Reporting

SRM warnings

When should Sample Ratio Mismatch warnings be taken seriously?

Last updated: May 22, 20268-14 minBeginner to intermediate
Start free See pricingMethodology

Same data, three statistical views

Testvoy does not reduce every experiment to one shiny score. The same experiment data is read with Bayesian decision probability, Frequentist validation and SRM traffic quality checks. This helps reduce early-decision risk, allocation issues and explanation gaps across teams.

Bayesian: probability to beat control and expected loss

Frequentist: p-value and confidence interval

SRM: whether traffic allocation deviates from expectation

Revenue: lift can be read with revenue impact too

Stats engine
BayesianFrequentistSRM
Bayesian, Frequentist and SRM signals meet in one decision panel.

When should I decide?

A high lift on a small sample can look exciting while still being risky. Ending a test without reading minimum sample, traffic split, SRM, bot filtering and segment behavior together can produce the wrong winner.

  • Is sample size sufficient?
  • Is there an SRM warning?
  • Any bot or dropped event signal?
  • Does the CI support the business decision?
  • Do revenue and segments point the same way?

Example report-reading flow

Example: Variant B looks strong in total conversion. Before deciding, read funnel steps, segment breakdown, revenue impact, bot-filtering signals and SRM warnings. This order reduces the risk of deciding early on one metric.

  1. 01

    Read the Best variant card

  2. 02

    Check Bayesian/Frequentist signals

  3. 03

    Look for SRM warning

  4. 04

    Inspect funnel drop-off steps

  5. 05

    Compare device, country and UTM segments

  6. 06

    Check whether revenue or RPV supports the same direction

  7. 07

    Create a shareable report

Example decision note
Variant B shows +8.4% lift on signup_completed.
No SRM warning.
Mobile segment is stronger; desktop is neutral.
Checkout drop-off did not change.
Decision: move Variant B to 25% rollout.

Practical scenario

A growth team changes CTA copy on the pricing page, connects signup_started and signup_completed goals, then reads Google Ads and returning visitor segments separately. If the result is promising, they share a report link with the client or leadership.

Common mistakes

Most mistakes come from wrong project keys, overly broad selectors, missing goals, staging/prod mixups, early decisions on small samples or skipping mobile QA.

Naming an experiment 'test'

Launching without a goal

Deciding on total CVR only

Sending screenshots instead of shareable reports

Related pages

ReportingStatistics methodology

Frequentist, Bayesian, probability to beat control, confidence, SRM and early decision risk.

Security and data qualityBot filtering

How bots pollute results, which signals Testvoy checks and how review workflow works.

Frequently asked questions

Who is SRM warnings for?

Marketing, growth, product and developer teams can use the same guide at different depths.

Do I need a developer?

Not for basic visual changes and report reading; SDK, custom events, custom JS or enterprise security benefit from technical support.

How do I know it worked?

Verifier, recent events, preview, QA checklist and report warnings show setup and experiment health.

What is the common mistake?

Wrong project keys, missing goals, fragile selectors, early decisions on small samples and skipped mobile QA are common.

Docs

Try this guide in the app

Finish setup, create your first experiment and read the report with cleaner data in Testvoy.

Start free Docs