An AI interview case study with 1,000 applicants and 10 hires
In this AI interview case study, Prospire, a technology consulting firm in India, gave all 1,000 applicants for engineering, product management and UI/UX roles a role-specific AI video interview. The hiring team met only the 15 people the reports pointed to and hired 10 of them, in weeks rather than months.
What did Prospire's funnel look like before and after?
The first 3 rows come from the customer's record. The last 2 are our arithmetic on those numbers.
| Measure | Before | After |
|---|---|---|
| Applicants interviewed (customer-reported) | Manual review of resumes | 1,000 AI interviews |
| Shortlist (customer-reported) | No shortlist method at this scale | 15 candidates |
| Hires (customer-reported) | Months of effort expected | 10, in weeks |
| Shortlist to hire (our arithmetic) | Not measured | 10 of 15, about 67% |
| Interviews per hire (our arithmetic) | Not measured | 100 |
Customer figures from our case study record dated 6 April 2026. Percentages and ratios are our arithmetic on them.
Why was Prospire stuck with 1,000 applicants?
Prospire is a technology consulting firm in India. It needed to stand up 3 teams at the same time: software engineers, product managers and UI/UX designers. The openings drew 1,000+ applicants across those roles.
Each discipline needs its own bar. An engineer is judged on technical depth, a product manager on judgment and trade-offs, a designer on process and craft. Holding 3 different bars steady across a thousand applications is beyond a small talent team, and the record says manual review would have taken months of recruiter and hiring manager time.
Consulting adds a deadline on top. A consulting firm sells delivery capacity, and it cannot staff client work with a team it has not built yet. For a firm in that position, how fast the 3 teams come together matters almost as much as the bar each hire clears.
How does Prospire's funnel compare with typical tech hiring?
Prospire's figures next to two large 2026 datasets from applicant tracking vendors. Both sources were read on 5 October 2026.
| Measure | Prospire | Published benchmark |
|---|---|---|
| Applications per hire | About 100 (1,000 applicants, 10 hires) | 291 per hire across Ashby customers in Q1 2026; roughly 200 in Gem's data |
| Share reaching a human screen or interview | 1.5% met the hiring team | 8% of applicants advance past initial screening (Gem) |
| Interviews per hire, engineering | 1.5 human interviews per hire | 17.9 interviews per engineering hire (Ashby) |
| Interviews per hire, product management | 1.5 human interviews per hire | 18 interviews per product management hire (Ashby) |
| Interviewer hours per hire, technical | Hiring team time went on 15 meetings | 24.7 hours per engineering hire, 23.5 per PM hire (Ashby) |
| Time from application to offer | Weeks, not months | About 48 days for technical roles, 38 for business roles (Ashby) |
Ashby, Recruiter Productivity, 2026 Talent Trends Report (28 April 2026). Gem, 2026 Recruiting Benchmarks Report (1 December 2025).
We had a thousand applicants and no way to fairly evaluate all of them. The AI interviewed every single one. Out of the 15 it shortlisted, we hired 10. Hiring Lead, Prospire
What would the usual approach have cost them?
The default move is to filter on resumes and interview a small subset. With 1,000 applicants in 3 disciplines, the record puts it plainly: the shortlist ends up decided by whoever reads fastest. Candidates with unconventional backgrounds rarely reach a conversation, because a resume shows where someone worked, not how they think.
Prospire chose the opposite order. Instead of narrowing on paper and interviewing a few, it interviewed everyone and narrowed on what people showed in the interview. That only works if the interview itself costs the team no calendar time, which is where the AI interviewer came in.
How did one team hold three different hiring bars?
Setting up evaluation criteria for a role: each criterion gets a name, a description and a weight that adds to 100.
Each discipline became its own role with its own written criteria, agreed before the first invite went out. Every engineering candidate was scored against the engineering criteria, every product candidate against the product criteria. A product shortlist and an engineering shortlist were each internally comparable, even though the two bars were different.
What might criteria for each discipline look like?
