When One Role Eats 94% of Your Screening Volume

Nine roles were open across one HireQwik deployment in July 2026. If volume had split evenly, each role would have handled roughly 2,700 resumes and 119 interviews. It didn’t split evenly. One posting — for a DL/ML Research Intern working on LLMs and VLMs — pulled in 22,761 of the month’s 24,327 resumes and ran 621 of its 1,074 interviews. That’s one role accounting for about 94% of resume volume and 58% of interview volume, out of nine roles open at the same time.

This isn’t a data anomaly worth explaining away. It’s the normal shape of hiring volume the moment a second opening goes live alongside the first, and it has real consequences for how a screening pipeline and a review queue need to be built. This post covers what that concentration looked like, why it happens, and what it means for planning your own hiring month if you expect anything close to nine roles running at once.

The split, role by role

The full month’s totals were 24,327 resumes, 1,074 interviews, and 54,024 credits across 9 roles. The DL/ML Research Intern posting alone accounted for 22,761 resumes, 621 interviews, and 48,627 credits — leaving the other 8 roles combined to share 1,566 resumes, 453 interviews, and 5,397 credits. Put differently: the average of the other 8 roles was under 200 resumes each, while the one high-interest posting drew more than 22,000 on its own.

That’s not a split anyone would design on purpose. It’s what happens when one job posting hits a nerve — in this case, a technical internship in a hot skill area — while the other eight roles get the ordinary volume a specific, targeted posting usually gets. Lopsided volume is closer to the rule than the outlier whenever several requisitions run side by side, and treating it as surprising is a planning mistake.

The split in one table

Numbers are easier to compare side by side than spread across paragraphs:

The table makes the pattern this post is built around visible at a glance: the dominant role’s share of resume volume (94%) and credit usage (90%) sit close together, while its share of interview volume (58%) sits meaningfully lower. Three roughly-90%-plus numbers and one roughly-60% number, side by side, is the concentration in one picture.

This is a Pareto pattern, not a fluke

Hiring teams that have run more than one requisition at a time will recognize this shape even without the exact numbers — it’s a familiar pattern from outside recruiting too, sometimes called the 80/20 rule: a small share of inputs producing a large share of outcomes. Here it shows up as roughly 1 role in 9 producing roughly 9 in 10 resumes. That’s not a coincidence specific to DL/ML hiring or to July 2026 — it’s what tends to happen whenever a set of postings compete for attention unevenly. The lesson isn’t “something went wrong with role diversity.” It’s “expect concentration, and build the review process to handle it.”

Why some roles pull disproportionate volume

A handful of factors reliably produce this kind of skew. The sheer scale involved is worth grasping first: a single recruiter now reviews upward of 2,500 applications a year spread across their open requisitions, so even a modest imbalance between postings translates into thousands of extra resumes landing on one person’s plate. Role type matters a lot — entry-level and internship postings in high-demand technical areas (LLMs and VLMs being an especially visible one right now) draw far more applicants than a specialized senior role with a narrow qualifying bar. Distribution matters too: a listing pushed out across job boards and campus channels reaches a far bigger pool than one shared only through a niche network. None of that has anything to do with the screening tool itself — it’s upstream of screening, in how many people see and apply to a given posting in the first place.

The reason this matters for a post about AI screening specifically is that the screening layer has to absorb whatever distribution shows up, without breaking down under the high-volume role or under-serving the low-volume ones. A pipeline that only works smoothly when volume is evenly spread across roles isn’t ready for a real hiring month.

What this does to a shared review queue

If every candidate from all 9 roles landed in one undifferentiated list, the DL/ML Research Intern candidates would swamp it — 22,761 resumes’ worth of scored candidates sitting beside the other 8 roles’ much smaller pools, with no quick way to separate one from another. That’s exactly why HireQwik’s Needs Review queue at app.hireqwik.in/dashboard/hr filters by role alongside resume-match-score band. A recruiter working the low-volume roles that month could filter straight to them without wading through the dominant posting’s much larger candidate list, and a recruiter dedicated to the high-volume role could work that queue without losing track of where they’d gotten to.

Without role-level filtering, volume concentration like this doesn’t just slow reviewers down — it actively hides the smaller roles. A queue sorted by recency or score alone, with no role filter, would surface DL/ML Research Intern candidates almost exclusively simply because there are so many more of them, burying the other 8 roles’ candidates further down the list even if some of them are strong matches.

Why the per-JD rubric doesn’t bend under volume

The mechanism that scores each candidate doesn’t change based on how many other people applied to the same role. Every candidate for the DL/ML Research Intern posting was scored against that role’s own per-JD rubric, the same way every candidate for the other 8 roles was scored against theirs — relevance-aware resume scoring evaluates each profile against whatever that posting actually asks for, and applicant count never enters into it, whether 200 people applied or 22,000 did.

