Deploying new recruitment technology requires a rigorous cost-benefit analysis. As talent acquisition budgets face increased scrutiny, HR leaders must justify every software investment with hard data. When evaluating AI voice screening tools, the ROI model is heavily front-loaded and immediately measurable.
AI recruiting is worth it when three conditions hold: you screen more applicants than your team can reach, your time to fill is long enough that candidates are lost while waiting, and you can measure the result afterwards. Where those conditions do not hold, the return is small and the honest answer is no. This page gives the model, the published benchmarks behind each input, and the places the standard business case overstates itself.
The short answer
- The savings are real but smaller than most vendor math. The defensible line items are recruiter screening and scheduling hours, agency dependency, and vacancy days.
- Time to hire is the largest lever, and the least well measured. Median time to fill is about a month and a half, and most of that is queueing, not deciding.
- There is now independent evidence. A randomized field experiment across roughly 70,000 applicants found candidates interviewed by AI voice were 12 percent more likely to receive an offer.
- Most organizations do not realize the value. Gartner reports 88 percent of HR leaders saying their organizations have not seen significant business value from AI tools.
What the benchmarks actually say
Almost every article on this topic repeats cost-per-hire figures that its own cited source does not contain. Fixing that first is worth more than another vendor claim, because your business case will be checked against the primary source.
Scroll to see all columns
| Widely repeated claim | What the source actually says |
|---|---|
| "Average cost per hire is $4,129, per SHRM" | Correct but stale. That is SHRM's 2016 benchmarking report, covering fiscal 2015. |
| "Average cost per hire is $4,700, per SHRM" | From SHRM's 2022 article The Real Costs of Recruitment, which puts average cost per hire at about $4,700. |
| "Cost per hire is $4,700 for non-executive roles, per SHRM 2025" | Not what the 2025 data says. SHRM's 2025 Recruiting Executives Benchmarking report reports a median of $1,200 for non-executive roles and $10,625 for executive roles. |
| "Average time to fill is 42 days, per SHRM 2025" | The same report describes a median time to fill of about a month and a half, roughly 45 days, rising to 61 days at extra-large organizations. |
| "Agencies charge 15 to 25 percent" | Roughly right. Industry survey data puts 20 percent as the most common direct-hire fee, used by 42 percent of firms, with professional placements at 18 to 22 percent. |
Scroll to see all columns
Median and average are also different animals. SHRM's $1,200 median and its earlier $4,700 average describe different distributions of the same messy underlying data. Use the median for a typical requisition and the average when a few expensive searches dominate your spend.
The model, one line at a time
The usable form of the calculation is simple. Everything difficult is in the inputs.
Annual return = recruiter hours recovered + agency spend avoided + vacancy days avoided, minus platform and implementation cost.
Scroll to see all columns
| Line item | How to compute it | Benchmark input and source |
|---|---|---|
| Recruiter hours recovered | Requisitions x screening and scheduling hours per requisition x fully loaded hourly cost x share automated | 17.7 hours of manual admin per vacancy, of which 3.6 hours is reviewing applications and 2.5 is scheduling interviews (Totaljobs and The Stepstone Group, 2025, a UK survey of 748 HR leaders) |
| Fully loaded recruiter cost | Base hourly rate x benefits multiplier | Median $36.51 per hour for HR specialists (US BLS, May 2025), times about 1.43, since wages are 69.9 percent of total compensation (US BLS Employer Costs for Employee Compensation) |
| Agency spend avoided | Agency hires x average salary x fee percentage x reduction | 20% most common direct-hire fee (Staffing Industry Analysts) |
| Vacancy days avoided | Roles x days saved x daily value of the role | Median time to fill of about a month and a half (SHRM 2025). Daily value is a modeling assumption, not a measured figure |
| Requisition load | Requisitions per recruiter per year | Median 20 overall, and 50 to 60 at large and extra-large organizations (SHRM 2025) |
| Cost side | Platform cost + integration + training + the first quarter of lower productivity | Usually understated. Most vendors quote license only |
Scroll to see all columns
A worked example
The following uses published benchmarks rather than Braintrust client data, so you can swap in your own numbers and see which assumptions carry the result. Assume a team hiring 300 non-executive roles a year at an average salary of $65,000.