Example criteria to show the shape of 3 separate rubrics, with the reason each one earns its place. They are illustrations, not Prospire's own wording.
| Example role | Example criterion 1 | Example criterion 2 | Example criterion 3 | Why these, for this role |
|---|---|---|---|---|
| Engineering (example) | Explains a past technical decision and its trade-offs | Debugs out loud when given a failing scenario | Works with product and design without handoffs breaking | A resume lists tools. Talking through a trade-off shows whether the person made the decision or watched it. |
| Product management (example) | Frames a problem before naming a solution | Prioritises with a stated reason | Talks to engineers in their terms | Product work is judgment under constraint, so the interview should hear the reasoning, not a framework name. |
| UI/UX design (example) | Walks through a design process from research to handoff | Defends a design choice with user evidence | Takes critique without losing the thread | A portfolio shows the output. The conversation shows the process and how the designer handles pushback. |
| Suggested length (ours) | 20 minutes for engineering | 20 minutes for product | 20 minutes for design | Each needs room for 2 or 3 follow-ups per criterion. 10 minutes suits a quicker first filter. |
The record says each interview covered technical depth and team fit. These rows illustrate that split for each discipline.
How did Prospire run 1,000 interviews, step by step?
The steps in today's product terms. Prospire's setup dates from April 2026, and the product has changed since.
- Agree the bar for each discipline The record says this took more internal debate than the technology. Write a short set of weighted criteria per role and put what good looks like into each description. Get the engineering lead, the product lead and the design lead to sign off their own role's criteria before a single invite goes out.
- Create one role per discipline Engineering, product management and UI/UX each get a role, a rubric and an interview length of 10 or 20 minutes. Add any questions you want covered; the AI adds its own. Starting points for added questions: the [software engineer](/interview-questions/software-engineer), [product manager](/interview-questions/product-manager) and [UX designer](/interview-questions/ux-designer) question banks.
- Load every applicant Import applicants from your ATS through one of 60+ ATS integrations, or upload a CSV. Nobody is filtered on paper first.
- Bulk invite each role's pool Select the whole pool and invite them in one go. Candidates book their own slot, with no account to create, and get email reminders.
- Let the AI interview, then read the reports Each live, two-way video interview produces a 1 to 5 score per criterion, overall written feedback, a weighted score out of 100, a transcript and a recording.
- Decide who the hiring team meets Sort by weighted score and read the borderline reports. The report suggests a verdict; the team decides. At Prospire the hiring team met 15 people in total. A workable review rule: read every report with a 4 or 5 on the heaviest criterion, and open the recording when two criteria disagree by 2 points or more.
- Tighten the wording after the first batch Prospire's rough patch was calibration. Some evaluation language was tightened so technical depth and team fit meant the same thing to every reviewer. Have 2 reviewers score the same 10 reports by hand. Wherever they disagree, the criterion description is too loose. Rewrite it before the next batch.
What did the hiring team read before meeting anyone?
An example interview report: a score per criterion, a weighted score out of 100 and resume claim checks.
The hiring team never had to sit through 1,000 conversations. It read reports, opened a recording when a score looked borderline, and chose whom to meet. The AI probes up to 5 resume claims per interview and marks each one VERIFIED, REFUTED or UNCLEAR, with evidence from the conversation.
The funnel arithmetic, line by line
Inputs are customer-reported. Everything after an equals sign is our arithmetic.
INPUTS Applicants interviewed by AI: 1,000 Shortlisted for the hiring team: 15 Hired: 10 SHORTLIST RATE 15 / 1,000 = 1.5% of applicants reached a human interview. SHORTLIST TO HIRE 10 / 15 = 66.7%, about 2 in every 3 people the team met. INTERVIEW TO HIRE RATIO AI interviews: 1,000 / 10 = 100 per hire. Human interviews: 15 / 10 = 1.5 per hire. APPLICANT TO HIRE 10 / 1,000 = 1%. WHAT THE TEAM SKIPPED 1,000 minus 15 = 985 applicants assessed without a human interview slot. INTERVIEW TIME THE AI CARRIED At 10 minutes: 1,000 x 10 = 10,000 minutes, about 167 hours. At 20 minutes: 1,000 x 20 = 20,000 minutes, about 333 hours. At 8 hours a day, 333 hours is about 42 working days of one person doing nothing but first interviews. AGAINST A TYPICAL TECH FUNNEL (Ashby, 2026) 10 hires x 17.9 interviews = about 179 interviews at the engineering average. Prospire's hiring team met 15 people for the same 10 hires.
