A BPO hiring case study from 35 to 120 agent hires a month
In this BPO hiring case study, a 2,000-seat Phoenix call centre moved first-round agent screening to The Cognitive's live AI video interview. Customer-reported results: cost per hire fell from $1,800 to $230, hires rose from 35 to 120 a month, and 90-day retention climbed from 54% to 78%. Recruiters made every decision.
What changed at the Phoenix BPO?
The first 3 rows are the customer's own before and after figures. The last 2 are our arithmetic on those figures, so you can check them.
| Metric | Before | After | Change |
|---|---|---|---|
| Cost per hire (customer-reported) | $1,800 | $230 | 87% lower |
| Agent hires per month (customer-reported) | 35 | 120 | 3.4 times as many |
| 90-day retention (customer-reported) | 54% | 78% | Up 24 points |
| Hires per recruiter per month (our arithmetic, 6 recruiters) | About 5.8 | 20 | 3.4 times as many |
| Agents still employed at day 90, per month of hiring (our arithmetic) | About 19 | About 94 | About 5 times as many |
Customer figures from our case study record dated 3 April 2026. Per-recruiter and day-90 figures are our arithmetic on them.
Who is the customer, and what was broken?
The customer is a 2,000-seat BPO in Phoenix, Arizona, named in our record only as BPO Company. It hires customer service agents, and it needed 100+ new agents every month to keep its floor staffed. Its 6-person recruiting team could get through 35 hires a month, at a cost the team put at $1,800 per hire in recruiter time.
Screening ran on rushed phone screens fitted around everything else. The Head of Recruitment describes the old approach bluntly: the team "used to hire anyone who showed up". Only 54% of new agents were still there 90 days later, so a big share of each month's recruiting went on replacing the agents hired the month before.
The agents who get through actually stay. Head of Recruitment, BPO Company, Phoenix AZ
How bad is agent turnover across call centres?
Turnover above 50% a year is common. In ICMI survey data reported by Informa Connect on 9 June 2026, more than 4 in 10 contact centre organisations still reported annual agent turnover above 50%. The same article cites Insignia research from April 2026 that puts the industry average at 40 to 45%, and up to 60% in high-stress sectors.
The US Bureau of Labor Statistics projects about 289,500 openings a year for customer service representatives from 2025 to 2035, while total employment in the job shrinks by 5%. It expects every one of those openings to come from replacing people who leave the occupation. Agent hiring is a replacement market, so first-round volume never goes away.
That makes early churn expensive. Verint's 2026 State of the Agent Experience study, also cited in that article, puts the cost of replacing one agent at $20,000, counting recruitment, training and lost productivity. This BPO was losing 46 of every 100 new agents inside 90 days.
What did the AI interview assess for each agent?
A live, two-way AI video interview: the candidate checks camera and microphone, the AI asks and follows up, and the room closes on time.
Every applicant got a short first-round interview with the AI over video, which the customer reported as 15 minutes. Today a recruiter sets each role to 10 or 20 minutes, in one of 9 languages. The criteria were communication, problem-solving and empathy, probed with scenario questions written for customer service work.
What does an example rubric for a customer service agent look like?
Example rubric, the kind you can set up. It is not this customer's actual rubric. Each criterion has a weight and a description of what a 1, a 3 and a 5 look like.
| Example criterion (weight) | A 1 looks like | A 3 looks like | A 5 looks like |
|---|---|---|---|
| Spoken clarity (25) | Hard to follow. Long pauses, or answers that drift away from the question. | Clear and polite, but describes what they would do rather than what they would say. | Short, plain sentences. Checks the customer understood before moving on. |
| Empathy and de-escalation (25) | Argues with the caller or jumps straight to policy. | Stays calm and apologises, using stock phrases that never name the actual problem. | Names the customer's frustration in their terms, then moves to a fix with a clear next step. |
| Problem-solving in a scenario (25) | Cannot say what they would check first. | Reaches a sensible fix but skips verifying the account or confirming the outcome. | Asks questions in a sensible order, verifies, offers an alternative when the first fix is not allowed. |
| Policy and accuracy (15) | Promises something the scenario's policy does not allow. | Follows the policy when reminded, but misses a step such as noting the account. | Applies the policy unprompted, explains it without blaming the company, records what was agreed. |
| Shift and role reality (10) | Has not thought about schedules, back-to-back calls or handle-time targets. | Accepts the schedule but stays vague about what high call volume is like. | Names the shifts they can work and describes a past high-volume job concretely. |
Weights add to 100. The report turns the 1 to 5 scores into a weighted score out of 100.
