Every conversation is read afterwards for what it taught. Almost nothing survives that reading, and the part that does has been corroborated across separate people, separate sessions and a recent window — because a system that believes what it hears once is confidently wrong at scale.
What was said, what they were trying to do, and where they got stuck. Notes, not conclusions — nothing is decided from one conversation.
Four things are thrown away before anything is kept — and what was thrown away is recorded too.
So a quiet conversation and a conversation that went wrong are never mistaken for each other. Your team can see which of the two it was, and so can we.
A figure read off one customer's dashboard is their data, not a fact about your product. It is never learned as one.
“Sorry, can you say that again.” A conversation about the conversation teaches nothing about the software.
Anything Amvio itself said. A system that learns from its own output drifts, and it drifts confidently.
A claim that nothing in the session actually backs up. If the evidence is not there, neither is the observation.
Not observations about your users and not a report on sentiment. Statements about your product, of a kind that could be printed in your documentation tomorrow.
“Member invites are under Settings → Team, not Settings → Workspace.”
The single most common class of question in most products, and almost never in the documentation.
“A scheduled export keeps running after the person who created it is deactivated.”
Behaviour nobody wrote down because it was obvious to whoever built it.
“Trailing whitespace in the carrier column causes an import row to fail silently.”
Not a complaint and not a summary of complaints — a specific, checkable thing about the software.
Each one is a reason to throw a claim away. The useful output of all this is how little comes out of the end of it.
Nothing is concluded from a single session. It waits until there is a body of evidence to look across, because one person's afternoon is not a pattern.
What comes out has to be a checkable statement about your product — what a thing is called, where a function lives, what a rule is. Never a summary of how conversations went.
If it merely restates something already in your knowledge, it is dropped. Otherwise the same fact ends up in two places at two levels of trust, and there is no single answer to what Amvio believes.
Corroborated across separate conversations with separate people. One customer saying something twice in the same session is one data point, and it is treated as one.
Measured from when the evidence happened, not from when the record was touched. A pattern nobody has run into for a long time may simply have been fixed.
The bars show the shape of it, not a measurement. Amvio has no production numbers to show you and will not invent any — and a page arguing for evidence is the last place to start.
Not a summary of conversations and not a report on how things are going. A specific claim about your product, with the evidence attached to it.
Member invites live under Settings → Team, not Settings → Workspace.
What sort of thing this is — where something lives, what it is called, how a task is done, what a rule is, where people get stuck.
How strongly the evidence supports it. It rises as more conversations confirm it, and nothing quietly nudges it up on its own.
Separate sessions, with separate people. This is the number that had to be reached, not how many times one person said it.
The individual things people said that support it. “Why does Amvio believe this” has an answer you can open and read.
The last time somebody's conversation backed it up. A claim nobody has run into for months stops outranking one confirmed this week.
Which is the part that makes the rest of this page comfortable rather than alarming. It proposes; somebody on your side decides; the queue is short because almost nothing survives the tests.
One sentence, with the conversations behind it attached. You read the claim, and if you want to know why, you open the evidence and read what people actually said.
Accepting puts it into what Amvio can answer from. Declining keeps it on file so you can see what was proposed, and it never gets used.
Because the tests run first. You are reading the handful that survived, not everything anybody said this month — which is the difference between a review queue and a second job.
You ship a release, the limit moves, the menu is renamed. Newer evidence can replace a belief outright, and the replaced one keeps a pointer to what replaced it, so the history is readable.
Retirement is the honest shape for “this used to be true”. Decay is not — a fact that changed did not become gradually less true.
Which is also why nothing here silently lowers a number until a stale claim stops winning. It either still holds, or it is moved out of the way with a record of why.
Everything it works out this quarter is still there next year. That is the only asset in your support stack that improves without anybody being assigned to it.
Ramps for months, gets genuinely good, and then leaves — taking every undocumented thing they learned with them. The next hire starts at the bottom of the same curve.
Learns more slowly than a good hire at first, and never gives any of it back. Nothing it worked out in month three has to be worked out again in month thirty.
The shape, not a measurement — the vertical axis has no number on it on purpose. The reset is the only part being claimed, and it is a fact about employment rather than about anybody’s ability.
Everything it works out this month is still there next year, and it is the only asset in your support stack that improves without anybody being assigned to it.
A map with nobody to ask. A conversation with nothing to draw on. A memory of a product it never learned. The system is the point.
A conversation inside the product, on the screen they are stuck on — spoken, and able to see what they are looking at.
read the page →Where it stops. It says it does not know, refuses the three subjects you never want answered, and promises nothing it cannot keep.
read the page →It walks your product screen by screen and writes down what it found, in a 22-section record per page — including what it could not establish.
read the page →Your maps and docs compiled into things it knows, held at two grains so a single fact inside a long document can still be found.
read the page →It remembers the company, not the seat — so the fourth person from an account does not start the conversation over.
read the page →Every conversation is read afterwards for what it taught, and only claims that survive four gates become something it believes.
you are hereWhat it handled, what it learned, what it could not answer, and which accounts have gone quiet — for the people who own the product.
read the page →Where you set what it is, how it speaks, what it must never touch, and which pages it appears on. Settings that become instructions.
read the page →Your users are already having the conversation. Right now it is with nobody.