A practical look at the most effective and least used method for surfacing what scientists don't know they're doing
What ethnography isEthnography is a type of market research and voice-of-customer (VOC) study where the Product Manager (PM) watches a customer do their work in their environment, instead of asking them to describe it. You watch the bench, not the interview room, long enough that the scientist stops performing for you and goes back to working normally.
A survey or an interview asks the customer to report on her own behavior. Ethnography does not, and that difference is the actual reason to use it, not a lack of budget for a bigger survey. A larger sample size gives you more confidence in what scientists say they do. It does not tell you what they actually do, because a habit, repeated daily, stops registering as information to the person performing it. The postdoc who pulls a second aliquot before every run, off protocol, will not list it on a survey. She isn't hiding it. She has stopped noticing she does it.
Pros and cons
| What it gives you | What it costs you |
|---|---|
| Surfaces habits the customer cannot articulate, because they are no longer decisions to her. | Slow. One useful visit runs half a day including travel, and one visit rarely tells you if what you saw is a pattern or a habit. |
| Rated among the most effective idea generation methods by Cooper and Edgett's study of new product development practices. | Does not scale. You cannot easily run ethnographic research on forty accounts the way you can survey forty. |
| Produces checkable physical evidence: an annotation, a workaround, an artifact the customer built, or a photo or video where the site allows it, instead of an opinion. | Photos and video depend on site approval. Many regulated environments (GMP, biopharma QC) restrict or prohibit them, same as notebooks. |
| Works even where the customer cannot name the problem, which is exactly where novel workflows and academic labs live. | Access is often restricted, and it cannot tell you whether a new-to-world product would be adopted, because there is no current behavior to watch. |
How to apply it
A PM needs to be prepared to scale down the observations based on the kind of lab access they have. Full bench access and no access at all are both rare. Most sites land somewhere in between.
| Access Level | What You Can Collect | What You Can't Get |
|---|---|---|
| Tier 1. Direct bench access. An academic core, your own beta site. | The method as performed, including steps not in the SOP. Photos or video if the site approves them. | Nothing. Take this access whenever it's offered. |
| Tier 2. Escorted walkthrough. Biopharma QC, gowned, time-limited, with an escort. | Physical evidence: instrument placement, annotations, what is taped to what, where people queue and wait. Photos or video only if pre-approved. For capital equipment, note whether the site lead checks the instrument's maintenance log herself before a purchase decision or defers to a technician. That pattern says as much about who actually owns the buying decision as an interview would. | Continuous, unsupervised observation. You see a room, not the full process. |
| Tier 3. No physical access. Competitor-sensitive site, remote customer. | Artifacts: the SOP, the run log, the error log, the maintenance log, training material the lab wrote itself. | Anything never written down, and anything you cannot see. |
Tier 3 is the one PMs abandon and the one worth defending. The highest-value document in any customer's building is the one they built themselves: a laminated quick-reference card, a homemade dilution calculator. Someone spent unpaid hours creating it because your official documentation failed. Probe to get access to these types of documents.
Three signatures of a workaround
The annotation. Sharpie on a reagent bottle, a card taped to an instrument that may circumvent part of an SOP. Every annotation marks a point where the official method proved insufficient and someone patched it.
The detour. A step performed out of sequence, or absent from the document entirely. Ask what happens if it is skipped. The answer is usually a failure mode nobody escalated to you.
The duplicate. Anything performed twice for reasons other than a built-in replicate, like a triplicate run every method calls for regardless of outcome. The signal is duplication that only happens when a specific result looks wrong: an extra aliquot pulled because the first one seemed off, a confirmatory assay run because nobody trusted what the first run showed. That kind of duplication is a scientist spending sample or hours to reduce an uncertainty that shouldn't have been there.
Do not ask what the product should do while watching. The moment you ask, the scientist stops working and starts designing, and you have converted your one source of unstated evidence into a feature request. If you are developing something with no existing equivalent, you cannot observe your future product's absence. You can only observe how the job gets done today, with whatever tools exist. Watch for how the gap is compensated for right now, a manual workaround, a spreadsheet, a competitor's tool, a shortcut nobody documented. If you observe several sites and find nothing filling that gap in any form, that is a demand problem, not an observation problem, and no site visit fixes that.
What goes in the trip report
Three categories:
1. What was performed, in order, including the step not in the SOP and pictures/video (if approved by the site).
2. What it costs in the customer's own units (minutes, aliquots, repeat runs) and what you observe in the moment, a sigh, a muttered comment, a visible pause, rather than a paraphrase you write down afterward.
3. What the customer built to cope, and where they keep it.
Catching a weak product concept before it goes into full development is far less expensive than catching it after launch, and the evidence that would catch it earliest is usually sitting on the bench, taped to something.
Book the visit before you write the product concept
Use whatever access tier you actually get rather than canceling the visit because full bench access was refused. Come back with one undocumented step, one cost in the customer's own units, and one artifact they built themselves. Then write the product concept. This is what VOC looks like when someone goes and checks.
FAQs
Quality will never let me onto the floor. What do I actually do?
Ask for the paper instead of the room: the SOP, last quarter's run log, any training material the lab wrote itself. Those three documents carry the annotation and the duplicate even when you cannot watch them happen. A QC manager who will not gown you in will usually screen-share a controlled document under an existing confidentiality agreement, because it costs her nothing.
Sales already runs customer visits. How is this different?
A commercial visit returns intent, what she says she'll do. Observation returns method, what she actually does. Neither substitutes for the other. Easiest entry point: join a visit already on the calendar and take the final twenty minutes at the bench after the commercial conversation closes.
How do I know one visit was enough?
You have enough when you can name one undocumented step, one quantified workaround, and one artifact she built herself. That is the bar for a problem statement, not a business case. A second visit tells you whether the workaround was a local habit or worth watching for elsewhere. Confirming an actual pattern takes a larger sample, twenty sites is a reasonable floor, and from there you are into qualitative or quantitative market research, not site visits.
I manage a team of six PMs. How do I get this to happen consistently, not just when one PM feels like it?
Build the three trip report categories into whatever template your team already files, a concept brief, a gate document, so filling them in is not extra work on top of an existing process. Review trip reports in your regular one-on-ones the way you would review a forecast: check for one undocumented step, one quantified cost, and one self-built artifact. Consistency comes from making the three categories the bar for a completed visit, not from mandating how many visits each PM runs.
