Launching soon on the Shopify App Store
For Shopify merchants
Run your products, packaging, pricing and ads past simulated shoppers built from census data and your store's own patterns. Get a purchase intent score, a clear verdict and the reasoning behind it before you spend real money.
Purchase intent
Put to work by
How it works
Install the app and pick any product from your catalog. That's what gets tested.
Open-ended questions, like “What would stop you buying this?” testfeed blocks question types it can't answer honestly before they run.
Simulated shoppers start from US or Australian census data, shaped by your store's aggregate purchase patterns. No personal customer data is used.
A purchase intent score, a verdict, and every shopper's full answer, inside Shopify admin.
Results
Every study returns a purchase intent score with a confidence range, the level of agreement among shoppers, and every individual answer in full. The verdict is one of four bands, based on the score.
Ties are reported as ties
testfeed never manufactures a winner. If two variants perform the same, it tells you, and you can stop debating.
Pack test · Sunny Crisp Corn Flakes
N = 100
Recurring themes
“I like that it looks less sugary than the big brands. I would try a box if the price was close to what I pay now.”
Guardrails
In an industry that claims 90 percent accuracy, we are upfront about what simulated shoppers can and can't answer. testfeed blocks the questions they can't, before they run and before they cost you anything.
“How much would you pay?”
Stated willingness to pay is unreliable, so we don't ask it.
Rejected“Rate this 1 to 5.”
Asking an AI for a rating produces inflated scores. Shoppers answer in words only.
Rejected“How does it taste?”
Simulated shoppers have no senses. Taste, smell and texture are out of scope.
Rejected“Heard of our brand?”
A simulated shopper has no real memory of your brand, so awareness questions would be meaningless.
Rejected“What would stop you buying this?”
An open question about purchase barriers. This is the kind that runs well.
Passed
Validation
We have run two head-to-head comparisons against real human panels. Here are both results, including the one we got wrong.
The favorable case · Bae Juice, 2026
The miss · B2B software, 2026
Reproducibility
0%
variation across repeat runs of the same study
Most reliable question
Purchase intent
about three times better aligned with human answers than understanding-style questions
Nine test types
Would they buy it?
Does the pitch land?
Which pack wins the shelf?
Does the price hold up?
Does the ad persuade?
Do they believe you?
Does the claim convince?
Does the offer tempt?
Would they show up?
The methodology
Asking an AI to rate something 1 to 5 produces inflated scores. It's a documented failure mode. So shoppers never see a scale. They answer in their own words, one or two sentences at a time.
Each response is embedded and compared against anchor statements for every rating level, producing a probability distribution over ratings. The method is published, independent research from PyMC Labs. The score comes from the language, not from a leading question.
Shoppers start from government census microdata: 100,000 US profiles and 10,000 Australian. Your store's aggregate patterns shape that base. No personal customer data is used.
From answer to score
Method: Semantic Similarity Rating · published research · PyMC Labs, Maier et al., 2025
Data and privacy
Audiences are shaped by aggregate store signals like repeat purchase rates, spend bands, product interests and location. Individual customers are never read or modeled.
White glove service
For eight figure brands and FMCG teams, we design bespoke simulated audiences: custom populations built for your category, calibrated against your own data, with studies run for you by our team.
Book a demo and we'll run your first study with you.
Direct install from the Shopify App Store arrives in a few weeks.