Which example scenario questions suit a call centre agent interview?
Examples you could add to an agent role as your own questions. The AI asks added questions in its own order and adds questions of its own. These are illustrations, not the customer's questions.
A customer was charged twice this month and says they will cancel today. Walk me through your first 2 minutes on that call.
Acknowledges the double charge, verifies the account before promising anything, explains the refund timeline in plain words and gives a reference number.
The refund takes 7 to 10 days and they want it now. What do you say?
A caller asks you to waive a fee that your policy does not let you waive.
Says no without hiding behind the word policy, explains the reason, then offers what is possible, such as an escalation route or a payment plan.
The caller asks for your supervisor. What happens next?
Your CRM goes down in the middle of a call and the queue is building.
Tells the customer honestly, takes notes by hand, sets a realistic callback time and keeps the call short so the queue can move.
How do you make sure that callback actually happens?
What does the recruiter see after each interview?
An example interview report, shown here for an engineering role: scores per criterion, a weighted score out of 100 and resume claim checks.
Each criterion gets a 1 to 5 score, and overall written feedback names strengths and areas to improve. The scores roll up into a weighted score out of 100, and the report suggests a verdict. Nothing is rejected automatically. A recruiter reads the report, opens the recording when a score looks borderline, and decides who moves on.
How did cost per hire fall from $1,800 to $230?
The record gives cost per hire in recruiter time: $1,800 before and $230 after, which is 87% lower. Multiply by volume and monthly hiring spend went from $63,000 to $27,600 while hires rose 3.4 times.
Retention moves the picture more than cost does. At 54%, 35 hires leave about 19 agents on the floor after 90 days. At 78%, 120 hires leave about 94. Divide spend by the agents who stayed and the cost of each one fell from about $3,333 to about $295.
Replacement cost makes the retention gain bigger still. Per 100 agents hired, 46 used to leave within 90 days; now 22 do. That is 24 fewer early leavers per 100 hires. At Verint's $20,000 per replacement, it is $480,000 of replacement cost avoided for every 100 agents hired.
The working, line by line
Every input below is customer-reported. Everything after the equals signs is our arithmetic.
COST PER HIRE Before $1,800. After $230. $1,800 minus $230 = $1,570 less per hire. $1,570 / $1,800 = 87.2%, the 87% in the headline. MONTHLY HIRING SPEND (hires x cost per hire) Before: 35 x $1,800 = $63,000 After: 120 x $230 = $27,600 $27,600 / $63,000 = 44% of the old spend for 3.4 times the hires. COST PER AGENT STILL EMPLOYED AT DAY 90 Before: 35 x 54% = 18.9 agents kept. $63,000 / 18.9 = about $3,333 After: 120 x 78% = 93.6 agents kept. $27,600 / 93.6 = about $295 HIRES PER RECRUITER PER MONTH (6 recruiters) Before: 35 / 6 = about 5.8 After: 120 / 6 = 20 AGAINST THE TARGET OF 100+ HIRES A MONTH Before: 35, at least 65 short every month. After: 120, about 20 over. EARLY LEAVERS PER 100 HIRES (Verint: $20,000 per replacement) Before: 46. After: 22. 24 fewer. 24 x $20,000 = $480,000 avoided per 100 hires. OVER A YEAR ($63,000 minus $27,600) x 12 = $424,800 less hiring spend.
How do you replicate this in a BPO?
These are the real steps in The Cognitive today. Give it at least one week of reports before you judge the result.
- Create the agent role and write the rubric Add 4 or 5 weighted criteria, like the example above. Put what a 1, a 3 and a 5 look like into each criterion's description, because both the interviewer and the scoring read it.
- Add your scenario questions Paste in the 3 or 4 scenarios your floor actually sees. The AI asks them in its own order, adds its own questions and follows up when an answer is thin.
- Set the length and the language Choose 10 or 20 minutes per role. Pick one of 9 languages, so a Spanish-language queue can run as its own role in Spanish. 10 minutes covers 2 scenarios for a high-volume first screen. 20 minutes leaves room for a follow-up on every scenario.
