An interview that survives a dropped call

Calls drop

A voice interviewruns live in the browser, over the candidate’s own connection, from wherever they happen to be, quite often in a borrowed meeting room or on the walk back from one. Some of those calls will drop. Not many, but the number is never zero, and any product that behaves as though it is will eventually punish somebody for standing in the wrong place.

The interesting question is not how to stop it happening, because you cannot. It is what the system does in the thirty seconds afterwards, while the candidate is looking at a screen that says the session has ended and wondering what has just happened to their application.

A dropped call does not cost the slot

The first rule is that losing the connection does not lose the opportunity. The interview is something the candidate was invited to do. A network failure is not a decision by either side to withdraw that invitation, and treating it as one is simply a penalty applied to whoever has the worst coverage.

So the candidate whose call drops is still a candidate with an interview to do, rather than a candidate who has used up their chance. That distinction costs nothing to honour and it decides how the whole process feels from the other end of the line.

Half an interview is never passed off as a whole one

A dropped interview can be resumed from where it stopped, within a day. The candidate gets one retry, and unlimited retries when the fault is ours. Nobody starts again from the first question.

The second rule protects you rather than the candidate. If the interview is never resumed, what was said is still graded, but it is marked as cut short. A fragment scored against a full rubric can produce a report that looks exactly like a good one, so it is never passed off as a full interview. When you see a completed interview against a candidate, the interview finished.

Finishing is recorded once, however often it is reported

This next part is invisible when it works, which is most of the time. The end of a session can be reported more than once: the same ending arriving by more than one route, or a report sent again because the first was not acknowledged. Anything that depends on a live connection has to expect this.

Completion is therefore written exactly once, no matter how many times the ending turns up. Not two interview records for one conversation. Not a second scoring pass over the same transcript. Not a candidate appearing twice in the same stage of your pipeline. The first report settles it, and the ones that follow change nothing.

An ending reported three times is still one interview. A candidate should never be able to tell how many times the network told us it was over.

And the sessions that never end at all

There is a third case, quieter than the other two. A session neither finishes nor reports a failure: the candidate’s battery went, the tab was closed, something upstream stopped speaking. Nothing arrives to say the interview ended, because nothing ended it.

Left alone these sit open indefinitely. They hold a candidate in a stage that stopped being true days ago, and they are why a pipeline review turns up somebody who has apparently been halfway through an interview since a fortnight last Tuesday. Stalled sessions are cleaned up rather than left hanging, so what the record says about a candidate is what is true about them now.

What this looks like from the candidate’s side

Put yourself at the other end of it. Someone has taken time out of a working day, found a quiet corner, and been talking to an automated interviewer for eleven minutes about the hardest thing they have ever shipped. The call cuts out. They have no way of knowing whether a system recorded that as an interview, as a failure, or as nothing at all.

Most people, in that position, assume the worst and say nothing. A few write in. Almost nobody enjoys any part of it, and the ones most likely to walk away are the ones with other options, which is the group the whole exercise existed to reach.

The behaviour that fixes this is not clever. The interview they were invited to is still there to be done, the fragment is not held against them, and nothing about the incident needs chasing. It is worth being explicit about because the alternative is not a rare edge case; over enough interviews it is a steady trickle of people with a bad story about your company.

The test is an email nobody has to send

All of this exists to prevent one small, miserable message: the candidate writing in to ask whether their interview counted. It is an entirely reasonable thing to want to know and a bad thing to have to ask, because asking means the system left them holding a question it should have answered itself.

It is also a fair question to put to any vendor running interviews on your behalf. Not “how reliable is it”, which invites a number, but “what happens to the candidate when it fails”, which invites a description of behaviour you can check.

What this means on your pipeline

• Completed means completed. An interview marked done is one that ran to the end, so a thin report is a thin candidate rather than a dropped connection.
• A drop is not a withdrawal. Nobody loses their place because of their signal, which matters most in exactly the markets where connections are patchy.
• Nothing stays open forever. Sessions that stall are closed out, so the stage a candidate sits in still describes where they actually are.