What Inspired me
I got interested in how companies figure out what customers actually want after seeing a friend make decent money taking paid surveys. Diving into market research, I discovered retail ethnography—where researchers observe and interview shoppers right in the store aisle as they make buying decisions.
While ethnography produces rich insights, traditional studies are broken: they cost an average of $150,000, take months to complete, rely on tiny sample sizes, and suffer from the fact that humans often can't accurately explain their own subconscious buying choices. Because of this, product teams are forced to either make decisions on thin evidence or wait too long for good data. I wanted to find a way to give teams real-world shopper insights without the massive time and budget barriers.
What It Does
Prova is an agentic AI tool that delivers retail ethnography at scale, simulating shopper decisions across hundreds of synthetic customers in hours instead of months for a fraction of the cost.
Instead of relying on abstract text prompts, Prova captures the visual reality of a physical store in four steps:
- Capture: A researcher uploads a video of a product in its real selling environment (e.g., an air fryer sitting on a retail shelf next to competitors, with pricing, packaging, and placement visible).
- Simulate: Prova deploys a swarm of hundreds of cost-efficient, multimodal AI agents—each assigned a distinct persona, like a busy parent or a broke college student—to watch the video and decide whether to buy. Each agent logs its exact internal reasoning, including what caught its eye, what caused confusion, and what it wished were different.
- Synthesize: A central orchestrator analyzes all the individual agent responses to extract broader trends on what worked and what failed.
- Deliver: The researcher receives a structured, actionable report detailing why customers buy and what needs improvement, with every insight traceable back to individual agent decisions.
Built With
- agents
- growthsignals
- velocityai
