The instrument
The instrument
One AI ethnographer, a library of topic packs, and a memory that accumulates. What the lab actually operates.
Most organizations know how their people feel in aggregate and have no idea why. The aggregate arrives as a score, and the reason the score moved stays with the people who could have explained it. The instrument exists to close that gap: it holds a real conversation with every person in scope, and reports what it learns at the level of the cohort.
Riley, the AI ethnographer
Riley is the AI interviewer built by Latent Variables. It conducts a voice conversation of about ten minutes with one person at a time. It is not a chatbot fielding a questionnaire and not a survey with a synthesized voice reading the items aloud. It opens with rapport, follows what the person actually says, and presses for specific recent incidents rather than general sentiment, because an account of one concrete Tuesday is worth more than an agreement rating.
Riley works from topics, never from a script. Two people asked about the same rollout will be asked different follow-up questions, because the follow-up is chosen from what they just said. That is what makes the method ethnographic rather than psychometric, and it is also what makes it scale: the same instrument can hold five hundred of these conversations in the time a research team would need to schedule twenty. It also means the read is not capped by what you thought to ask. When people keep raising something the topics never covered, that comes back to you as a finding rather than being dropped, and it is often the most useful thing in the engagement.
The topics are set before anyone is interviewed, and that set is called a topic pack: what the conversation covers, and the categories the answers come back in. Nothing about your industry is built into the instrument itself, which is why the same interviewer can work a plant floor, a hospital, a public agency, or a claims operation.
How something is said is treated as part of what was said. Hesitation, hedging, and flatness where enthusiasm is expected are signal, and the gap between what a cohort says and how it sounds is one of the findings the instrument reports. It cannot judge, retaliate, or gossip, which is the point: the social cost of saying the true thing is most of why the true thing never reaches a leader.
The longitudinal read
A single round of interviews is a photograph. The instrument is built for the film. Conversations are distilled into a memory that persists between rounds, so the second read knows what the first one heard, and a claim made in March can be checked against what the same cohort says in September.
This is what makes the instrument a standing read rather than a study. It can tell a leader that a workaround they were told had been retired is still in use, that a group which was cautious about a rollout has quietly given up on it, or that the person everything routes through has changed and nobody updated the chart. Those findings are only available to an instrument that remembers, which is where the lab takes its name: the thing that explains the behavior is never the thing being measured.
How an engagement runs
- 01Pick the topicsChoose a topic pack, edit it, or author a new one: a rollout, a merger, attrition, a set of behaviors. Topics, never a script; the instrument prepares the conversation.
- 02The ethnographer conversesTen minutes per person; invitations, scheduling, and reminders are handled. It opens with rapport, adapts, and asks about specific recent incidents.
- 03Intelligence returnsCohort-level findings, never traceable to a person: a queryable picture of the real organization, and the assets to act on it, on demand.
What the instrument will not do
People speak candidly when the terms are clear and enforced. These are the terms.
- REPORTING FLOOR
- Findings are reported at cohort level only, and a cohort has to clear a minimum number of voices before it is reported at all. The floor lives in code, not in a setting someone can turn off.
- SEALED TRANSCRIPTS
- An individual transcript is not a deliverable. Employers receive patterns across a cohort, not a named person's interview.
- BARRED USES
- Inferences from these conversations are barred by contract from decisions about an individual's compensation, promotion, or discipline.
- ALWAYS DISCLOSED
- Participants are told they are speaking with an AI interviewer, what the conversation is for, and who will see what.
The contractual language is in the terms and privacy policy.