AI · listening
The decision is made in a room nobody recorded, then reconstructed from memory.
Four sets of notes that disagree, one person's recollection three days later, and a hire worth a year of somebody's salary resting on it.
A participant joins the call and records it, with consent on the file and a retention date attached. What comes back is not a summary — it is findings, each one seeking the second it happened, and the requirement nobody thought to raise.
- Every finding plays its moment
- Unverifiable findings are dropped
- Coached against your own team
Coaching · against your own median
- Talk ratioyou 62% · team 48%
- Questions askedyou 4 · team 9
- Next step agreedyou 55% · team 81%
Measured against what your team actually does, not a benchmark from somebody else's customers. Every bar opens into the calls beneath it.
The problem, specifically
Notes record what was said. They cannot record what was missed.
Below is one round, returned. Three of the rows are things that happened and can be played back. The fourth is the one no set of notes has ever contained: a requirement the role called essential that nobody raised in thirty-two minutes.
That row is invisible to every scorecard product in this category, because they collect opinions about the hour rather than a record of it.
What the hour established
round 2 · 32:10
- System design06:20
Walked the payments split end to end, and named the failure mode
- Ownership14:02
Took the migration over when the original lead left
- People management22:41
“I ran the on-call rota” — trailed off, and nobody followed up
- Regulated domains—
Marked essential on the role. Never raised in the hour at all.
The panel, checked
Two people scored the same hour differently. That is the useful part.
Every scorecard product collects ratings and almost none of them do anything with the disagreement — which is the single most informative thing in a debrief and the reason debriefs run long.
Each panellist's rating sits beside what the interview itself produced, competency by competency, and the rows where they are furthest apart are marked before the debrief rather than argued through during it.
Interview evidence
Claims, risks and half-answered questions, each seeking the second it happened. Nothing that cannot be verified against the recording appears at all.
Prep briefs
What to probe in the next round, drawn from what the last one failed to establish — so the gap above does not survive to round three.
Consent and retention
Consent captured on the file, a retention date attached, and deletion that actually runs when it arrives.
How the hour is normally lost
Three failures that a scorecard cannot see.
None of these is a failure of note-taking. They are failures of the format — what a set of notes is structurally unable to hold.
The confident recollection
Notes are written afterwards, by which point the hour has been compressed into whichever moment was most memorable — usually the last one.
A candidate is remembered as “vague on ownership” because the weakest answer came at minute thirty, after four strong ones.
The unasked question
A scorecard records a rating for every competency, including the ones nobody raised. Absence and weakness produce the same low number.
A requirement marked essential is scored two out of five by both panellists, neither of whom asked about it at all.
The disagreement nobody reconciles
Two panellists file different ratings and the debrief averages them, because there is nothing to adjudicate against.
Four and two on the same competency becomes a three, and the reason one of them was wrong is never established.
The same hour, on the sales side
A call is evidence too, and a forecast built from it is a different number.
Objections nobody answered, commitments somebody made, competitors weighed twice — each with the moment it was said. What changes is not the quality of the note-taking. It is that the deal stops being described by the person with the most to lose from describing it plainly.
- Coached against your own medians —
Talk ratio, questions asked, next step agreed — measured against what your team actually does rather than a benchmark drawn from somebody else's customers. A median from six reps is more useful than a number from six thousand strangers.
- Risk surfaced on the deal —
What was said that should worry somebody appears on the record it concerns, rather than in a report a manager opens on a Friday.
- Follow-ups from the commitments —
A draft built from what was actually agreed on the call, not a template with the customer's name merged into it.
What you receive
From one hour, five things.
- Findings, each seeking its moment —
A claim the candidate made, a risk worth knowing, a question answered only halfway — each one playing the second it happened rather than summarising it.
- The panel beside the evidence —
Each panellist's rating against what the hour itself produced, with the rows furthest apart marked before anybody opens the debrief.
- What was never established —
Requirements the role called essential that the hour did not touch, stated as absent rather than scored low.
- Coaching against your own medians —
Behaviour rather than outcome — questions asked, talk ratio, whether a next step was agreed — measured against your team, not a benchmark drawn from strangers.
- A consent and retention record —
Captured on the file, with a date attached to the recording and deletion that runs when it arrives.
Objections first
What gets asked once somebody realises this records people.
›Does it conduct the interview?
No. It joins as a named participant, records, and reports. The product that conducts interview rounds is Cognisense, a companion product from CognityHire with no integration into this platform. A recorder and an analyst of a human interview is a different thing, and a demo exposes the difference immediately.
›Can we read the full transcript?
The recording is transcribed and every finding is verified against it, but what you open is the finding and the moment it came from rather than a wall of text. The check that matters is hearing it said.
›What about consent?
Captured on the file, with a retention date attached to the recording rather than to a policy document. Deletion runs when the date arrives.
›Will interviewers object to being measured?
Coaching is per rep and per interviewer, against your own team's medians, and it is about behaviour rather than outcome — questions asked, whether a next step was agreed. Teams tend to object to rankings, not to being shown the question they forgot to ask.
›What happens if the audio is poor?
Findings that cannot be verified against the recording are dropped. A bad recording produces fewer findings, never invented ones.
Book a demo
Make us play the moment.
Bring a recording of your own if you have one. The test is whether you can check every claim it makes.
- Open a finding and play the second it came from
- Show me a requirement the interview never touched
- Show me a finding that was dropped for not being verifiable
- Open a rep's coaching and show me the median it is measured against