- Bring applicants in Import them from your ATS through one of 60+ ATS integrations, or upload a CSV. Only candidates imported from the ATS get a note written back to it.
- Bulk invite the whole batch Select every new applicant and invite them in one go. Candidates book their own slot without an account and get reminders by email. Invite on the day people apply, so the gap between applying and interviewing stays short.
- Review reports and decide Sort by weighted score, read the feedback on borderline candidates and open the recording when unsure. A recruiter moves each person forward or declines them. A simple rule: read in full anyone with a 2 or lower on empathy, whatever the overall score, before deciding.
- Check scores against 90-day retention After each cohort reaches 90 days, compare who stayed with their interview scores. Adjust the weights on the criteria that predicted staying.
What does this case study not show?
- One customer, one site, one role family: a data point for your planning, not a benchmark.
- Retention gains need the rubric to match the job; criteria that reward polish can hire the wrong people faster.
- Recruiters still read reports daily. At 120 hires a month, an unread queue undoes the speed.
- Training, pay and floor management still decide whether a good hire stays past day 90.
- The setup dates from early 2026. The product has changed since, and this page describes today's workflow.
What did a recruiter's month look like before and after?
One recruiter's month on the 6-person team, built from the reported figures. Per-recruiter numbers are our arithmetic.
| Part of the job | Before: rushed phone screens | After: AI first interviews |
|---|---|---|
| First-round screens | Squeezed between everything else, by phone | Run by the AI; applicants book their own slot |
| Hires per recruiter a month | About 5.8 | 20 |
| How a decision is made | A judgment from a short, rushed call | Weighted score, written feedback and the recording on borderline cases |
| Agents still there at day 90, per recruiter | About 3 a month | About 15.6 a month |
| Cost per hire in recruiter time | $1,800 | $230 |
| Where the saved hours went | Replacing last month's leavers | Onboarding the agents who were hired |
Can AI replace an interview outsourcing company for BPO hiring?
An interview outsourcing company runs your first-round screens with its own screeners, working from your criteria. It adds capacity fast. The cost is consistency: each screener applies their own judgment, and unless calls are recorded you see notes rather than the conversation.
A live AI interview covers the same first round from inside your own account. Every candidate is scored against one written rubric, with a recording and transcript you can open. Your recruiters still read the reports and make each call. In Phoenix, the team put the hours it got back into onboarding.
What should a BPO do and avoid when it copies this?
Do
- Write what a 1, 3 and 5 look like before the first invite goes out.
- Invite every applicant in one batch, the day they apply.
- Read the written feedback on borderline scores before you decide.
- Track 90-day retention by interview score, cohort by cohort.
Do not
- Treating the suggested verdict as the decision. A recruiter accepts or overrules it.
- Criteria that reward polish or confidence over what the answer contains.
- Putting this customer's numbers into your business case as a forecast.
Try it on one agent role
Set up one role with the example rubric, bulk invite this week's applicants and read the reports before you change anything else. Start free at https://app.thecognitive.io/signup.
Where to go next
AI recruiting for BPO and staffing · High volume hiring · Customer support interview questions · How the AI interviewer works · All case studies · Hourly hiring software
Sources
Customer figures and quote: The Cognitive case study record for BPO Company, dated 3 April 2026. Industry turnover: Roberta Gamble, "Contact Center Tech Keeps Improving. So Why do Agents Keep Leaving?", Informa Connect, 9 June 2026, citing ICMI, Insignia (April 2026) and Verint (2026). informaconnect.com/contact-center-tech-keeps-improving-so-why-do-agents-keep-leaving/
Openings and employment: US Bureau of Labor Statistics, Occupational Outlook Handbook, Customer Service Representatives, last modified 27 August 2026. bls.gov/ooh/office-and-administrative-support/customer-service-representatives.htm. Both sources read on 4 October 2026.
Interviews per hire by department: Ashby, "Recruiter Productivity", 2026 Talent Trends Report, 28 April 2026, ashbyhq.com/talent-trends-report, read 5 October 2026.
Frequently asked questions
Can an AI interviewer handle BPO hiring?
Yes. In this case study a 2,000-seat Phoenix BPO ran its first-round agent interviews through The Cognitive's live AI video interview and reported going from 35 to 120 hires a month. The AI interviews and scores against a rubric the recruiter writes. A recruiter still decides who moves on, and the product rejects nobody automatically.
