Back to all posts

Interview Drift: How Hiring Loops Lose Calibration Over Time and How to Fix It

Rishit Chaturvedi, CEO of Mazle AI
Rishit Chaturvedi, CEO of Mazle AI

A senior interviewer at a public tech company had a habit. When a strong candidate walked in, he would let them "seduce" him into a 40 minute career story instead of running the structured loop. He liked the conversation. The rubric got ignored. The candidate got an enthusiastic yes that did not actually map to the role.

This is interview drift. It is not malicious. It is gravitational. Every interview loop drifts toward the interviewer's personal interests, the candidate's strongest topics, and the most recent conversation pattern. Without active correction, the loop quietly becomes a different loop than the one you designed.

Industry research surfaces the math on this. After five rounds, only two thirds of the JD's required skills are well covered. Technical depth sits at 55 percent. The other 33 percent of the JD is theater. Everyone assumes it is being assessed. Nobody is actually assessing it.

The fix is not more training. Training fades. The fix is observability. You need to see drift while it is happening, not in a quarterly retrospective.

The practical version looks like this. After each interview, the system compares what the rubric said to assess against what the transcript shows was assessed. If the gap is more than 20 percent, the loop owner gets a notification. Not a punishment. A nudge. By the next loop the interviewer adjusts.

Platforms like Mazle measure drift continuously. Most loops self correct once the gap is visible. The interviewers were never trying to drift. They just could not see themselves doing it.

You cannot fix what you cannot measure. Interview drift is the most measurable failure mode in hiring.