Common questions

Latent Variables is an applied-research lab, spun out of the University of Michigan and Johns Hopkins, that focuses on improving organizational change management with AI. Our system interviews every person in an organization at the depth of an interview and the scale of a census, then synthesizes those conversations into a longitudinal memory of how the organization actually works.

The lab has expanded into partnerships with leading organizations across industries, supporting public and private companies as they work to better understand organizational change, operational dynamics, and workforce experience. Today, the platform conducts thousands of frontline conversations each week, providing a continuous, organization-wide view of how work is actually experienced on the ground.

A survey asks fixed questions and captures what people are willing to put on a form. The instrument holds open, anonymous conversations using indirect-elicitation methods, so it surfaces the workarounds, frustrations, and hidden dynamics people never write down, then connects them across the whole workforce.

Interviews are run entirely by an AI agent, with no human in the loop, and usually last 5 to 15 minutes. The conversation follows a structured framework while adapting its questions to each participant's responses, using predefined topic areas and conversational memory to probe where more detail is needed. It runs in the participant's preferred language where supported.

Employee data is stored on dedicated AWS infrastructure rather than shared environments, and the platform is SOC 2 and HIPAA compliant. To protect anonymity, if a team is too small to preserve anonymity, results are aggregated at a higher organizational level. Country-specific compliance can be addressed as needed, and NDAs are available on request.

Deployments typically see response rates between 55% and 75%. Outreach and follow-up reminders are handled automatically across email, text, and phone, and multilingual support, including Spanish, is available. Monetary incentives can be used if participation falls below target, and organizations are only charged for completed interviews.

Deployment needs two things: email whitelisting so invitations reach people reliably, and a roster with each participant, their role, and their preferred language. Once those are in place, the platform manages outreach, interview scheduling, and data collection automatically.

Yes. Deployments can run with organizations as small as 10 to 20 participants, and smaller proof-of-concept studies are possible. The same anonymity safeguards apply at any size: a theme needs multiple independent mentions before it is reported, and results are aggregated when necessary so no individual can be identified.

Yes. The platform is built to complement existing survey systems such as Peakon or Qualtrics, adding qualitative interview data alongside quantitative survey results. Interviews can run on their own or be used to explore the findings of a prior survey in greater depth.

A structured, queryable read of the organization: where change is taking hold and where it isn't, who actually moves opinion, where the workarounds run, and the gap between the board's view and the floor's.

The methodology draws on research in psychology, behavioral economics, and ethnography. The platform supports longitudinal benchmarking by comparing baseline and follow-up measurements over time, and findings are paired with evidence-based intervention recommendations drawn from published research and prior case studies. Cross-organization benchmarking is in development and not yet available.

The lab's second partner cohort, one hundred organizations, closes August 7. Book a conversation with the lab to learn more about the research and explore whether it may be relevant to your organization.