AI note taker ROI for recruiting agencies: the formula and a worked example

The ROI of an AI note taker for a recruiting agency is the value of the recruiter time it saves (or the extra placements that time produces) minus everything it costs, divided by what it costs. In practice the answer turns on one number you have to measure yourself: minutes saved per screen, from the end of the call to notes saved and the submittal sent. Every figure on this page is an invented example for showing the arithmetic. Replace each one with your own before you put it in front of anyone.

Most teams never do this. In SHRM's research for The State of AI in HR 2026 (published April 2026), 56% of HR functions said they do not formally measure the success of their AI investments, and only 16% use ROI as a metric. An agency that measures can at least tell whether the tool is paying for itself.

What is the formula?

INPUTS
  R  recruiters using the tool
  S  screens (or interviews) per recruiter per month
  M  minutes saved per screen       (measured in your trial)
  H  loaded hourly cost of a recruiter
  T  tool cost per month
  O  other monthly costs: review and admin time, valued at H
  U  one-time setup cost: training and configuration time

MONTHLY VALUE OF TIME SAVED
  V = R x S x M / 60 x H

ONGOING MONTHLY ROI
  ROI = (V - (T + O)) / (T + O)

FIRST-MONTH ROI
  ROI = (V - (T + O + U)) / (T + O + U)

BREAK-EVEN MINUTES SAVED PER SCREEN
  M* = (T + O) / (R x S x H / 60)

Loaded hourly cost means salary plus benefits, payroll taxes and overhead, divided by hours actually worked. If you pay commission, leave it out of H: commission follows placements, and placements belong in the separate capacity scenario below.

Where does the value actually come from?

An AI note taker can produce value in four ways. Count only the ones you can measure.

Speed matters on a contingency desk, and the screen to submittal workflow shows where the hours go between a call and a client inbox. But a faster submittal is only worth money if it changes who the client interviews, and that is hard to prove from one quarter of data.

How do you measure minutes saved per screen?

1. Baseline week. Before the trial, each trial recruiter times five ordinary screens: the clock starts when the call ends and stops when the notes are saved in the ATS (and, for a submittal, when it is sent).
2. Settling-in week. Ignore the first week of the trial; recruiters are still learning where things go.
3. Measured weeks. Time five screens per recruiter per week in weeks two and three, including the time spent checking the AI notes against the transcript.
4. Use the median, not the average, per recruiter. One messy call should not decide the case.
5. Keep the template the same. If the submittal template changes during the trial, you are measuring the template.

The checking time belongs in the measurement. A tool that saves 15 minutes of typing but needs 10 minutes of correcting saves five.

A worked example: a five-recruiter agency

All inputs below are invented for illustration.

V   = 5 x 60 x 12 / 60 x $40   = 3,600 minutes = 60 hours x $40 = $2,400 a month

Ongoing cost        T + O      = $250 + $120          = $370 a month
Ongoing ROI                    = ($2,400 - $370) / $370 = 5.49, about 549%

First-month cost    T + O + U  = $370 + $560          = $930
First-month ROI                = ($2,400 - $930) / $930 = 1.58, about 158%

Break-even          M*         = $370 / (5 x 60 x $40 / 60)
                               = $370 / $200           = 1.85 minutes per screen

The break-even line is the most useful one. In this example, the tool covers its ongoing cost if it saves each recruiter less than two minutes per screen. That shifts the question from "does it pay?" to "are the notes accurate enough that recruiters stop retyping them?", which is what the trial should test.

How sensitive is the answer to minutes saved?

Same invented agency, ongoing costs of $370 a month, different measured savings:

If your trial lands near the top row, the case depends on accuracy and adoption, not price. If it lands near the bottom, look at why: recruiters still typing during calls, phone screens not being captured, or a submittal template the tool's output does not fit.

Should you count extra placements?

Time saved is only worth money if it is used. On an agency desk the natural use is more screening, and more screening can mean more placements. Model that as a separate scenario, with your own conversion rates from recruiting agency metrics:

CAPACITY SCENARIO (all numbers are examples)

Hours saved a month                          60
Share reused for extra screening             50%   -> 30 hours
Minutes per extra screen, prep to write-up   45    -> 40 extra screens
Screen to submittal                          30%   -> 12 submittals
Submittal to client interview                40%   -> 4.8 interviews
Client interview to placement                20%   -> 0.96 placements a month
Average fee                                  $15,000

Expected extra fee a month = 0.96 x $15,000  = $14,400

Three rules keep this honest:

• Do not add it to the time value. The same 30 hours cannot be counted once as $1,200 of saved cost and again as $14,400 of fee.
• Check demand first. Extra screens only produce placements if there are job orders without enough candidates. A desk short of job orders converts saved time into business development, not placements.
• Report a range. Change each conversion rate by a few points and the result moves a lot. Show the low and high case, and say which assumption drives it.

Which costs do teams forget?

• Review time. Someone has to check AI notes, especially in the first month. Put it in O.
• Setup and training. Configuration, templates, the consent routine, the first session. Put it in U.
• Minimum seats and annual commitments. A plan with a seat minimum costs more per recruiter for a small desk.
• Phone setup. If phone screens must run through a computer to be captured, a softphone or dialer may be a new cost.
• Leavers. Seats paid for recruiters who have left, and the time to reassign their notes.

Published prices for the main tools, with free tiers and a worked ten-recruiter budget, are in AI interview note-taker pricing.

How do per-seat and per-hour prices change the math?

Per-seat pricing rewards volume: the more screens a recruiter runs, the lower the cost per screen. Per-hour pricing suits a desk with uneven volume or a part-time recruiter. Work out cost per screen both ways:

Per seat (example):  $50 a month / 60 screens          = $0.83 per screen
Per hour:            pack price / hours in the pack x screen length in hours

As an example of per-hour pricing, as of October 2026 Interview Signal (our product) sells pay-as-you-go hour packs alongside its subscriptions: 1 hour for $9, 5 hours for $29 and 15 hours for $69, bought once and never expiring. At the 15-hour price, that is $4.60 an hour, so a 30-minute screen costs about $2.30 in transcription. Current plans are on the pricing page. Whatever tool you price, use the same formula above with T set to what your volume would actually cost.

How do you present the case?

One page is enough. Put these lines on it, in this order:

1. Measured minutes saved per screen, per recruiter, with the baseline and trial dates.
2. Monthly value of time saved, with H stated.
3. Total monthly cost, including review and admin time.
4. Ongoing ROI and break-even minutes.
5. Accuracy: corrections found in manager review during the trial.
6. The capacity scenario as a separate range, with its assumptions.
7. What you will measure in month three to confirm it, and who owns that check.

If the case passes, the rollout guide covers the pilot, training and the problems that show up in week two. For the wider set of desk measures to report alongside it, see recruiter productivity metrics.