A definitional guide
Synthetic shoppers
TL;DR
- A synthetic shopper is a simulated buyer modeled from aggregated behavioral signals, used to evaluate site variants without exposing real customers.
- Squoosh calibrates its synthetic shoppers from 250M+ behavioral signals — device, funnel position, intent, category affinity — so the cohort matches your real traffic.
- Synthetic shoppers contain no PII, no cookies, no tracking of real humans. They’re a population, not a database of people.
What is synthetic shoppers?
A synthetic shopper is a simulated buyer — a model — used to traverse variants of a website and report on their behavior. Synthetic shoppers don’t exist as real people; they’re statistical constructs calibrated to behave like the population of real shoppers a site actually sees.
The reason they exist: running a traditional A/B test means exposing real customers to every variant, including the losers. That burns conversions. Synthetic shoppers let you evaluate variants without that exposure, by running a modeled population through every variant in parallel.
How does it work?
Calibration from your data
Squoosh reads your Shopify orders, GA4 events, Segment streams, or BigQuery exports. The signals — device mix, funnel drop-off points, average cart size, time-on-site — are aggregated into cohort parameters that drive the synthetic population.
Generation
Squoosh instantiates a population of synthetic shoppers whose collective behavior matches your calibrated cohort. Each one has a buyer segment, an intent, a device, a funnel position — but is not tied to a real person.
Variant traversal
Every shopper traverses every variant. The same shopper sees the control and each test variant — paired comparison — so the difference in their behavior is attributable to the variant, not to noise in the cohort.
Aggregation and ranking
Squoosh aggregates the per-shopper outcomes into per-variant conversion rates, revenue per visit, drop-off rates, and ranked confidence intervals.
When should you use it?
| Synthetic shoppers | Real-customer A/B test | |
|---|---|---|
| Population source | Modeled from aggregated behavioral signals on your site | Live traffic — whoever shows up that day |
| PII / cookies required | None | Yes (for tracking + segment matching) |
| Time to a usable result | Under an hour | Days to weeks |
| Cost of a losing variant | Zero — synthetic, not real | Lost conversions on real customers |
| Statistical design | Paired (every shopper sees every variant) | Independent samples (each visitor sees one variant) |
Frequently asked questions
Are synthetic shoppers real people?
No. They’re modeled. A synthetic shopper has a buyer segment, an intent, a device, and a funnel position, but they’re a statistical construct, not a database row tied to a real human. No PII is involved.
How are they calibrated?
From aggregated behavioral signals on your site — orders, sessions, GA4 events, segment streams. Squoosh reads device mix, funnel drop-off, cart sizes, time-on-site, etc., and tunes the synthetic population to match.
How accurate are they?
On our 81%-match-rate benchmark, the cohort matched the live A/B test outcome in 81% of completed tests across e-commerce, SaaS, and fintech. Accuracy is highest when calibration data is rich (Shopify + GA4 + Segment) and lowest on brand-new pages with no historical traffic.
Why paired observations?
Every synthetic shopper traverses every variant. That makes the comparison paired — same brain, different page — which removes population-mix variance from the result. The remaining delta is attributable to the variant.
Can synthetic shoppers replace user research?
No, and we’re explicit about that. Synthetic shoppers tell you which variant is most likely to lift conversion. They don’t tell you why a customer felt confused or what new feature they want. Pair synthetic testing with qualitative research.
See it on your site
Bring a site change. We'll test it live in 30 minutes — no live traffic exposed.