Scroll to see all columns
| Input | Value | Basis |
|---|---|---|
| Requisitions per year | 300 | Assumption |
| Screening and scheduling hours per requisition | 6.1 | 3.6 reviewing plus 2.5 scheduling, Totaljobs and Stepstone (UK data, applied to US wages) |
| Fully loaded recruiter cost | $52 per hour | BLS median wage x BLS benefits multiplier |
| Share of that time automated | 70% | Assumption, and the single most sensitive input |
| Recruiter hours recovered | About $66,600 | 300 x 6.1 x $52 x 0.70 |
| Agency-sourced hires | 45 (15% of hires) | Assumption |
| Agency fee | 20% of $65,000, so $13,000 | Staffing Industry Analysts |
| Reduction in agency reliance | One third | Assumption |
| Agency spend avoided | About $195,000 | 45 x $13,000 x 1/3 |
| Days saved per fill | 10 | Assumption, from removing the scheduling queue |
| Daily value of an open role | $125 | 50% of the daily salary equivalent, a common convention and not a measured figure |
| Vacancy days avoided | About $375,000 | 300 x 10 x $125 |
Scroll to see all columns
Two observations about that table. First, the vacancy line, often labeled cost of vacancy, is the biggest and the softest. It rests on a convention rather than on measurement, and a finance partner will challenge it. Lead with recruiter hours and agency spend, which are auditable in your own ledger, and treat vacancy value as upside.
Second, the automation share drives everything. At 70 percent you recover $66,600 of recruiter time. At 30 percent you recover $28,500. Adoption, not capability, decides where you land.
"This is doing to job recruiters what Kayak did to travel agents," Braintrust cofounder Adam Jackson told Fortune.
Time to hire, measured
Time to hire is where the return concentrates, because most of a 45-day median is queueing rather than deciding. Applications wait for review. Reviewed candidates wait for a scheduling slot. Scheduled candidates wait for a panel.
The independent evidence is now stronger than the vendor case. In a randomized field experiment covering roughly 70,000 job applicants, candidates routed to an AI voice interview were 12 percent more likely to receive a job offer than those interviewed by human recruiters, with higher job starts and retention. That is the strongest available answer to whether the speed gain costs you quality, and on the evidence so far it does not. We cover the study and its caveats in more detail in our explainer on what an AI interviewer is and how it evaluates candidates.
The queue is also getting longer. Robert Half found 67 percent of HR leaders reporting that AI-generated applications have slowed hiring, with 20 percent seeing delays of more than two weeks and 84 percent reporting heavier workloads. Application volume rose because candidates automated their side first. A screening capacity problem created by AI is unlikely to be solved by hiring more screeners.
What the ROI case usually overstates
An honest business case names its own weak points before finance does.
- Realised value is rare. Gartner found 88 percent of HR leaders saying their organizations have not realized significant business value from AI tools. Buying the tool is not the intervention. Changing the workflow is.
- Recovered hours are not banked savings. Unless headcount changes or requisition load rises, the hours are redeployed, which is valuable but is not a line item finance will credit as cash.
- Nobody measures quality of hire. Only 20 percent of organizations measure quality of hire at all, down from 27 percent in 2022. A quality-of-hire benefit you cannot measure should not appear in the model.
- Governance is immature. Deloitte reports 60 percent of executives using AI in decision making while only 5 percent say they manage it well.
- Bad-hire cost multipliers are folklore. The commonly cited figure of 30 percent of first-year salary traces to no retrievable primary source. Leave it out.