That consistency matters because a high-volume role is exactly where scoring drift would do the most damage. If a scoring mechanism got looser or stricter under load, the role with 22,761 resumes would be the one most affected by it — a small shift in threshold behavior applied 22,761 times produces a much bigger swing in outcomes than the same shift applied to a role with 200 applicants. Per-JD scoring that doesn’t vary with volume is what keeps a high-interest posting from becoming a screening bottleneck or a quality risk.

What this means for the interview stage

Volume concentration carries through to interviews too, just less sharply — 58% of interviews for one role out of 9, compared to 94% of resumes. That gap between the resume-share and the interview-share is itself informative: it means the DL/ML Research Intern posting’s resume-to-interview conversion ran narrower than the blended month-wide average, likely because a large applicant pool for one specific, in-demand technical role produced a wider spread of fit — some strong matches, but also a long tail of resumes that looked relevant on the surface but didn’t hold up against the role’s actual rubric.

For reference, general funnel benchmarks land near a 3% applicant-to-interview rate, so this posting’s roughly 2.7% sits close to the wider norm despite its extreme volume — the concentration made the pool bigger without making the filter looser.

That’s a useful sanity check for any high-volume role: if the resume share and interview share track closely together, the resume scoring isn’t discriminating much within that pool. A meaningful gap between the two, like the one here, is a sign the scoring is doing real filtering work even inside a single dominant role, not just filtering roles against each other. It’s a check worth running on any role that pulls outsized volume, precisely because that’s the role where a scoring mechanism that isn’t actually discriminating within the pool would do the most damage — a flat pass-through at 22,761 resumes wastes far more review time than the same flat pass-through at 200.

What this means for staffing the review queue

Volume concentration has a direct staffing consequence: a recruiting team that splits review responsibility evenly across roles — say, one reviewer assigned per role — will end up with one badly overloaded reviewer and eight who are mostly idle. That’s a natural instinct when nine roles are open; it’s also exactly the wrong way to allocate attention once one role is pulling most of the volume. A better default is to staff toward the dominant role first, then cover the remaining eight with lighter, shared coverage, and adjust as the month’s actual split becomes clear rather than guessing evenly at the start.

This is also where role-level filtering earns its keep beyond just tidiness. A reviewer assigned to the DL/ML Research Intern queue specifically can work through 22,761 resumes’ worth of scored candidates without the other eight roles’ much smaller pools interrupting their flow, and a reviewer covering the remaining eight roles can move through a genuinely smaller, more varied set without the dominant role crowding it out. Splitting the work by role, once you know which role is dominant, is a straightforward staffing decision — the harder part is not assuming the split in advance, since which role turns out to dominate isn’t always obvious before applications actually arrive.

Planning for a lopsided month before it happens

If you’re opening multiple roles at once, the mistake to avoid is sizing your review capacity around an even split. Nine roles doesn’t mean nine similarly-sized queues — plan for one or two roles to pull the majority of volume, and make sure whoever’s reviewing has a way to filter down to the roles that aren’t the loud one. A few concrete things worth doing before a multi-role hiring push, rather than after the volume has already arrived:

• Don’t pre-assign reviewer capacity evenly across roles. Wait for the first few days of applications to show which role is pulling volume, then reallocate.
• Set up role-level filtering in the review queue before applications start arriving, not after the queue is already unmanageable.
• Check in on the split mid-month, not just at the end. A role that looks evenly matched in week one can pull far ahead by week three once a posting has had time to circulate.
• Treat the dominant role’s rubric with extra scrutiny. Since it’s producing the most volume, a scoring issue there affects far more candidates than the same issue on a low-volume role.

It’s also worth checking where credit and cost actually accrue once volume concentrates like this — a role pulling 94% of resumes is very likely also pulling close to that share of usage cost, which matters if you’re budgeting a hiring season role by role rather than as one blended total.

The broader lesson from July 2026’s data isn’t specific to DL/ML hiring or to this one deployment. Any time you run multiple concurrent postings, expect the volume to be lopsided, expect one or two roles to dominate it, and build your review process around filtering by role from the start rather than retrofitting it once a queue gets unmanageable.

It’s worth adding one caveat before treating “94%” as a number to expect every time: this was one particularly high-interest posting in one particularly high-demand skill area, in one hiring season. A different mix of nine roles — say, nine roles with genuinely similar seniority and similar levels of applicant interest — would produce a flatter split than this one did. What’s reliably true across mixes, though, is that an even split is the exception, not the default assumption to plan around. See how role-level filtering works on a live review queue before your own next multi-role hiring push.