Every On Hold Candidate Needs a Decision Date

Open the review queue two weeks after a campus drive finishes and you can usually predict what you will find. The Strong Go rows have been actioned. The No Go rows are closed. And sitting in the middle, untouched, is a block of On Hold candidates that nobody has opened since the interviews ended.

Not rejected. Not advanced. Just parked, in the one tier that asked a person to make a call.

This is the most common operational failure I see in AI screening, and the model is not what caused it. The screening worked. It produced a verdict for every candidate and separated the confident cases from the genuinely uncertain ones. What broke afterwards is that the uncertain pile was handed to a team with no deadline attached to it, and work without a deadline loses to work with one.

What On Hold actually means after a screen

The post-call verdict set is Strong Go, Go, On Hold and No Go. On Hold sits in the middle, and it means something specific: the conversation did not produce enough signal to confidently advance or reject this person.

That is not a hedge. A screening system that never produced an uncertain verdict would be lying about how much a fifteen to twenty minute conversation can tell you. Some candidates give a strong answer to one question and a weak answer to the next. Some are hard to read on communication because the line was bad. Some clear the bar on skills and sit right on it for the role’s actual demands.

Worth keeping separate from a related number, because the two get confused constantly: the resume match score bands you triage before a call, where Strong Go runs 90 to 100 percent and Go runs 75 to 89, are not the same thing as the verdict the AI reaches after the interview. One ranks a document, the other summarises a conversation. We wrote about where each one applies because acting on the wrong one is its own category of mistake.

Why the middle band grows while the edges empty

The two confident tiers carry their own instructions. Strong Go means move this person forward. No Go means the file is closed. Anyone can pick those up, in any order, in whatever ten minutes they have free.

On Hold carries a question instead of an instruction. Answering it means opening the transcript, reading the evaluation, and forming a view. That is real work, it needs a block of attention rather than a gap, and it competes with every other thing in a recruiter’s week during placement season.

So the band grows. Not through neglect exactly, but through a completely rational preference for work that can be finished. If your JD also uses auto-decide thresholds to close out the obvious accepts and rejects automatically, the effect gets stronger, because the queue a human sees is now made almost entirely of the cases that need judgement. That is the feature working as designed. It also means the remaining pile is denser than it used to be, and needs a plan to match.

The deadline is your working rule, not a setting

Be clear about this, because it would be easy to imply otherwise: HireQwik does not ship a review deadline. There is no SLA field, no timer on the band, no escalation that fires on day four. The verdict tiers, the review queue and the per-JD thresholds are product. The deadline is a decision your team makes and writes down.

I would rather say that plainly than sell a governance feature that does not exist. The rule works anyway, because the constraint was never technical. Nothing stopped anyone reviewing those candidates in week one. What was missing was a named owner and a date.

So the rule is two sentences long. Every On Hold candidate gets a decision within three working days of the interview. One named person owns the band for each drive.

Why three days, in an Indian campus context

There is nothing special about three days on its own. It fits the calendar this audience actually works to. Campus offer-to-joining windows are short, competing employers are interviewing the same students in the same fortnight, and a candidate who has heard nothing assumes the worst.

The external research lines up with that. Roughly 48 percent of candidates lose interest if they have not heard back within a week of applying, and more than half of 18 to 25 year olds expect a response inside seven days. That is precisely the age band a campus drive is built around. Meanwhile 36 percent of candidates report waiting one to two months for next steps after applying, and among candidates who withdrew from a process, 32 percent had accepted another offer.

Three working days keeps you inside the window that student is measuring you against. If your drives run on a different rhythm, pick a different number. Pick one.

Work the band in slices, not in one sitting

Sixty undecided candidates is intimidating as a single block and manageable as four groups of fifteen. Filtering is what turns one into the other.

The score filter on the Needs Review tab narrows the pre-call queue by resume-match band, and the same instinct applies after the calls: take the On Hold rows for one role at a time rather than the whole drive at once. Reviewing twelve candidates for a single JD is one context in your head. Reviewing sixty across five roles is five contexts, switched between at random, which is how reviews get abandoned halfway.

Two things to avoid while working through a band this way. First, do not let the pile itself become the argument for a bulk action; clearing a queue by selecting everything and rejecting it is the bulk-action mistake that turns a review backlog into a hiring problem. Second, if you find yourself agreeing with the AI on every single row without opening anything, you are not reviewing, you are rubber-stamping. A periodic audit of the No Go pile is the honest counterweight to that.