Can AI interview customer service agents?
Yes. A live, two-way AI video interview can test spoken clarity, empathy and problem-solving with call scenarios, then follow up when an answer is thin. Each criterion gets a 1 to 5 score, with overall written feedback, a transcript and a recording.
How many interviews does it take to hire a customer support agent?
Ashby's 2026 Talent Trends data puts customer support at 9.5 interviews and 8.9 interviewer hours per hire, the lowest of any department it tracks, against 17.9 interviews for engineering. At BPO volume that still adds up: 100 hires a month at 8.9 hours each is about 890 interviewer hours. That load is what the AI first round takes off recruiters.
How much does call centre agent turnover cost?
Verint's 2026 study, cited by Informa Connect, puts the cost of replacing one agent at $20,000 once recruitment, training and lost productivity are counted. In this case study, 90-day retention rising from 54% to 78% means 24 fewer early leavers per 100 hires, about $480,000 of replacement cost avoided per 100 agents hired.
What should a call centre agent interview assess?
Spoken clarity, empathy and de-escalation, problem-solving in a realistic call, applying policy accurately, and whether the candidate understands shifts and call volume. Use scenario questions from your own queue and weight empathy and problem-solving highest. See the call center agent interview questions for more prompts.
How do you reduce call centre agent turnover with AI screening?
Screen on what predicts staying, and check it against retention. This BPO reported 90-day retention rising from 54% to 78% after it interviewed every applicant against fixed criteria instead of rushing phone screens. To repeat it, compare each cohort's interview scores with who was still there at day 90, then adjust the weights.
What is the cost per hire for call centre agents?
It varies widely by employer. This BPO reported $1,800 per hire in recruiter time before the change and $230 after. That differs from the cost of replacing an agent: Verint's 2026 study, cited by Informa Connect, puts replacement at $20,000 once training and lost productivity are counted. Measure your own baseline over one month before comparing.
What is an interview outsourcing company?
An interview outsourcing company runs first-round interviews or phone screens for an employer, using its own screeners and the employer's criteria. It adds capacity quickly, but each screener applies their own judgment, so results vary with who took the call.
Is there an AI alternative to BPO interview outsourcing?
Yes. The Cognitive runs the first-round interview itself as a live, two-way AI video interview, so every applicant can be interviewed without adding screeners. Each one is scored on one written rubric, with a recording and transcript. Offers and human interview scheduling stay in your ATS, because The Cognitive is not an ATS.
Where can I find AI recruitment case studies?
The Cognitive publishes its customer case studies at thecognitive.io/case-studies, and this BPO page is one of them. When you read any AI recruiting case study, check who reported the numbers, whether there was a control group and which months were measured.
Does the AI reject candidates automatically?
No. The Cognitive rejects nobody automatically. Each report suggests a verdict from the weighted score, and a recruiter accepts or overrules it. Integrity flags such as tab switches or the camera turning off are logged for the recruiter to review and do not change the score.
How long is the AI interview for an agent role?
A recruiter sets each role to 10 or 20 minutes. This customer described its interviews as 15 minutes when it reported its results. 10 minutes suits a fast first screen. 20 minutes gives the AI more time for follow-ups on each scenario.
Do candidates know the interview is run by AI?
The invitation tells candidates the interview is run by AI. Candidates then book their own slot from that invitation, without creating an account, and get reminders by email.
Does The Cognitive connect to our ATS?
Yes, through 60+ ATS integrations, including Greenhouse, Lever, Workday, iCIMS and Bullhorn. Import applicants from the ATS and The Cognitive writes a note with the score, summary and report link back to each one after the interview. People sourced inside The Cognitive are not written into the ATS automatically.
How much does an AI interviewer for high-volume BPO hiring cost?
AI Interview starts at $99/month and AI Sourcing starts at $49/month, on monthly plans you can cancel anytime. Each AI Interview plan includes a monthly interview allowance, and extra interviews are billed at the plan's overage rate. Signup is self-serve at app.thecognitive.io/signup, with no sales call needed to set up an agent role.
Explore: AI screening software · Pre-screening interviews · Recruitment process outsourcing · Recruiting metrics · High-volume hiring software · One way video interviews explained