The screening ran, the ratings generated, the shortlist went over. Three weeks later the manager is still asking for "the full list, I'll go through it myself".
That is the real failure mode of automated screening, and it is a trust problem rather than a technology one.
Send evidence, not scores
A number by itself asks the manager to trust a process they did not design and cannot inspect. A number with the sentence that produced it is checkable in five seconds.
"Communication 4/5 — explained a system-design decision clearly and unprompted" is a claim they can verify against the transcript. "Communication 4/5" is an assertion they have to accept or ignore, and busy people ignore.
Cap those reasons short — around 25 words. A paragraph per column is another transcript, and they will not read it.
Send the misses too
The single most effective trust-builder: alongside the top thirty, include five that scored badly, with the reason.
It sounds counterproductive and does the opposite. A manager who reads five rejections and agrees with all five stops worrying about what the process is throwing away. Without that, the nagging question — "what if the good one got filtered out?" — is exactly what drives them back to the full list.
Let them disagree in one click
The shortlist is a reading order, not a verdict. Make the transcript reachable from every row, so a manager who thinks a score is wrong can check immediately rather than losing confidence in all of them.
Trust survives disagreement about individual calls. It does not survive being unable to check.
Calibrate against them, not against theory
Before the first real batch, run ten calls and have the manager rank them independently. Compare with the rubric.
Where you differ, one of you has the wrong standard — and about half the time it is the rubric's prompt, which you can fix in a minute. This exercise costs an hour and is the difference between a rubric that reflects what this manager actually wants and one that reflects what you assumed.
Be honest about what it cannot see
A five-minute screening call cannot assess culture fit, and scoring it anyway is false precision that will eventually be caught — at which point every other column becomes suspect.
Score what the call observed. Say plainly what it did not. A shortlist that admits its limits gets trusted with the things it is actually good at.