A practitioner guide
A/B testing without traffic
TL;DR
- Most A/B testing tools fail on low-traffic pages — they need thousands of conversions per variant before declaring a winner.
- Synthetic A/B testing decides in under an hour by running modeled buyer cohorts through your variants instead of waiting on live visitors.
- It’s the only practical way to test pages that get a few thousand visitors a month and the only practical way to test pre-launch changes.
What is a/b testing without traffic?
Most teams hit the same wall: a traditional A/B test needs thousands of conversions per variant before the result reaches statistical significance. On a page that gets 2,000 visitors a month with a 2% conversion rate, that means months — and most ideas don’t survive the wait.
A/B testing without traffic is the synthetic alternative: run the variants through modeled shoppers calibrated from the small amount of traffic you do have. The result comes back in under an hour with confidence intervals you can act on.
How does it work?
Identify the constraint
The page either has too little traffic for a meaningful live test, or the change is pre-launch (no live traffic possible at all). Either way, traditional A/B testing isn’t a viable path.
Calibrate from what you have
Squoosh uses whatever behavioral data you do have — even small samples — plus aggregate industry signals to calibrate a synthetic cohort that resembles the population you’d expect to see on that page at scale.
Rank variants synthetically
Every synthetic shopper traverses every variant. Squoosh ranks variants by predicted conversion lift with confidence intervals.
Decide and ship
You ship the winning variant — either straight to production, or as the single live test if your stakes warrant a confirmation. Either way, you didn’t spend months waiting for traffic that wouldn’t have produced significance.
When should you use it?
| Synthetic on low-traffic page | Traditional A/B test on low-traffic page | |
|---|---|---|
| Time-to-decision | Under an hour | Months — or never reaches power |
| Minimum traffic needed | Calibration sample (often a few hundred sessions) | Thousands of conversions per variant |
| Useful on pre-launch pages | Yes | No (no live traffic exists yet) |
| Risk of stale test | None — runs to completion immediately | High — calendar-time changes confound results |
Frequently asked questions
How small can my traffic be and still get a useful synthetic result?
Squoosh has been used on pages with as few as a few hundred sessions per month. Below that, calibration relies more on industry priors and less on your specific data, but it still produces a directional result that’s better than no test at all.
What about brand-new pages with zero live traffic?
These are pre-launch pages. Squoosh calibrates from sibling pages (same site, similar funnel position) and from industry priors. The result is directional rather than precise — but it’s actionable, which is more than a traditional A/B test can offer.
Will this result ever match a live A/B test on the same page?
When live traffic is sufficient to power a live test, the synthetic result matches the live winner 81% of the time on average. On low-traffic pages, the live test usually never reaches power, so there’s no live result to compare against — the synthetic result is the only result.
Can I use this in addition to my existing A/B testing tool?
Yes. The common pattern: synthetic testing for screening (rank 5 ideas down to 1), then your existing A/B tool (Optimizely, VWO, AB Tasty, Adobe Target) for live confirmation on the winner.
How is the cohort calibrated when I have very little data?
Squoosh blends your available signal with industry priors weighted by similarity (vertical, funnel position, device mix). Calibration uncertainty propagates through to the confidence intervals on the synthetic result, so you see when the data is thin.
See it on your site
Bring a site change. We'll test it live in 30 minutes — no live traffic exposed.