A reorg. A change to benefits. A new return-to-office policy. When an announcement derails, it derails on a Monday morning, in the break room, over one sentence nobody saw coming. Kapari lets you hear that sentence first. A panel of simulated voices reacts to your decision: you see who buys in, who pushes back and where the blind spot sits, while an adjustment still costs nothing but a rewrite. No real person is interviewed.
The question that was meant to catch you off guard in the meeting, you have already heard it, cold, in your office, with your answer ready. That is what Kapari gives you, and it does it without going anywhere near your workforce. No employee is interviewed, no questionnaire goes out, no employee data is collected or processed. The panel is made of simulated voices built from public data. A tool that claimed to know what your employees think, so you could skip talking to them, would be indefensible; Kapari is built the other way around.
Our position fits in one sentence: Kapari prepares the conversation with your employees, it never replaces it. Whatever process you owe your people stays exactly where it was. Notice requirements on a layoff, union obligations where a contract applies, straight talk with your managers: none of that moves. Here is what changes. You walk into the room having already heard the likely objections. You have worked your answers. You know what your announcement left in the dark. Pressure-testing a conversation is not dodging it. It means the hard question lands and you have a real answer for it.
A draft policy, a set of talking points, the deck you plan to show the leadership team: attach a file (.txt, .docx, .pdf) and the panel works from your facts. The text is pulled out locally, in your browser. The file itself never reaches our servers, and only the text you approve goes into the simulation. The panel reasons on what is in your document, never on opinions collected from your workforce.
The engine and the database run in Paris, on European infrastructure. Your decisions, folders and context stay there.
Your decision text is processed by a model from Mistral AI, a French company. Never your identity, never your credentials: only the text the simulation needs.
We train no model on your decisions. Your content produces your result, and nothing else.
Close your account and everything goes with it: decisions, folders, panels, context. No courtesy copy kept on the side.
Every one of these claims is spelled out, and checked against the infrastructure we actually run, on the Security and sovereignty page.
A machine-generated "employee approval rate" would not survive its first meeting. Someone would wave it around, someone else would ask where it came from, and both would be right. So Kapari does not produce one. On sensitive topics, the read stays qualitative. You get arguments, fault lines, and computed signals: the voice that breaks with its own side, the common ground between two camps, the empty chair at the table. Never an opinion score. The panel always shows how it was built, and we explain in public, on Are synthetic audiences accurate?, why we refuse to put a number on opinion. That leaves you a process you can document and defend.
A transformation plan lives or dies on how it is announced as much as on what it contains. The test bench gives you material to work on both, before the day.
The panel imagines your announcement failing: where the plan derails, who checks out first, what sets the whole thing off. You deal with those scenarios while they are still on paper.
Run several framings of the same decision past the same panel: the one that lands, the one that backfires, and why, group by group.
A reaction does not add up? Ask that voice up to three follow-up questions, or put one question to the whole panel. How the panel responds, graded on five steps, is replayed across three passes, so you can see whether it holds.
PDF and Excel exports, a read-only share link you can revoke for your CEO or the leadership team, and a folder library where you log what actually happened after the announcement.
Access is free on request, no credit card, with an answer within two business days, and the demo needs no sign-up: you can see for yourself in two minutes. Kapari is in beta, and we would rather say so than dress it up.
What we publish speaks for us: an exam on 16 well-known US decisions, replayed without their ending. Several are employer calls aimed at their own staff: the Airbnb and Stripe layoffs, Coinbase's ban on internal political debate. The answer key is 48 objections people actually voiced at the time, verified one by one. The panel recovered 78% of them on average, and the worst sheet, 33%, is published as is on the exam page. Put a real HR decision on the test bench during the beta and you will see what it is worth on your own ground before you commit to anything. We have not published a dedicated HR case yet; the first one may well be written at your company.
Kapari assembles a panel of simulated voices and has them react to your announcement before the day comes: who buys in, who pushes back, and where the blind spot sits. No one is interviewed and nothing is predicted: that transparency is exactly what makes the read defensible in front of your people. It prepares the conversation with your employees, and with their union reps where there are any; it never substitutes for it.
Try the demo, no sign-up, or put a real HR decision on the test bench during the beta. No real person interviewed, no employee data processed.
And when the day comes: how to talk about it with your people ->
Kapari measures its audience with Google Analytics, to know how many people read these pages. Nothing is stored on your device until you accept.