Finance tends to ask for the return on investment (ROI) of an artificial intelligence (AI) recruiting tool at the most awkward moment: during the pilot, before a single hire has come through it. The textbook formula wants the value of the people you hired, and you don’t have any yet. So most teams either guess or tell finance to wait.

Three numbers from your own records answer the real question, whether the subscription is earning its keep, long before anyone is hired: cost per qualified candidate, recruiter capacity, and time to decision.

I’d keep the pre-hire case small and true. Hours saved is the figure everyone reaches for first, and on its own it rarely survives the first question from finance: payroll is the same size it was last quarter, so where’s the saving? This article gives you the answer before the meeting: how to work out the return from your own records, a break-even to write down before the pilot starts, and a worked example to rerun on your own numbers.

Why AI recruiting ROI breaks the textbook formula before the first hire.

The standard formula weighs the value your hires create against what it cost to hire them, and the guide to recruitment ROI as a measurement problem walks through it. Before the first hire, the value those hires create is still unknown. Retention, revenue per hire, and how well a new starter performs all need people who have been in their seats for months.

What you do have from the first week is your own records: the invoice, your applicant tracking system (ATS), and your recruiters’ calendars. They’re enough to answer the three questions finance is really asking:

• What are we paying for each qualified candidate, and is that cheaper than our other channels?
• What happened to the recruiter time the tool freed up?
• Are candidates getting a decision faster?

Until those hires arrive, the return you can calculate is narrower: the recruiter hours the tool frees that move into other work on record, valued at what an hour of recruiter time costs you, plus any spend the tool replaced, minus what the tool cost. The sections below build each part, and the break-even section puts them into one formula with a worked example. The value of the hires themselves comes later, and the business case for recruitment software covers how to price it once they arrive.

Set the baseline before you switch anything on.

Every number in this case is a comparison, so the before period has to be measured the same way as the pilot. Pick the roles and the stage the tool will touch, usually inbound screening or sourcing, and record the following over an equal period before go-live:

• Recruiter hours spent on that stage, from calendars or a simple weekly time log.
• Qualified candidates from that stage. A candidate counts as qualified when a recruiter advanced them to the next step and the hiring manager didn’t send them back within a set number of days, fixed before the baseline starts. If a candidate is still inside that window when a period ends, count them once it closes without a send-back.
• Three dates: when each application arrived, when a recruiter recorded the accept or reject, and when each requisition opened.
• What you already pay for the stage, such as job board add-ons or a screening tool you plan to replace.

Be honest about what a before-and-after comparison can show. It can’t separate the tool from a seasonal dip in applications, a change in role mix, or a recruiter on leave. If you have the volume, also run comparable roles without the tool in the same weeks, as a second check. With or without that check, the write-up should say “the pilot, compared with the baseline”, and not claim causation it can’t prove.

Cost per qualified candidate.

Finance will follow this number most easily. Take the tool’s cost for the pilot period, add the internal time spent setting it up (security review, legal, and systems work), and divide by the qualified candidates from the same period. Price that setup time at loaded cost, meaning salary plus benefits and overhead per hour.

Which candidates you divide by depends on where they come from. A tool that brings in its own candidates, such as a sourcing tool, is compared on all the qualified candidates it produced. A tool that screens candidates other channels bring in, such as inbound applicants from a job board, brings in no one new. It earns its cost by helping recruiters find more qualified candidates in the same pile, so count only the additional ones: the qualified candidates above the baseline. Dividing its cost across all of them would spread it over candidates those channels were already paid to bring in. If the pilot produces no more than the baseline, there’s nothing to divide by, and that’s the finding to report. The additional ones also carry the before-and-after caveat from the baseline, since part of any rise could be a busier quarter.

Then compare the result with your other paid channels: a job board, a sourcing seat, or an agency’s early-stage screening. Don’t compare it with cost per hire. The question it answers is narrower and more useful before the first hire: does this tool buy us qualified candidates more cheaply than the other things we pay for?

A screening tool only adds qualified candidates if the applications it ranks highest are ones your recruiters agree are worth advancing. Application Review, the Metaview tool that reads each inbound application against the role’s criteria and sorts it by fit, gives one read on this. In Metaview’s application data, among applications with a recorded recruiter decision, recruiters advanced the ones it rated a great fit at about three times the rate of poor-fit ones.

Read the 17.2% and 5.6% as agreement: recruiters at Metaview customers tended to advance what the fit ranking rated highly, which is the condition a screening tool has to meet before it can add qualified candidates. Two limits apply. Those recruiters saw the rating before they decided, so the agreement isn’t independent, and the figures don’t forecast your cost per qualified candidate. Your pilot will produce its own rate.

Application Review shows the reasoning behind each fit rating and never auto-rejects: every accept and reject call stays with a person.

Recruiter hours count once you can say where they went.

Most pre-hire cases stumble here. A saved hour looks like money, but your recruiters are paid the same whether they spent the week screening or doing something else. A case that multiplies saved hours by salary and calls the result a return will come straight back from finance.

The fix is to count a saved hour only when you can name the work it moved to and point to a record of that work growing against the baseline:

• More intake meetings or calibration sessions with hiring managers, on the calendar.
• More requisitions carried per recruiter, in the ATS.
• Sourcing for a hard role that would otherwise have gone to an agency. It enters the return once, as the agency fee you didn’t pay or as the hours at loaded cost, never both.

Value those hours at loaded cost. Hours with no destination on record count as zero, however busy the team felt. The rule is strict, but finance can check every hour it lets in.

