I came across a study recently that I've been thinking about since. Research firm Conveo ran an AI-led ethnographic study on dishwasher behavior — 91 U.S. households, two days, one-tenth the cost of traditional in-home research. That last number is the one that stopped me.
In this industry, research budgets are a constant negotiation. Traditional in-home ethnography is expensive, slow, and logistically demanding — and after all of that, most studies yield 10 to 20 households. This study returned 91, in two days, at a fraction of the price, with findings that were arguably richer than what a researcher sitting in someone's kitchen would have produced.
The Methodology
The setup was straightforward: participants filmed themselves loading their dishwasher, narrating their choices. After the cycle finished, they returned on camera to document results. An AI interviewer guided both sessions — asking follow-up questions in real time, probing for specifics, and maintaining context across visits. No observers. No lab. No travel budget.
What that enabled was something traditional fieldwork consistently struggles to deliver: people behaving naturally, on their own schedule, in their own environment, without anyone watching over their shoulder.
The Cost Case
A conventional mobile ethnography of this scope typically requires four to five months of coordination, observer training, and field execution. The economics described in this study represent a structural shift — not an incremental improvement.
The sample size alone changes what's analytically possible. At 10 to 20 households, findings are directional. At 91, patterns become statistically meaningful. That distinction matters enormously when research is being used to make product or marketing decisions.
What the Data Actually Showed
The study's findings on brand detection illustrate why this methodology produces a different quality of insight. The AI analyzed not just what participants reported using, but what appeared on camera in their actual kitchens.
| Brand | Self-Reported | Observed on Camera | Gap |
|---|---|---|---|
| Cascade | 37 mentions | 37 observations | None |
| Great Value | 4 mentions | 10 observations | +150% |
| Finish | 10 mentions | 15 observations | +50% |
Cascade showed perfect alignment — a brand with genuine top-of-mind salience. Great Value appeared on camera more than twice as often as it was mentioned — people use it regularly and don't think to bring it up, even when talking to an AI that is not capable of judging them. Finish was identified down to the specific SKU — Finish Powerball Quantum — without a single participant naming it.
"Fine means I only had to hand-dry a few things."
— Study participant, on rating their dishwasher results as "as expected"There is a meaningful distinction between brands people are consciously loyal to and brands they reach for on autopilot. The latter are used regularly but not particularly valued. They are, as a category, more replaceable than their sales figures suggest — and this kind of research is one of the few ways to surface that vulnerability before a competitor does.
The Broader Implication
The research industry has been discussing how to do more with less for years. What this study documents is a methodology that actually delivers on that — not by cutting corners, but by removing the structural inefficiencies that make traditional ethnography so expensive in the first place.
For research and insights functions operating under budget pressure, this model addresses something that has been genuinely difficult to solve: scaling qualitative depth without proportionally scaling cost. The dishwasher category is the vehicle in this study. The methodology is transferable to any category where stated behavior and observed behavior have quietly diverged — which, in practice, is most of them.
That's worth paying attention to.