Introducing Digital Twins

What a Digital Twin is

One of your customers, represented digitally, built from a real long-form AI-moderated interview with that person.

One twin. One human. A specific person with their own life experience, idiosyncratic opinions, and endless nuance to dig into.

Once they're built, anyone in your org can talk to them — 1:1 for a quick question, as a group for a gut check, or across the full audience as a long-form AI interview. Available through MCP, so your team can reach them from wherever they already work.

That's the shift: your customers become something the whole company can learn from continuously, not just during the six weeks a study is in field.

What a Digital Twin isn't

Not generic synthetic data. Not a persona averaged out of a dataset frozen in time. Not a population model guessing what someone like you would say.

Averaging is where the value dies. Aggregate enough people and you get a plausible, agreeable, entirely fictional customer who never surprises you.

But, the nuance lives in individuals, so we ground each twin in one.

How they're built

1. Recruit the audience — custom-built for you, from your most important customers.

2. Interview the human — one long, training-oriented AI-moderated interview (a Grounding Interview) built to capture how someone reasons, not just what they answer.

3. Train the twin — we build a persona core on key traits: how they reason, how they weigh risk, where what they say diverges from what they do.

4. Blind test — we remove a question from training, ask the twin, measure against what the human actually said.

5. Refresh over time — re-interview to capture change; humans grade their own twin's answers. A twin is not static, it evolves, just as a human does.

Confidence and traceability

With every answer, you see:

A confidence score. The idea that all synthetic answers are equally accurate is impossible — we show you the confidence score transparently. Confidence reflects how many of that person's real answers support the response, and how much they agree.

Traceability. Open any answer to see the human evidence underneath: the spread across real answers, the context that drove the match, and the words that person actually said.

Beyond simple surveys

Most validation in this space grades twins against multiple-choice answers. But there is a fundamental issue: it’s easier for Digital Twins to correctly identify the right answer out of a group of 4, rather than reason the way a human does and answer like them in a long-form answer.

But the nuance lives in the long-form answer.

We've developed the leading training methodology and eval for digital twins, one that grades twin accuracy on long-form, open-ended answers.

And the eval isn't just a report card. What it learns goes back into how we build twins — which grounding questions actually predict someone, where an interview needs to go deeper, how confidence should be calculated. Measuring the twins is how we improve them.

We've published our thinking on all of it.

The loop

The more human research you run, the better the twins. The better the twins, the more you find that's genuinely worth asking a human.

Twins point you at the contested question. Then you go field it — with the actual people behind those twins, or a fresh audience — from the same place.

Human research isn't a cost the twins help you avoid. It's the input that makes them work.

This doesn't replace human research. It extends it.

How teams are using it

• De-risking marketing campaigns and product concepts in minutes

• Brainstorming and early discovery before committing to a full study

• The follow-up question that surfaced after fielding closed

• Building bespoke twin audiences of their most important customers

• Talking to twins from anywhere, via MCP

Want to learn more about Digital Twins? Book a demo