Kapari helps you decide better.
Before you announce a plan, see who will back you, who will push back, and why.
No card needed; access opens by email.
- 1You describe your plan.
- 2The AI runs it past fictional people.
- 3You see how they react, before you announce.
Your plan
- Leadership: for
- Partners: for
- Customers: split
- Suppliers: split
- Employees: against
- Managers: leaning against, the sticking point
Already using Claude or ChatGPT?
They play these characters when you ask them to. At Kapari, the AI writes what the people say; the conclusion comes from a fixed rule.
Kapari or Claude?Where do these people come from?
For every decision, Kapari builds them from what you describe, from its library of 543 sourced profile types (group portraits, each backed by public sources) when the topic is covered, from research in psychology and the social sciences, and from public statistics. No real person is interviewed.
And your plan: who will back it?
Describe it in one sentence and watch the network take shape. To see the reactions, you’ll need an account: 30-day trial, no card needed.
The reactions are written by Mistral AI, in the European Union; a few checks go through US services. Details: the Security page.
Kapari in action.
You describe your plan; Kapari runs it past fictional people and shows you how they react, down to the objection nobody saw coming.
Every decision rests on assumptions. Test them before you commit.
Customers will follow, the team will buy in, the price will hold: behind every decision sit bets nobody has checked. Kapari puts them to the test, in front of the people who will receive it.
What if it had already failed?
Psychologist Gary Klein suggests assuming the decision has already failed, then working out why. Kapari runs the exercise: the fictional people look back from a year out, tell what went wrong, and name the warning signs to watch from day one.
What if you did it differently?
What if you announced it in two steps? What if the increase were smaller? Run each variant past the same people and see what shifts in their reactions.
Four steps, a few minutes.
Your fictional people ready in minutes, a reading the same day, the adjustment right after. You describe, you choose who reacts, you question them, you read their reactions. The AI walks you through from start to finish.
Frame the context and the decision
State your situation and the decision to confront, in a few words. The AI rephrases, clarifies the stakes and suggests the points of tension to test.
Choose who reacts
Choose the groups that matter to you, and how much each one counts. The AI suggests the right groups and turns them into fictional people, built on real data.
They debate it
Run the simulation. Each voice reacts with its feeling and its arguments, then the debate plays out over several rounds: voices hear each other, answer each other, some switch sides. At the end, you see who buys in, where opinions split and which objections come back.
Read the reactions and the next steps
You get the map of reactions: where it lands, where it sticks, and why. The AI sums up where opinions split and suggests ways to adjust.
Variety over headcount. Very different fictional people cover angles an eight-person focus group leaves in the dark.
In plain terms. One AI builds fictional people from public data, another makes them react to your decision. You get plausible reactions to decide before you announce; no real person is ever interviewed.
Your context, your people.
Kapari is not a generic audience. You choose the groups that matter to you, and how much each one counts.
Far more than a reaction.
You do not just read how it lands. You question the voices, compare your options, rewrite, spot the risk, and walk away with a deliverable.
Compare up to 4 options
Test several versions of a decision or a message on the same panel, and see which one rallies, and where each one sticks, family by family.
Rewrite in one click
From the objections that surface, Kapari suggests a version that defuses them. You leave with the talking points already drafted.
Test a visual
A poster, a campaign visual, a slide: the panel reacts to what it sees, not only to a text.
Up to one hundred voices
Compose the panel that really matters: dozens of contrasting profiles, grounded in data, never a generic audience.
The risk profile
Before you announce, a risk scan dimension by dimension: internal climate, operations, reputation, buy-in.
A deliverable, not a screen
Export the reading as a branded PDF or an Excel workbook with charts, and share a read-only, revocable link that never exposes your context.
Ask them why
A reaction intrigues you? Question the voice: it answers in character, faithful to its profile. And ask the whole panel a question after the deliberation.
What if one family weighed more?
After the verdict, vary each family's weight and watch the verdict recompute before your eyes, without rerunning the simulation.
A file that lives
Each decision keeps its thread: the variants put on the test bench, and the reality check you record after the announcement. Kapari holds its own sheet against what actually happened.
Walk in prepared, in front of those who will judge.
Board, leadership team, employee representatives, investors: your decision will be judged. And the first question will be whether you did your homework.
The objections, already heard
You walk into the room having already heard the criticisms waiting for you, answers ready. The question meant to destabilize you becomes the one you were expecting.
A document to put on the table
The PDF report, with its transparency notice, written to be shown around you, the Excel export with charts, and a read-only share link you can revoke at any time. Proof that the work was done.
A public method, with published grades
The method behind your file is published and graded on real past decisions, good and bad report cards included. You are not defending a black box.
And when you need to explain it around you: the exact wording to announce it to your people.
We hid the end of the story from it. Its job: find it again.
It is not a poll and it predicts nothing: so we grade it. 100 real decisions by companies and institutions, in the US and France, from the Bud Light backlash to the Reddit blackout, submitted to Kapari without their ending. The answer key: 421 objections actually voiced at the time, verified one by one in dated articles. On average, it recovers 77% of them, more than three out of four.
It surfaced the retail investors' betrayal on its own
Robinhood halting GameStop purchases, 2021: Kapari recovers 100 percent of the objections actually voiced, from the broken democratization promise to the class actions filed the same day. Its advice on decisions received this way: do not expose them as is.