Cockroach Labs, the company behind the CockroachDB database, is a useful test of this rule, because its Metaview case study reports both a saving and a plan for it. “With Metaview, our recruiting team has saved over 14 full work weeks,” says Lynette Estrada, vice president of global recruiting, in the Cockroach Labs case study, which credits automated notetaking and follow-up reports. When Metaview’s Notetaker records an interview, it captures every spoken word and turns the conversation into structured notes. That’s where the case study puts part of the saving: the team’s recruiters no longer have to type up detailed interview notes. Megan Mueller, senior manager of global go-to-market recruiting, names where that time is meant to go: “We’re really focused on our recruiters not just being tactical recruiters,” she says, “but transitioning to strategic hiring partners to our hiring managers and to the entire business.” Under this method, only the part of those 14 weeks that shows up on record as partnering work enters the return, such as more intake meetings and calibration sessions on the calendar.

Time to decision and time to a first qualified candidate.

Candidates and hiring managers notice speed first, and it’s easy to measure wrong. Two readouts cover it, both straight from your ATS.

Time to decision runs from the date an application arrived to the day a recruiter recorded an accept or reject. Measure inbound applications only, and report the median. The trap is the undecided applications. A median over decided applications looks great precisely because it leaves out the people no one got to. Siadhal Magos, Metaview’s co-founder and chief executive officer, described how that happens:

So keep every application in the count, in both periods, and rank each undecided one as slower than every decided one, since it’s still waiting. If more than half are still undecided at the end of the period, the median lands on an application with no decision date, so report the share that got a decision within a set number of days instead, and fix that number before the baseline.

Time to a first qualified candidate runs per requisition, from the day the role opened to the day a recruiter advanced the first candidate who counts as qualified. Report the median in days and list the requisitions still waiting. Neither readout gets a dollar value yet, because pricing a day of vacancy needs the value of the person who fills it, and that’s post-hire work.

The break-even, and a worked example.

To calculate the pre-hire return, multiply the hours moved into named work by your loaded hourly recruiter cost, add any spend the tool replaced, and subtract the tool’s cost. Divide that by the tool’s cost and you get a capacity-based ROI, not a cash return: most of it is recruiter time valued at loaded cost, and the worked example below reports the cash separately. The break-even is the number of hours the tool has to move into named work for the return to reach zero: the tool’s cost, minus any spend it replaces, divided by your loaded hourly recruiter cost.

Write the break-even down before the pilot starts, and set it against the hours the stage takes today. If the break-even needs most of the stage’s time, you know before signing that the tool has to take over most of the stage to pay for itself.

Take a team of four recruiters with a loaded cost of $75 an hour, running a 12-week pilot on inbound screening. Before the pilot, screening took each recruiter 10 hours a week. The vendor invoices $7,500 for the pilot, and setup takes 20 hours of internal time, also at $75 an hour.

At 10 hours a week each, screening takes the four recruiters 480 hours over 12 weeks. The 96-hour break-even is a fifth of that, and the pilot freed 192 hours, or 40%, though only the hours moved into named work count toward the break-even. The 42 freed hours that went nowhere on record count for nothing.

This team clears break-even by 54 hours, and 54 hours at $75 is the $4,050 return, an ROI of 45%. Had only 80 hours landed in named work, the return would be minus $1,200 and the team would be 16 hours short. The tool would be doing the same job either way. The difference is entirely in what the team did with the time, which is why I’d write the redeployment plan into the business case before the pilot starts.

Be clear with finance about which part is cash. In cash, the pilot cost $5,700 net: the $7,500 invoice minus the $1,800 add-on the team canceled. The $1,500 of setup time is already on payroll, so it stays in the $9,000 for the return and the break-even and out of the cash line. The $11,250 is capacity at cost, and it turns into cash only when it avoids spend, such as an agency fee or an extra hire you no longer need.

Because this tool screens applicants other channels bring in, set the $450 beside what you pay per qualified candidate on those channels: that channel’s cost for the period divided by the qualified candidates it produced. Label the two plainly, because they aren’t the same calculation: the $450 is the screening tool’s cost per additional qualified candidate, and the channel figure is that channel’s cost per qualified candidate. Dividing by all 110 would give about $82, the version the cost section rules out. The rise from 90 to 110 is still a before-and-after comparison, so present the $450 as the pilot compared with the baseline. Report the drop from 8 days to 3 in days, without a dollar figure.

If the renewal is a yearly contract, work the break-even out again for the full term before that conversation, instead of scaling up the pilot’s break-even: the year’s invoice, minus the spend the tool replaces over the year, divided by your loaded hourly recruiter cost. The setup hours were a one-off, so they stay out.

What to put in front of finance.

The pre-hire case fits on one page: what you paid, what you stopped paying, the break-even, where the freed hours went, cost per qualified candidate against your other channels, and the two speed readouts in days. Put the caveats on the same page. A finance team that sees the limits stated up front trusts the rest of the numbers more.

Numbers get the pilot approved, but someone still has to want to run it. In the video below, Siadhal Magos gives two ways to get a boss to buy into AI in recruiting: tie the tool to a business goal the team already has, or build a culture of experimentation that develops intuition about how AI can augment the work.

A pilot measured this way ties the tool to a hiring goal and gives the team a test to learn from. One warning: keep these numbers as a test of the purchase and never turn them into recruiter targets. A target on cost per qualified candidate is an invitation to advance marginal candidates.

If the hours moved into named work fall short of the break-even, report the shortfall and make the case for the tool on cost per qualified candidate and speed until the hires arrive, and then on a post-hire ROI. If they clear it, you have a return finance can check line by line, months before the first new starter’s review.