A worked example: sixty rows, one afternoon

Say a drive across four colleges finishes with sixty candidates in the middle band, spread over three open roles. The instinct is to book a two hour block and get through all of them. That block gets moved twice and never happens.

What works better is three shorter passes, one per role, each with its own question in mind. The support role’s band is mostly candidates whose communication was serviceable but not warm, so the question is whether the desk they will sit at needs warmth or accuracy. The analyst role’s band is mostly candidates strong on the technical questions and thin on the scenario ones, so the question is how much of that a first month of training closes. The field role’s band is mostly people who answered well and hedged on location, which is not really an On Hold at all; it is a question somebody can settle with a two minute phone call.

That last group is the one that surprises teams. A meaningful slice of any middle band is not genuinely uncertain, it is waiting on a single missing fact. Separating those out first usually clears a third of the pile in twenty minutes and leaves you with a smaller set of real judgement calls, which is a much easier thing to schedule.

When the band is too big to be about candidates

If the middle band is a handful of candidates per role, that is the system working. If it is 40 percent of everyone who interviewed, stop reviewing and look at the configuration instead.

A band that large usually means the scoring is not separating people, and that is almost always a rubric problem rather than a model problem. A rubric with one real criterion can only produce a few distinct scores, which bunches candidates together in the middle where no threshold can split them. We took that lesson publicly when every candidate on a role scored 100 percent, and the same shape shows up as an oversized On Hold pile after the calls.

The fix is upstream: give the rubric enough criteria to spread candidates out, and revisit the per-JD thresholds once you can see where real candidates actually land. Reviewing your way through a band that the configuration created is treating a symptom, and it is a symptom that returns on the next drive.

When you genuinely cannot decide

Some candidates stay ambiguous after you have read everything. The transcript is fine. The evaluation is fine. Nothing is wrong and nothing is compelling.

The answer is to convert the ambiguity into a next action rather than leaving it as a verdict. Three options, in the order we usually suggest them: a short human conversation with a specific question the screen did not resolve, a move to a different open role where the same profile is a better fit, or a clear decline.

Declining is a legitimate outcome and it is better than silence. A candidate who is told no can move on to their next option. A candidate left in the middle band for six weeks eventually works out what happened, and the college’s placement cell remembers it next season. We have written before about what never replying to campus applicants actually costs, and the On Hold band is where most of that silence is generated.

What to actually write when the answer is no

The decline is the part teams stall on, which is one reason candidates sit in the band. Nobody wants to write it, so nobody opens the row.

Keep it short and do not manufacture a reason you did not use. Three lines is enough: that a decision has been made, that it is a no for this role, and whether their profile stays on file for other openings. Skip the detailed feedback unless you genuinely have something specific and defensible, because vague feedback invites a reply asking for more, and a reply nobody answers is worse than the original silence.

What matters more than the wording is that it goes out on a schedule rather than whenever somebody feels up to it. If your rule is three working days, the decline is part of that rule, not a separate task that happens later. A batch of declines sent together on day three is a completed loop. The same declines sent whenever someone remembers is how a placement cell concludes you went quiet on their students.

One caution worth stating: a decline from the middle band is a human decision, and it should read like one. If you are declining an entire band without opening any of it, that is not a deadline working. That is the bulk action problem wearing a deadline as cover.

What the deadline is really protecting

It is tempting to frame this as candidate experience, and it is that. But the thing a decision deadline really protects is the credibility of the screening itself.

The whole argument for automated screening is that it lets a small team make decisions about thousands of people without the pipeline collapsing. Across pilot campaigns we have run 1,099 interviews on that basis. If the output of all that work is a queue of undecided candidates that quietly ages, then the screening did not compress the hiring timeline. It relocated the bottleneck from the phone screen to the review queue, and a relocated bottleneck is much harder to see, because it does not look like a backlog. It looks like a dashboard.

Write these three lines down before your next drive

Not a framework, just the smallest thing that works. Before the interviews start, agree on the review owner for each drive, the number of working days an On Hold candidate may wait, and the default action if the deadline passes with no decision made.

That last line is the one people skip and the one that does the work. A deadline with no default is a suggestion. Decide now whether an undecided candidate at day four gets a human call or a decline, and the band stops accumulating on its own.

Then hold yourself to the same standard you hold the tooling to. We can already tell a recruiter whether one specific candidate was actually contacted rather than only how many messages a campaign sent, and the review side deserves the same treatment: a named person, a date, and an outcome for every row. You can see the full band and every verdict behind it in the HR dashboard.