision rules and a kill switch are named.
• Sample notes in the log has an owner and a review date.
• Constraint is treated as a gate: underpowered tests should not change policy.

Frequently asked questions