What changed, number by number?
1,000 interviews. Every applicant got a conversation instead of a resume skim. That is the change the other numbers depend on: the shortlist came from what people said and did in the interview, not from keywords on a page.
15 shortlisted. Only 1.5% of the pool reached the hiring team. A small shortlist is usually a sign that nobody had time to read more. Here it was small because every applicant had already been assessed against the same written criteria for their discipline.
10 hires. 10 of the 15 people the team met were hired, about 2 in 3. The hiring team's own interview hours went almost entirely on people who ended up joining, and the 3 teams were built in weeks. Ashby's 2026 data puts a technical hire at about 48 days from application to offer, so weeks rather than months is the gap that matters.
What did the hiring lead's week look like, before and after?
The usual approach the record describes, set against how Prospire actually ran it. Figures are the customer's; the hour estimates are our arithmetic.
| Part of the job | Resume-first approach | Interview-everyone approach at Prospire |
|---|---|---|
| Monday morning | A queue of 1,000+ resumes across 3 disciplines, read one by one | A list of completed interview reports, sorted by weighted score per role |
| Who gets a conversation | Whoever the fastest reader picks from paper | All 1,000 applicants, each interviewed against their discipline's criteria |
| Time on first conversations | Only a small subset could ever be called | None. The AI carried about 167 to 333 hours of interviews |
| Hiring team meetings | Many first rounds before anyone looks strong | 15 meetings in total, with the report read beforehand |
| Hires from those meetings | Unknown until months in | 10 of 15, about 2 in 3 |
| Where the hard work went | Reading and scheduling | Agreeing the bar and tightening criteria wording |
What does this case study not show?
- One customer and one hiring push. Treat it as a data point, not a benchmark.
- The interview covers how people explain their work. Hands-on tests and portfolio reviews still belong in later rounds.
- A shortlist of 15 depends on criteria being right; weak criteria would produce a confident but wrong list.
- The 10 hires were made by people. The AI suggested; Prospire's team decided.
- The setup dates from April 2026. The product has changed since, and this page describes today's workflow.
Who does this approach fit, and who should skip it?
Do
- Hundreds of applicants for a handful of roles.
- Several disciplines hiring at once, each with its own bar.
- Teams that want to look past resumes at how people explain their work.
- Hiring managers whose calendars are the bottleneck.
Do not
- A single senior hire from a shortlist of 5. Interview them yourself.
- Roles where the first conversation must be a client meeting or a site visit.
- Teams unwilling to write criteria before inviting anyone.
What still needed a human at Prospire?
Defining the bar. The record describes the hardest part as internal debate about what each discipline's bar should be, not the technology. No tool can settle that argument for you, and the interviews are only as good as the criteria written into them.
Calibration, the final meetings and the offers. Prospire's team tightened prompts and evaluation language after the first round, met the 15 people itself and made every hiring decision. The Cognitive is not an ATS, so offers and later interview rounds stayed in Prospire's own process.
Try it on one role
Pick the role with the biggest applicant pile, write its criteria, bulk invite the pool and read the reports before you meet anyone. Start free at https://app.thecognitive.io/signup.
Where to go next
High volume hiring · How the AI interviewer works · Software engineer interview questions · Product manager interview questions · Pricing · All case studies
Sources and notes
Figures and quote reported by the customer in The Cognitive's case study record for Prospire, dated 6 April 2026. Setup dated April 2026. Percentages, ratios and hour estimates are our arithmetic on those figures. The example criteria are illustrations, not Prospire's own.
