squoosh
  • Blog
  • Partners
Sign in
squoosh
  • Blog→
  • Partners→
  • Sign in→

A practitioner guide

Pre-launch testing

TL;DR

  • Pre-launch testing means deciding which variant of a site change to launch before real customers ever see it.
  • It replaces the slow, traffic-burning screening phase of traditional A/B testing with a synthetic-shopper pass that ranks variants in under an hour.
  • Used right, it cuts live-traffic exposure on losing variants by 90%+ — you only run live tests on the strongest survivor.

What is pre-launch testing?

Pre-launch testing is the practice of evaluating site changes — copy tweaks, layout overhauls, pricing experiments, urgency banners, redesigns — before any of your real customers are exposed to them. Instead of pushing all the variants live, splitting traffic, and waiting for one to win, you run them through a synthetic-shopper pass and decide which one (if any) deserves live traffic.

Pre-launch testing doesn’t replace live A/B tests. It replaces the part of A/B testing where you spent live traffic just to find out which idea was a loser. The losers never make it to your customers.

How does it work?

  1. Bring a hypothesis

    Define the change you want to test. It can be one variant or ten. Squoosh handles paired ranking across as many variants as you bring.

  2. Run the synthetic pass

    Squoosh runs every variant through synthetic shoppers calibrated from your traffic. Output is a ranked list of variants with predicted conversion lift, drop-off points, and confidence intervals.

  3. Decide what gets live traffic

    The losers stay in the lab. The winner — or the top 1-2 candidates — get pushed to a live confirmation test, or directly to production if your team is willing to take the synthetic result as the decision.

  4. Confirm with a live test (optional)

    For high-stakes changes, run a live A/B test on just the synthetic winner. You’ll get to statistical power faster because you’re only running two variants (control vs synthetic winner) instead of five or ten.

When should you use it?

Pre-launch synthetic testLive A/B test only
Conversions burned on losersZeroUp to 50% per losing variant
Variants you can screen at onceUnlimitedLimited by traffic — usually 2-3
Time-to-decision on losersUnder an hourDays to weeks
Best forScreening many ideas, low-traffic pages, pre-launch QAFinal confirmation on a single change

Frequently asked questions

Should we still run live A/B tests?

For high-stakes changes — pricing, checkout, anything tied to revenue — yes. But run them on the synthetic winner, not on every candidate. Pre-launch testing is the screening pass that gets your live test to a single confident comparison.

What about low-traffic pages where live A/B tests never reach significance?

This is where synthetic testing is highest-leverage. A page with 2,000 visitors a month will never reach significance in a live A/B test before the test goes stale. Synthetic shoppers, calibrated from those same 2,000 visitors’ behavior, can rank variants in under an hour.

Can we test changes that aren’t live yet — staging URLs?

Yes. Squoosh can test variants from any URL, including pre-production. That’s the canonical pre-launch use case.

How does this affect our experimentation roadmap?

Most teams find they can run 5-10× more variants per quarter because the synthetic pass is fast and free of traffic constraints. The roadmap becomes idea-throughput-bound, not traffic-bound.

What if the synthetic winner doesn’t win the live test?

It happens — the benchmark is 81%, not 100%. When it does, you’ve lost a single live test, not five. The math still favors pre-launch screening because you saved live traffic on the other four ideas.

See it on your site

Bring a site change. We'll test it live in 30 minutes — no live traffic exposed.

Back to home →
squoosh

Rapidly increase conversions with synthetic shopper UX A/B tests.

Product

  • Overview

Company

  • Partner
  • Press

Resources

  • Blog
  • Documentation
  • Support

Legal

  • Terms
  • Privacy
© 2026 Squoosh, Inc.