Methodology
How we measure what we claim.
Every statistic published on this site is listed below with what it counts, how it is derived, and what it does not show. All of it is first-party platform data rather than independent research, and we say so plainly — a number is only useful if you can see how it was produced.
Every published figure, and how it is derived.
A single batch of up to 500 resumes is scored against a job description in roughly two minutes.
- What it counts
- Wall-clock time from batch submission to a complete ranked shortlist — every resume in the batch carrying a fit score, a skills breakdown, and specific red and green flags.
- How derived
- Observed processing time for full-size batches run on the platform. Scoring is model- and rules-based and runs in parallel across the batch, so elapsed time is governed by batch size rather than by reviewer availability.
- Basis
- Production batches run through the standard bulk screening pipeline at maximum batch size.
- Limits
- This is per-batch throughput, not a sustained hourly rate, and it varies with resume length and file format. It describes processing speed only — accuracy is a separate measure, covered below.
- Where used
- Homepage · /features/bulk-screening · /product
Structured evaluation scoring reaches 96% average correctness.
- What it counts
- Agreement between the platform's fit-to-role score and an independent recruiter judgment made on the same candidate profile.
- How derived
- Platform scores are compared against a held-out, human-labelled set of candidate profiles, where recruiters assessed the same candidates independently of the AI score. Agreement rate across that set is what the figure reports.
- Basis
- A held-out labelled sample of candidate profiles drawn from platform activity across customer organisations.
- Limits
- Agreement with recruiter judgment is not the same as predicting on-the-job performance. Recruiter labels carry their own variance, so this measures consistency with experienced human screening decisions rather than objective correctness. Accuracy holds at batch scale because scoring does not depend on reviewer attention span.
- Where used
- /resources/state-of-ai-hiring-2026 · /resources/5000-ai-interviews-analyzed
Teams running the full pipeline save up to 40 hours per recruiter each week.
- What it counts
- Recruiter hours removed from two specific stages: first-pass resume review, and the scheduling and conducting of first-round screening calls.
- How derived
- Recruiter and pipeline activity is compared before and after a team moves core screening and first-round interviewing onto the platform. Both stages scale linearly with applicant volume in a manual process and are automated on the platform, which is where the saving originates.
- Basis
- Aggregated platform activity for teams running the complete source-to-shortlist pipeline.
- Limits
- An upper bound, not an average. It reflects teams operating the full pipeline at high applicant volume; teams using only part of the platform, or hiring at lower volume, should expect proportionally less. It measures time removed from two stages, not total recruiting workload.
- Where used
- Homepage · /for/talent-acquisition · /resources/state-of-ai-hiring-2026
Teams reduce time-to-hire by up to 70%.
- What it counts
- Elapsed time from application received to scored shortlist, compared before and after the same team adopts platform screening and first-round interviews.
- How derived
- In a manual process every stage queues behind recruiter availability — applications wait for review, interviews wait for a calendar slot. Removing the human bottleneck from screening and first-round interviews means candidates progress continuously, at any hour. The reduction measures that change in elapsed time.
- Basis
- Before-and-after pipeline timing for teams that moved screening and first-round interviewing onto the platform.
- Limits
- Measured to shortlist decision, not to signed offer — later stages remain gated by human availability and candidate response. As an upper bound it reflects teams whose previous constraint was recruiter review capacity; teams already screening quickly will see a smaller change.
- Where used
- Homepage · /product · /resources/state-of-ai-hiring-2026
Platform-wide adoption and assessment volume to date.
- What it counts
- Organisations with at least one active job on the platform, and candidates who have completed at least one AI-conducted assessment or interview.
- How derived
- Direct platform counts, not survey responses or estimates. Each organisation and each candidate is counted once regardless of subsequent activity.
- Basis
- Cumulative platform totals since launch.
- Limits
- Cumulative rather than active monthly counts. A company that ran one job and a company hiring continuously each count once. The two figures count different things and should not be read as a ratio.
- Where used
- Homepage · /about · /resources/state-of-ai-hiring-2026
What first-party data can and cannot show.
This is a self-selecting group
These figures describe outcomes for teams that chose to adopt an AI hiring platform and then used it. That is not a random sample of the hiring market. Teams that adopt tooling of this kind usually have a volume problem worth solving, which is precisely the condition under which the gains are largest.
We publish it anyway, because real usage data at this volume is scarcer in this category than marketing copy, and because every figure here traces back to a describable process rather than a survey response. Read it as evidence about what the platform does for teams like the ones already using it — not as a forecast for every team.
Upper bounds are labelled as such
Where a figure is the best observed outcome rather than a typical one, the claim says “up to” and the Limits field says so again. The 40-hours and 70% figures are both upper bounds. We would rather a reader discount them correctly than overstate what an average customer should expect.
Corrections
If a figure on this site looks wrong, or you find a claim published without a corresponding entry here, tell us — we will correct it and note the change. Related reading: The State of AI Hiring 2026 and We Analyzed 5,000+ AI Interviews.
Questions about these figures
Is this independent third-party research?
No. Every figure on this page comes from JustInterview.ai's own platform data and should be read as vendor-reported first-party data, not independent or peer-reviewed research. We publish the measurement detail so the figures can be assessed on their merits rather than taken on trust.
Why do some claims say "up to"?
Because they are upper bounds observed among teams running the complete pipeline at high applicant volume, not averages across all customers. Where a figure is an upper bound we say so rather than presenting it as a typical result.
Why aren't exact sample sizes and measurement dates published here?
The underlying data contains customer and candidate information we cannot publish, and we would rather describe the method precisely than publish a figure we cannot stand behind in detail. Where we can add specifics without compromising that, we do — and this page is updated when we can.
Can these figures be independently verified?
Not directly, because they derive from internal platform data. What we can do is state exactly what each figure counts, how it is derived, and what it does not show — which is what this page sets out to do. If a specific figure matters to a decision you are making, contact us and we will tell you what we know about it.
How is screening accuracy maintained at batch scale?
Scoring is model- and rules-based rather than dependent on reviewer attention, so correctness does not degrade between a batch of 50 resumes and a batch of 500. This is what makes bulk and campus recruitment workflows viable rather than a trade-off against quality.
How often is this page updated?
Whenever a published figure changes, a new claim is introduced, or a measurement method is revised. The last updated date at the foot of this page reflects the most recent change.
Get started
See the numbers on your own pipeline.
Run a real batch through JIA and measure it against what your team does today.