Benchmarks: Ashby, "Recruiter Productivity", 2026 Talent Trends Report, 28 April 2026, ashbyhq.com/talent-trends-report, read 5 October 2026. Gem, "Key takeaways from the 2026 Recruiting Benchmarks Report", SJ Niderost, 1 December 2025, gem.com/blog, read 5 October 2026.
Frequently asked questions
Is it practical to interview 1,000 applicants?
Prospire did it across engineering, product management and UI/UX. The AI runs each live video interview, and candidates book their own slots, so throughput is not limited by anyone's calendar. The work that stays with people is writing the criteria and reading the reports.
What is a good interview to hire ratio?
It depends on who does the interviewing. In this case study the ratio was 100 AI interviews per hire but only 1.5 human interviews per hire, because the team met 15 people and hired 10. Track the human ratio separately; it shows how much of your team's time each hire costs.
How do you screen 1,000 applicants without reading every resume?
Interview them all against written criteria and read the scored reports instead. Prospire skipped resume filtering, invited the whole pool, and let the weighted scores and written feedback decide which reports deserved a close read.
How were three different roles assessed consistently?
Each discipline was set up as its own role with its own rubric, agreed before interviews began. Everyone applying for that discipline was scored against the same criteria, while the questions adapted to each person's answers.
Did the AI decide who got hired?
No. Each report suggests a verdict from the weighted score, and people decide. Prospire's hiring team chose the 15 it met and made all 10 hires. The Cognitive rejects nobody automatically.
How long is each AI interview?
A recruiter sets each role to 10 or 20 minutes. For engineering, product and design roles, 20 minutes gives the AI room for 2 or 3 follow-up questions on each criterion, which is where technical depth shows. 10 minutes suits a quick first filter on a very large pool.
How many applications does it take to make one hire?
It varies by role and employer. Ashby's 2026 data shows recruiters processing 291 applications per hire, against roughly 100 in early 2021, and Gem's 2026 benchmarks put it at roughly one hire per 200 applications. Prospire made 10 hires from 1,000 applicants, about 1 per 100.
How long does it take to hire a software engineer?
Ashby's 2026 Talent Trends data puts technical roles at about 48 days from application to offer, against about 38 days for business roles. Prospire reported building its engineering, product and design teams in weeks rather than months, because the first-round interviews ran without waiting for anyone's calendar.
How many interviews per hire is normal for engineering roles?
Ashby reports 17.9 interviews per engineering hire and 18 per product management hire, taking about 24.7 and 23.5 interviewer hours. At Prospire the AI ran the first-round interviews, and the hiring team met 15 people for 10 hires, about 1.5 human interviews per hire before final rounds.
What should an AI interview ask a UX designer?
Ask them to walk through one project from research to handoff, then defend a choice with evidence from users. Follow-ups should test how they took critique and what they changed because of it. Portfolio reviews stay with your design lead. See the UX designer interview questions for more.
Can an AI interview assess engineers, product managers and designers?
It can assess how each explains their work against criteria you write. The AI asks follow-up questions when an answer is thin and probes up to 5 claims from the resume. Hands-on work samples, portfolio reviews and final rounds stay with your team.
Where can I find AI recruitment case studies?
The Cognitive publishes its customer case studies at thecognitive.io/case-studies. When you read any of them, check who reported the numbers, which months they cover and whether anything else changed in the same period.
Does it work with our ATS?
Yes, through 60+ ATS integrations including Greenhouse, Lever, Workday and Ashby, or by CSV upload. Candidates imported from the ATS get a note with score, summary and report link written back. ATS stages never move on their own.
What does it cost to run interviews at this volume?
AI Interview starts at $99/month and AI Sourcing starts at $49/month, on monthly plans you can cancel anytime. Each interview plan includes a monthly interview allowance, listed on the pricing page, so check which tier fits a pool your size. Signup is self-serve, with no sales call.
Explore: What is high volume hiring? · What is a structured interview? · Hire a UX designer · Technical interview software · High volume recruiting strategies