Skip to content
Kapari
How it works
Who it's for
Executives HR Agencies Consultants
Examples Proof Security Pricing Demo Version française Log in Start your 30-day free trial
The guide

Synthetic audiences: a guide for business leaders

A synthetic audience is a group of characters created by an AI who react in place of real customers or employees. Used well, it shows you before the announcement what you would otherwise find out too late. Here is how to find your way.

What we are talking about

Hear your customers before you talk to them

You describe profiles to an AI ("a customer loyal for ten years", "a shop-floor team lead"), then put a question or a decision in front of them. Answers arrive in minutes, with no study to organize. It is ideal for spotting likely reactions before you act. For a reliable opinion percentage, you still need to ask real people.

The market

Three kinds of tools, three promises

The simulated poll

It promises an opinion percentage in minutes. Fast, but the most fragile promise.

The customer digital twin

A copy of your customers, built from your data, that you can question as often as you like. What it is worth.

The decision test

Where Kapari sits: varied voices react to your decision, so you see the objections before you announce it.

Apart: product testing

Simulated users try software before release, like MatrAIx. A different question.

Five ideas

What to know before you buy

1. Simulated answers, not surveyed customers

What it means for you: it is practice before the real game, not the real game. Specialists call it silicon sampling.

2. The flattening effect

What it means for you: an AI left to itself pulls everyone toward the average view and erases minority voices, often the ones that will cost you. A good tool makes them speak on purpose.

3. Made-up profile or documented profile

What it means for you: ask where the profiles come from. A label ("woman, 45, manager") produces clichés; a profile backed by published studies produces believable reactions.

4. Who reacts, rather than a number

What it means for you: a single percentage reassures but misleads, with gaps against a real poll of more than 23 points. Knowing who goes along, who pushes back and why lets you act. Why Kapari gives no single number.

5. A stable conclusion

What it means for you: submit the same case twice. If the conclusion changes, you are hearing the AI's mood. At Kapari, the conclusion comes from a fixed rule: same case, same result. The research behind it.

Common questions

What people ask us

What should I ask a vendor?

"Show me your results on real cases the tool had never seen." Kapari took that exam on 100 real past decisions and finds on average 77% of the criticism raised at the time. See the exam.

How many voices do you need?

Variety matters more than numbers: you need to hear every camp. Kapari brings together several dozen voices, depending on the decision. The reasoning.

What about my confidential information?

Simulated voices need no data about your customers. Just ask the vendor where the texts you hand over are stored, and whether they are used to train its AI.

For the curious: the sources
  • Argyle et al., Political Analysis, 2023: the founding paper on silicon sampling.
  • Bisbee et al., Political Analysis, 2024: a correct average can hide wrong links between answers.
  • Santurkar et al., ICML 2023; Wang, Nature Machine Intelligence, 2025: the flattening of minority views.
  • Park et al., Stanford, 2024: profiles built from simple labels are more stereotyped.
  • Verasight reports, 2025-2026: a gap of 4 to more than 23 points against a real poll.
  • Meehl, 1954; Dawes, American Psychologist, 1979: a fixed rule beats case-by-case judgment.
  • Reliability in detail: are synthetic audiences accurate?

Try it on your next announcement.

Write down what you plan to announce. In a few minutes, you see who goes along, who pushes back, and what to fix.