Frequently asked questions
How accurate are synthetic customers, and how do you validate them?
The model is one step; the grounding is the product. Every synthetic customer is built from real signal your customers already left, so its answers trace back to evidence rather than a guess. We validate against real outcomes: baselines are set before any change ships, and design-partner work is reviewed as before-and-afters rather than headline metrics.
Where does the data come from? Is this just scraped reviews?
It is more than reviews. We read the signal customers and the category already leave across reviews, support tickets, search, social, and the language people use when they switch. Reviews are mined at the aspect level for sentiment; search is clustered for demand; category leaders are read in full for friction and gaps.
How is this different from survey panels or traditional market research?
Panels ask strangers prompted questions and wait weeks. We build an interrogable customer from unprompted language your market already uses, and you question it directly. You leave with the specific move on the decision surface that matters, not a slide deck to interpret.
What about data privacy and provenance?
Every line we produce traces to real signal, and the method behind it stays ours. We work from publicly available and customer-provided signal, and design-partner names stay anonymous until each partner clears them.
What do I get, and how fast?
A presentation-grade decision, not a data dump: the specific change to make on the surface where the decision happens — a title, a page, a package, a campaign — with the customer signal that justifies it.
How is this different from a reporting-only tool?
A report is a one-shot snapshot. Action Signal gives you an ongoing, interrogable customer you can keep asking, plus an accumulating data asset that compounds as you use it. The audience is the product, not a PDF.