One Month of AI Screening: 24,327 Resumes, 1,074 Interviews

Most vendor pitches for AI screening lean on a case study from a single good week or a hand-picked pilot number. What we don’t see as often is a plain, unfiltered month: every resume that came in, every interview that got scheduled, and where the gap between those two numbers actually went. So here is one, without cherry-picking: a live HireQwik deployment’s full July 2026 usage, covering 9 open roles from July 1 through July 31.

The topline numbers: 24,327 resumes were AI-screened, 1,074 AI interviews were conducted, and the whole month used 54,024 total credits. Read as a funnel, that’s a lot of movement at the top and a much narrower stream by the time a live conversation happens — which is close to the point of running a screening layer at all. What matters more than the totals is what they tell you about where a screening pipeline actually spends its effort, and that’s what the rest of this post walks through.

Why one month, not the whole pilot

HireQwik’s broader pilot experience — 1,099 interviews across 14 campaigns — spans multiple customers, roles, and months since the pilot began. July 2026’s 1,074 interviews is a different, narrower slice: one deployment, one calendar month, 9 roles running concurrently. The two numbers happen to land close to each other, which is coincidence, not the same measurement — one is cumulative across a pilot program, the other is a single production month for one customer.

The reason a single month is worth publishing on its own is that cumulative pilot numbers hide monthly variance. A hiring team doesn’t experience “1,099 interviews spread across 14 campaigns” — they experience whatever volume shows up in their calendar that month, all at once, against whatever headcount they have free to review it. July 2026 is what that actually looked like for one real deployment, not a smoothed-out average.

Why July specifically

The month matters here, not just as a label on the dataset. July sits right at the start of India’s campus hiring season, when a large share of new-graduate technical roles open at once and applicant volume climbs fast in a short window. A month pulled from the middle of a quiet stretch would show a much smaller resume total and probably a flatter, less lopsided split across roles. July 2026’s numbers are closer to a stress-test month than an average one, which is exactly why they’re useful — a screening pipeline that holds up cleanly at peak-season volume tells you more about its ceiling than a quiet month would.

What 9 roles running at once actually means for volume

Running 9 roles concurrently doesn’t mean 9 roughly equal streams. It means one dashboard, one review queue, and wildly uneven volume across roles depending on how many people applied to each. Nine simultaneous openings is on the heavy side but not unusual — recruiters now carry around 14 open requisitions apiece on average, so juggling many roles at once is the working reality rather than an edge case. We cover exactly how lopsided that split got in a separate post on volume concentration — the short version is that a single role accounted for the large majority of both resumes and interviews this month, which is a pattern worth planning around rather than treating as unusual.

For HR, the practical effect of 9 concurrent roles is that the Needs Review queue at app.hireqwik.in/dashboard/hr mixes candidates from every open role in one place, filterable by role and by resume-match score. Without that filter, a recruiter reviewing one role’s shortlist would be wading through candidates for eight others at the same time. The filter exists because concurrent-role screening is the normal case, not the exception, for any account running several openings at the same time.

Where the funnel actually narrows

The gap between 24,327 resumes and 1,074 interviews is the single most informative number in this dataset, and we’ve written it up on its own in a post on the resume-to-interview ratio because it deserves more than a paragraph. Briefly: the vast majority of narrowing happens before a live interview is ever scheduled, at the resume-screening stage, where HireQwik’s relevance-aware resume scoring weighs what a person has genuinely done against what the posting genuinely asks for, rather than counting years worked or degree level in isolation.

That’s a deliberate design choice, not a side effect. A screening system that pushed most resumes straight to a live interview would be expensive to run and exhausting to review — every one of 24,327 resumes would need a human or an AI conversation before anyone could tell whether it was worth pursuing. It’s also broadly how hiring already works without any AI involved: industry funnel data puts the applicant-to-interview rate at roughly 3%, with the overwhelming majority of applicants filtered out before they ever speak to anyone. The question isn’t whether a funnel narrows sharply — they all do — it’s whether the narrowing is done on evidence or on a fast skim. Filtering at the resume stage means the 1,074 people who did get an interview were already resume-matched against that specific role’s requirements before anyone spent 15–20 minutes talking to them.

