Scorecards get filled from memory, hours later
Interview notes are assembled after the interview, in the gaps between interviews, from memory. Answers from the beginning are lost. Answers from the last ten minutes have the most influence. The scorecard reflects an impression rather than evidence.
That is where inconsistency enters hiring. When two interviewers assess the same candidate, they get different scorecards not because they disagree, but because they remember different parts.
The evidence has an automated step. For each criterion, the candidate's relevant statements are quoted so the interviewer can score from the words rather than from memory.
Scoring cannot be automated, and it should not be tried. Judging whether a person is suitable is a human judgment with employment consequences. A score from the scorecard would be treated as a second opinion even if it is only a summary.
How the ActionFlow runs on the canvas
Consent is the first requirement, and it is a procedure rather than a single step. Recordings are made only when the candidate has consented. This page assumes that consent has been handled properly.
A Document Upload node takes the recording. A Transcribe Audio node converts it.
An Airtable node defines the criteria for that role. Extraction follows those criteria.
When the candidate refers to a criterion, a Generate Object node returns those passages. Where the criterion is never mentioned, it returns a clear blank. That blank is useful: if an interview never referred to a must-have point, that is a process flaw.
No score, rating, or recommendation is given at any stage.
A Notion or Airtable node populates the scorecard. A Slack node notifies the interviewer that it is ready for scoring.
A Validate node makes sure each criterion has either evidence or an explicit gap marker, so a form that is only partly completed is not shown as complete.
The transcript is kept with the recording, because a passage that looks odd is most likely a transcription artifact rather than something that was actually said.
Nodes this ActionFlow uses
- Document Upload (Data): takes the recording, which exists only with the candidate's consent.
- Transcribe Audio (AI Core): converts it to text.
- Airtable (Integration): supplies the scorecard criteria for the role.
- Generate Object (AI Core): returns verbatim passages per criterion, and an explicit gap where none exist.
- Validate (Data): ensures every criterion has evidence or a gap marker before presenting the form.
- Notion (Integration): writes the populated scorecard for the interviewer to score.
- Slack (Communication): tells the interviewer the evidence is ready.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.