The metrics to track
Scroll to see all columns
| Metric | Why it belongs in the model | Measure it |
|---|---|---|
| Screening completion rate | The clearest early signal, and the one that moves first | Weekly, from application to completed screen |
| Time from application to first conversation | Isolates the queue that automation actually removes | Median and 90th percentile, not mean |
| Recruiter hours per requisition | Converts directly into the largest auditable saving | Sampled time study before and after |
| Agency-sourced share of hires | The second auditable saving | Quarterly, by job family |
| Pass-through rate by stage | Detects a screen that is too loose or too tight | Monthly, with adverse impact ratios by group |
| 90-day retention of AI-screened hires | The only credible quality check available early | Cohort against a pre-deployment baseline |
| Adverse impact ratio | Required for compliance, and a leading indicator of a bad rubric | Continuously, by protected group |
Scroll to see all columns
When AI recruiting is not worth it
- Low volume, high specialism. Ten senior hires a year through referrals and search will not recover a platform cost.
- The bottleneck is downstream. If offers stall in approvals, faster screening just lengthens the queue in front of the real blockage.
- No measurement baseline. Without a pre-deployment record of time to fill, completion rate, and recruiter hours, you will not be able to prove the result.
- No workflow change planned. That is the failure mode behind the Gartner finding above.
Where volume and speed do matter, Braintrust has published deployment results including a quick-service restaurant group expecting to save about 7,000 manager hours a year and a recruiting firm that placed a founding engineer in four weeks from more than 500 applicants. The budget-side argument against outsourced recruiting is set out in our analysis of AI displacing the RPO model.
Braintrust prices by interview volume, so per-interview cost falls as usage rises. To run this model against your own requisition load and salary bands, book a demo.
Frequently Asked Questions
Is AI recruiting worth it
AI recruiting is worth it when you have more applicants than your team can reach, a time to fill long enough that candidates drop out while waiting, and a measurement baseline to prove the change. Without those three, the return is thin.
How do you calculate the ROI of AI recruiting
Add recruiter hours recovered, agency spend avoided, and vacancy days avoided, then subtract platform, integration, and training costs. Weight the first two most heavily, because they are auditable in your own records, and treat vacancy value as upside.
What is the real time-to-hire impact of AI screening
Most of the gain comes from removing the scheduling queue rather than from faster decisions. Median time to fill is 45 days, and a randomized field experiment across about 70,000 applicants found AI voice interviews raised offer probability by 12 percent without a quality penalty.
What is a realistic cost per hire benchmark
SHRM's 2025 benchmarking data reports a median of $1,200 for non-executive roles and $10,625 for executive roles, while its earlier work put average cost per hire at about $4,700. Check which statistic a vendor is quoting, and from which year.
What is the payback period on AI recruiting
The recruiter-time and completion-rate effects appear within the first hiring cycle, which for a high-volume funnel is usually a quarter. Agency-spend reduction lags by a quarter or two because it depends on the internal funnel widening first. Retention effects take a year.
Will AI recruiting replace our recruiters
It replaces the screening and scheduling load, not the recruiter. The savings show up as redeployed hours and higher requisition capacity, which is why the model above counts recovered time rather than removed headcount.
Does AI recruiting reduce quality of hire
The available independent evidence points the other way, with higher offer rates, job starts, and retention in a randomized trial. The caveat is that only about 20 percent of organizations measure quality of hire at all, so most claims in either direction are unmeasured.
How much of the savings is recruiter time
Less than most vendor models suggest. Manual admin runs to about 17.7 hours per vacancy, of which roughly 6 hours is screening and scheduling. That is the automatable share, not the full 17.7.
What should we measure before deploying
Time from application to first conversation, screening completion rate, recruiter hours per requisition, agency-sourced share of hires, pass-through rate by stage, and adverse impact ratios. Capture all six for at least one quarter before go-live, or the after picture proves nothing.