It does not say no by reflex
Airbnb's 2020 layoffs and Stripe's in 2022, both praised as textbook cases: on the test bench they come out tense but manageable, objections to defuse, never a red light. Scored against how things actually turned out, both fall within the accepted margin, not exactly right. Bud Light's 2023 misfire gets a firmer call, hold off, and that one was exactly right. When it gives a green light to a goal that was then missed, as with the CFPB cap on overdraft fees, that is published too.
The exam sheet we could have hidden
Balenciaga's 2022 holiday campaign, replayed in the French half of the exam: its worst coverage score, 13 percent of the documented objections recovered on the latest exam. The grade is published as is, on the same page as the others. Seeing that one is what lets you trust the rest.
Every version retakes the exam
New setting, new knowledge, new model: before any engine update ships, reference decisions are replayed. If one of them changes its advice, the update waits.
The more you use it, the better it knows you.
Your decisions don't live in a heap. File them by client or engagement, give each folder its context, and Kapari reuses it on every simulation in that folder. A consultant juggling several clients keeps distinct contexts and histories, and never starts from scratch.
One folder per client
Sort your Kaps into folders and sub-folders, by client, by engagement, by topic. Each thread stays separate, findable, yours.
Context that compounds
The client's sector, their constraints, the people they must win over every time: describe them once at the folder level. That context enriches every new simulation. The more you feed it, the closer the panel fits that specific client.
The real world loops back
After the announcement, note what actually happened. The folder keeps that feedback next to the simulation, to hold Kapari's take against the facts, decision after decision.
Profiles grounded in the real.
Each voice draws its depth from public data and published research. The data feeds the composition, it never becomes a numbered verdict.
Public statistics
Profiles calibrated on US Census Bureau, Bureau of Labor Statistics and Gallup series: population structure, occupations, telework, ideology, stock ownership. Every figure checked against its primary source, and dated.
Social science
Published work in sociology, social psychology and behavioral economics gives each voice its priorities, reflexes and friction points: psychological contract breach, loss aversion, betrayal aversion, moral foundations. Voice texture is written by Mistral Large.
Deep contexts
Each voice backs what it says with public references, checked one by one before they enter the base. Credibility comes from the sources, honesty comes from the output.
Profiles calibrated on public data, built to compose the panel, not to produce a score. Serious sources guarantee the raw material; the published exam checks what Kapari does with it.
And if you are wondering what Kapari does that a general-purpose chatbot does not, the answer fits on one page, side-by-side case included: Kapari vs. a chatbot.
A software company raises prices 30 percent. It sees the split before the email goes out.
A 40-person software company in Denver sells scheduling software to small service businesses. On January 1 it raises every plan 30 percent and shuts down the free tier: roughly 18,000 accounts that bring in about 60 percent of new signups, 4 percent of revenue, and most of the support load. The company has 14 months of runway and has not touched its prices in four years. Kapari put that announcement in front of a panel of 66 simulated voices, inside the company and outside it. The test bench came back split down the middle: 40 percent leaned favorable, 20 percent came back mixed, 40 percent leaned opposed, and reception risk read high. The useful part is what the two ends argue about. Every free-tier voice sits at minus 75 or lower, and none of them argue about the 30 percent. They say a promise was broken.
The split, visible before the email goes out: who leans which way, and the broken promise to defuse first. Plausible reactions to adjust before you hit send; the real market stays the real market, and Kapari means you never walk in blind.
66 simulated voices, anonymous profiles
Your data, our discipline.
The same honesty as on our results: we describe the chain as it is, not as it would look good in a brochure.
Walled off per account
Your Kaps, folders and contexts are accessible only to you, protected at two levels: your signed session and an identity filter on every database access. A Kap is never visible to a third party, unless you create a share link yourself, and that link then strips the private context you had entered.
Minimization first
Kapari never sends an identifiable individual to the model. Panel voices are anonymous sociological profiles calibrated on public data, never real persons.
Powered by Mistral
Voice text is generated by Mistral Large, the frontier model of Europe's leading AI lab, with inference executed on Mistral's infrastructure in the European Union.
No training
Your content is never used to train a model, neither ours nor our providers'.
The method does not depend on the model
The AI writes the voices' texture. All the aggregation, Kapari's core, stays deterministic and stays with us.
Since September 18, 2026, the model call goes straight to Mistral AI, in the European Union, with no American intermediary. A few checks and public-fact searches go through US services: the Security page lists each one. We describe the chain as it is.
A compass, not a crystal ball.
The panel composes credible reactions to make you think, spot the blind spots and choose what to test for real. Kapari informs the decision, it does not make it: it does not predict the future and does not replace field research. That transparency is exactly what makes the grade credible.
"We do not promise you will make the best decision. We do promise that when you make it, you will have the most complete picture we can give you."
This promise is not talk: it is documented, exam sheet by exam sheet. One more thing: Kapari reads how a decision is received, not how it turns out. Netflix's password crackdown was received badly, then won. That is also why, scored against real outcomes, the French-market verdict on Netflix counts as a wrong-way call, and it is published as one.
What if you were the CEO?
Announce a company decision and watch a panel react, voice by voice, as the reactions sharpen. The public, playable demonstration of what Kapari does for your decisions.
Decide without blind spots.
Request access and run your first decision on the test bench. Kapari has launched: a 30-day trial at the Access volume, with no counter on screen, free and with no card. A work email opens your account in one confirmation click; otherwise, Pierre answers himself. The price list is public on the pricing page.
The Kapari that took this exam is exactly the one that will study your decision.