This case study describes a representative high-volume graduate hiring programme and the deployment pattern GoMeasure AI uses for campus recruitment. Figures are illustrative of a programme of this shape; the client is anonymised.
The challenge: 12,000 applicants, no track record, six weeks
A national graduate recruiter runs a single annual campus intake: roughly 12,000 applications for a few hundred entry-level seats across engineering, operations and commercial functions. The window is fixed — offers must go out before competitors' do — which compresses the whole funnel into about six weeks.
Two structural problems made this nearly impossible to do well. First, graduates have no track record, so the résumé signal is at its weakest: grades and university brand, both heavily biased and only loosely related to on-the-job capability. Second, the recruiting team could realistically first-round-screen only a few hundred candidates by phone, which meant more than 95% of applicants were filtered on paper alone, before anyone had spoken to them.
The team's honest description of the old process: "We reject eleven thousand people based on a document a chatbot could have written, and we interview the ones whose CVs happened to match our filters. We have no idea how many good candidates we never spoke to."
Why the obvious fixes did not work
Adding recruiters does not scale — the window is fixed and the volume spikes once a year, so permanent headcount sits idle for ten months. Aptitude tests helped a little but were gameable and told the team almost nothing about communication or applied reasoning. And a fixed-script "AI interview" — the same questions for all 12,000 — would have been fairer than résumé screening but still uninformative: it wastes a strong candidate's time and cannot tell the difference between a rehearsed answer and a real one.
What the programme needed was a way to give every applicant a real, adaptive interview that measured ability directly, ran without a human in the room, and produced a shortlist the team could defend.
The approach: adaptive interviews as the first round
The redesign inverted the funnel. Instead of screening on paper and interviewing the survivors, every applicant received an adaptive AI interview as the first step. The résumé became context, not a gate.
Layer 1: Calibration from the role
Each role's requirements were turned into an interview plan against the skills the job actually needed — a warm-up to settle nerves, a core band covering essential skills, and a stretch band to separate the strongest candidates. For graduates, the plan leaned on applied reasoning and communication rather than experience nobody had yet.
Layer 2: Adaptive interviewing at volume
Applicants interviewed asynchronously, on their own device, in their own time — no scheduling across 12,000 people. Each interview adapted in real time: computerized adaptive testing chose each next question by information gain, and item-response modelling tracked ability with a confidence bound, so the interview ran only as long as it needed to reach a reliable read.
Layer 3: Communication assessment (LSRW)
For client-facing and commercial roles, a listening-reading-speaking-writing track measured communication directly — the single most predictive and most résumé-invisible skill for early-career hires.
Layer 4: Integrity monitoring
Every session was integrity-monitored for presence, attention and liveness, so a remote, unsupervised interview at this scale still produced evidence the team could stand behind.
Layer 5: Evidence-ranked shortlists
Instead of a pass/fail filter, the output was a ranked list. Every candidate carried an ability estimate, a confidence bound, and the specific answers behind each score — so recruiters spent their limited human time on a shortlist built from demonstrated skill, with the evidence attached.
What changed
Beyond the mechanics, three outcomes mattered to the programme owner. Every applicant got a fair, identical opportunity to demonstrate ability — a meaningful shift in candidate experience and employer brand for a cohort that talks to each other. The shortlist rested on measured skill rather than pedigree, which widened the pool of universities represented. And every selection and rejection was backed by evidence, which turned "why didn't I get through?" from an awkward silence into an answerable question.
The strongest signal was not efficiency, though the time savings were real. It was that the team could finally answer, for any of the twelve thousand, exactly why the decision went the way it did.
Lessons from the deployment
- Interview first, screen later. The biggest gain came from inverting the funnel — measuring ability before filtering on paper, not after.
- Adaptive length is what makes 100% coverage affordable. Fixed-length interviews at this volume are either too short to be fair or too long to finish; adaptive stopping spends questions only where they change the outcome.
- Integrity is not optional at scale. Unsupervised remote interviewing only produces defensible evidence if every session is monitored — otherwise the shortlist is back to being a claim.
- The evidence trail is the real deliverable. Ranked scores are useful; the answers behind them are what let the programme defend decisions to candidates, universities and leadership.
Key takeaways
- Campus hiring breaks résumé screening completely: no track record, huge volume, a fixed window.
- Giving every applicant an adaptive AI interview as the first round measures ability directly, without adding headcount.
- Adaptive stopping makes 100% coverage affordable; integrity monitoring makes remote screening defensible.
- Recruiters move from reading thousands of CVs to reviewing evidence-backed shortlists — human judgement where it counts.
- The lasting value is the evidence trail: every decision, for every candidate, becomes answerable.