Research talks to twelve people who agreed to a call. Amvio talks to everyone who was confused, at the moment they were confused, and none of them were performing for a researcher.
Users read ‘archive’ as ‘delete’ and avoid it.
Admins look for SSO under Team, not Security.
Nobody understands what the sharing toggle affects.
“Export is too slow to be useful.” — fixed in 4.2
users in a typical study, all of whom volunteered, which is its own bias.
the conversations a year of an ordinary product holds — from people who were not performing for a researcher.
recruiting, scheduling or incentive budget.
Analytics tells you they abandoned the screen. This tells you they abandoned it because the field was called something they did not recognise.
Nobody opted in, nobody was paid, nobody knew a researcher was listening. This is the behaviour, not the report of the behaviour.
Forty phrasings of the same confusion collapse into one finding, with all forty transcripts underneath it.
Every belief carries confidence and evidence count, and gets dropped when the product changes and the evidence stops.
The same account, over eleven months. What they struggled with in March and what they never came back to.
Two of these are capabilities. The third is a limit, and it is the one that decides whether the first two are worth anything.
questions with a transcript behind them
What they said, in their words
The phrasing, not a category someone assigned afterwards.
How many hit the same thing
Every belief carries an evidence count and a confidence score.
on the record, without being asked
Groups forty phrasings into one finding
With all forty transcripts kept underneath, so you can audit the grouping.
Retires beliefs when the product changes
Findings expire. This is the part a research repository never does.
and will keep refusing
Claiming causation from volume
It reports what was said and how often, not why your metric moved.
Replacing your interviews
It tells you who to call and what to ask. The call is still yours.
Six research problems, each solved by the same fact: it was there when the thing happened, and nobody had to be recruited.
Twelve users, all of whom volunteered.
Everyone who was confused is in the record, and none of them opted in to be studied.
People describe what they think they did.
This is what they said while doing it, with the screen and the moment attached.
Users perform for researchers.
Nobody was performing. They were trying to get a file to import at eleven at night.
Analytics shows the drop, never the reason.
The reason is in the sentence they typed before they left.
The same account, over eleven months.
One continuous record per company, so you can see what they gave up on and never returned to.
Which twelve people should you actually call?
It tells you who hit the problem, how often, and what they said — so the interviews you do run are aimed.
One person calls archive “delete”.
logged · 1 session
Forty more phrase the same confusion differently.
grouped · 41 sessions
It becomes a scored finding your team can act on.
0.88 · 214 sessions
You rename the control in 4.2.
watched · post-release
The confusion stops appearing. Evidence dries up.
retired · kept in the record
A research repository accumulates. This one expires, which is the only way a body of findings stays true to a product that keeps changing.
Nothing here is done badly four times a year. It is done well four times a year, because recruiting and scheduling allow four. The honest variable is the gap between them.
Run generative studies to find unmet needs.
4/yrEvery unmet need arrives as a question somebody asked.
Run evaluative studies on new flows.
4/yrPost-release, it reports whether the confusion moved or vanished.
Synthesise findings into themes.
4/yrBeliefs scored by evidence count, transcripts attached.
Keep findings current.
2/yrBeliefs retire themselves when the evidence stops.
Democratise access to customer insight.
4/yrA record anyone can read, not a deck one team owns.
Research is a sampling exercise. This one has the whole population and no incentive budget.
12 people · recruited, scheduled, paid
40,000 conversations · nobody recruited, nobody paidone square = 100
It does not make your research team smaller. It makes the twelve interviews they run this quarter aimed at the right twelve people.
It is not a replacement for talking to your customers. It is the sampling frame you never had — it tells you which twelve people to call, and what they already said.
The first phrasings arrive.
Raw, unprompted, and time-stamped.
Themes form.
Forty ways of asking one thing collapse into one finding.
Beliefs get confidence.
Each one carries an evidence count, and the ones the product outgrows retire themselves.
You have a longitudinal record.
A year of it, per company, so you can see what an account gave up on and never came back to.
people to recruit, schedule or pay
sessions excluded for being inconvenient
belief opens to the transcripts behind it
line of code to start collecting
It is unprompted behavioural evidence with the context attached, grouped by meaning and scored by volume. The transcripts are underneath every claim.
Some, but the population is everyone who was stuck rather than everyone who agreed to a call — a far larger and less filtered frame.
No. It tells you which interviews to run and what to ask, which is the part that usually goes wrong.
Identifiers and sensitive patterns are redacted before anything is written, and the record is per company rather than per person.
Confidence and evidence count on every belief, and the grouping is auditable — you can open the forty transcripts behind a theme.
The evidence base is a by-product of conversations it was already having. There is no study to run and nothing to schedule.
<script src="https://cdn.amvio.ai/a.js" data-key="acme_live_8f2c"></script>Install it once
Every conversation from that day is evidence, time-stamped and in context.
Decide who can read the record
It is a company-wide body of evidence, not a deck one team owns.
Check the beliefs after a release
The ones that stop being supported retire themselves. That is the mechanism to watch.
Not eight products with eight logins. One specialist with one memory, which is the only reason the renewal conversation knows about the onboarding one.
Forty accounts known properly, and the nine that need a person handed back.
An implementation call for the eighty percent who got a welcome email.
The three people who know the whole product, available to six thousand at once.
Tier one resolved in the product, escalated with the transcript attached.
The technical answer at 2am, during the trial, not next Tuesday.
Everyone who was confused, at the moment they were confused.
The gap list, ranked by what each missing page actually costs.
Decided seven months before the date, with time left to fix it.
Your users are already having the conversation. Right now it is with nobody.