What actually happens in an interview, once one gets scheduled

Every one of those 1,074 interviews ran HireQwik’s two-evaluator scoring: one evaluator scores the content of what a candidate said against the role’s per-JD rubric, and a second evaluator scores the audio itself — pace, filler and hesitation rate, pronunciation. Neither evaluator runs alone, and a Reject verdict needs both to agree, which is the same asymmetric design used across every HireQwik deployment, not something specific to this dataset.

Before any of that scoring happens, a candidate has already cleared whatever role-specific disqualifier questions the JD carries, which fire at the very top of the call. Candidates who don’t clear those exit the conversation early with a tagged reason HR can see later, rather than sitting through a full interview that was never going to produce a Go verdict. That’s part of why the 1,074-interview figure isn’t inflated by conversations that were doomed from the first question — the interview stage still does its own filtering on top of what the resume stage already did.

The funnel, laid out stage by stage

It’s easier to see where a month’s effort actually went when the totals are broken into stages instead of read as two headline numbers:

Laid out this way, the month’s work is heavily front-loaded: five of the six rows above happen at or near full applicant scale, and only the last two — interview and HR review — operate on the much smaller, pre-filtered set. That’s the structural reason a topline “1,074 interviews” figure understates how much screening activity actually ran that month.

What this looked like week to week

A single monthly total also hides how unevenly the work lands across the four weeks inside it. Applications don’t arrive on a flat, predictable schedule — a posting can sit quiet for a few days and then get a wave of applicants after it circulates on a new channel or gets shared inside a college placement group. The 24,327-resume total for July is the sum of that uneven arrival pattern, not a steady 800-or-so resumes a day. A recruiter watching the dashboard in real time would have seen slow days and spike days, with the interview count trailing the resume spikes by however long a candidate takes to book their self-schedule slot.

Reading a month like this without over-interpreting it

One month of data from one deployment is exactly that — a real, unedited snapshot, not a universal benchmark. A different customer, a different mix of roles, or a different hiring season would produce different totals, and nothing here should be read as “this is what every HireQwik account looks like every month.” What it is useful for is showing, with actual numbers instead of a marketing claim, that the scoring stage and the conversation stage do genuinely different jobs in the funnel, and that most of the volume reduction happens before a live conversation, not after one.

It’s also worth being honest about what this dataset can’t tell you. It doesn’t say how many of the 1,074 interviewed candidates ended up hired, or how the roles compared in offer-acceptance terms — that’s downstream of screening and outside what a usage snapshot like this one measures. What it can tell you is where the effort in a real month actually went: overwhelmingly into scoring resumes at scale, with a much smaller, better-targeted slice of that volume reaching a live conversation.

Why this matters if you’re sizing a screening deployment

If you’re trying to estimate what a screening tool will cost you in review time or usage, the ratio in this dataset is a more useful planning number than either total in isolation. Roughly 1 interview happened for every 22–23 resumes screened this month. That ratio is what determines how much of your recruiter time goes to reviewing interview verdicts versus how much never needs a human at all, because the resume layer already resolved it. A funnel audit worth running on your own numbers is a natural next step once you have a month of data like this to look at — the framework there works whether your ratio looks like ours or nothing like it.

The other planning number worth watching is where cost actually accrues across that funnel, which is a separate question from interview count and one we go into directly in a post on how screening cost scales with resume volume, not interview volume. The short version: because most of the funnel’s work happens at the resume stage, resume volume is the number that moves your usage total the most, not how many interviews eventually get scheduled.

The number that should change how you read a “results” post

Most AI screening vendors publish outcomes — pass rates, time saved, candidate satisfaction — without ever showing the raw funnel that produced them. That’s not necessarily dishonest, but it makes it hard to tell whether a strong-sounding pass rate reflects careful screening or just a small, pre-filtered applicant pool to begin with. Publishing 24,327 resumes against 1,074 interviews is a deliberate choice to show the denominator, not just the headline number, because the denominator is what tells you whether a funnel is actually doing filtering work or just producing a number that looks good in isolation.

That’s the standard we think usage data should be held to, ours included. If you’re evaluating a screening tool for your own team, ask for the same thing: not just how many interviews it ran, but how many resumes it started from and where the gap between those two numbers actually went. A vendor that can’t or won’t share the resume-intake number alongside the interview number is asking you to trust a headline figure without the context that would let you judge whether it’s a good one.

See what a live month of screening data looks like on your own roles before deciding whether the funnel it produces is one you’d want to publish.