What does Listen Labs do besides simulation?

Per Salesforce, its agents design studies, recruit from a global network of more than 50 million participants, interview around the clock in more than 120 languages, and synthesize the findings, taking research from months to days. That is real research with real people, and it is the half that produces evidence.

Why would a behavior-grounded twin miss a bad reaction?

Because the behavior it learned from was recorded in a product that did not have the feature. My onboarding agent's stuck points were accurate. The problem was how it felt to be watched and helped, and that reaction only existed after the feature did. Three of my ten failed agents failed this way: correct, and beside the point.

What rule keeps simulated answers out of decisions?

Tag every row in the evidence log as observed, said by a customer, or simulated. Simulated rows are allowed to change a question or a test plan. They are not allowed to justify a roadmap decision.

How big does the real cohort need to be?

Small. The point is not statistical power, it is hearing a reaction the simulation could not produce. I prototype an intervention with a small group and ask about perception before scaling, which is the step that would have caught the onboarding agent before it cost a quarter.

What Can a Customer Digital Twin Tell You Before Launch

What can a customer digital twin tell you before launch?

THE SHORT ANSWER

It can tell you whether your instrument works: where an interview guide leads the witness, which edge cases break a flow, and how a segment you cannot recruit might move through a screen. It cannot tell you how people will feel about something that does not exist yet, because it is grounded in behavior recorded before the feature shipped. My onboarding agent found real stuck points in real first sessions and offered help at exactly those moments; users read it as surveillance, and activation came in below the control group for a quarter. Salesforce's own description of Listen Labs' digital twins places them before further customer or user testing. Keep that order: simulate, then a small real cohort, and measure perception next to behavior.

Salesforce agreed this week to acquire Listen Labs, a research platform that also sells simulated customers, and The Twin Would Have Shipped It is my case for where simulation belongs in a research plan. The example underneath it, an onboarding agent that was right about every stuck point and still lowered activation, is number seven in 10 AI Agents I Built That Failed. The Honest Retrospective. For products whose customer is itself software, the honest form of simulation is scripted agent sessions against your own endpoints, covered in Customer Discovery When Your Customer Is an Agent.

It belongs to the AI Product Management argument: running a study got cheap, and knowing which answers count as evidence did not.

Last reviewed 2026-09-30 · 1 min read

What can a customer digital twin tell you before launch?

SOURCES

THE LONG VERSION

RELATED ANSWERS