What AI in recruitment means for an agency
AI in recruitment, for an agency, is software that does the reading, matching, chasing and logging around a placement so the recruiter's hours go to candidates and clients. Done properly it cuts admin and shortens time to fill, and it leaves the relationship and the hiring decision with a person.
I'm Michael Batko, co-founder of Hourglass AI. We build and maintain AI for Australian companies of 20 to 100 people with no engineers on staff, and recruitment agencies are one of the client types we work with. The job is half conversations and half paperwork. The conversations are the business. The paperwork is where AI pays back.
The story: a recruiter who books interviews with one voice note
Story Recruitment
One-person CFO recruitment firm. 9 years in the role, a network of 30,000+ finance professionals.
Before: Tom booked first-round interviews by email, late at night, between 8 and 10pm. Ten candidates took him a couple of hours.
What we built: an AI agent he talks to in Telegram, connected to his email and calendar. He sends the CVs and the role as a voice note. The agent emails every candidate with the role and a time, books the replies into his calendar, sends the confirmations, and tells him who is booked and who has not replied.
After: 10 interviews booked from a single voice note, a saving he puts at thousands of dollars, and four more recruiters have come to Hourglass through him.
“I could send it 10 CVs of the people that I want to talk to and it goes away and emails them, books in times in my calendar and sends them confirmation of it, details that they need to know and all of that is done while I'm just sitting on my couch, talking to it on my phone, not even typing.”
That is one agency and one workflow. Across a whole desk, the same kind of admin adds up to a lot more.
Where an agency loses time and money
Five pain points recur across the research on recruitment desks. For each one I have put a public figure next to a plain assumption so you can check the working. Two assumptions run through all of them. A recruiter works four vacancies a month, the same assumption the Totaljobs study reported by HRreview used. And a recruiter's time costs about $42 an hour, from SEEK's average Australian recruitment consultant salary of $75,000 to $90,000, taken at the $82,500 midpoint over a 38-hour week, before super and on-costs. These are estimates, and conservative ones.
Reading applications that a model could rank
A live role brings more applications than a recruiter can read properly, so most get a skim. Totaljobs found recruiters spend 3.6 hours reviewing applications for every vacancy. An AI layer can read the full stack against the client's brief and hand back a ranked list with a written reason beside each name. A person still decides who goes forward.
How we estimated this
3.6 hours per vacancy (Totaljobs, UK, 748 HR leaders and 2,025 jobseekers, August 2025) x 4 vacancies a month x 12 months = 173 hours. 173 hours x $42 an hour = about $7,200.
Booking interviews by email, one reply at a time
Cronofy's survey of 272 recruiters using JobAdder, the Australian-founded ATS, found 60% send three or more emails or calls to book a single interview, and 41% take three days or more. A fifth of the companies in it had a no-show rate of 20% or more. Totaljobs puts scheduling at 2.5 hours per vacancy. This is the job Tom handed to his agent first, and it is the one I would start with.
How we estimated this
2.5 hours per vacancy (Totaljobs) x 4 vacancies a month x 12 months = 120 hours. 120 hours x $42 an hour = about $5,000. The Cronofy figures are from 2019.
Writing up calls and keeping the ATS current
After every interview someone types up the notes and files them against the right candidate and job. Totaljobs measured 3 hours of post-interview notes per vacancy. When it slips, the data rots. The Access Group found 74.4% of agency leaders name poor data and disconnected CRMs as a top blocker to getting value from AI. AI can transcribe the call, write the note in your house format and file it, with the recruiter checking it before it saves.
How we estimated this
3 hours per vacancy (Totaljobs) x 4 vacancies a month x 12 months = 144 hours. 144 hours x $42 an hour = about $6,000.
Searching for candidates you already have
Most agencies hold thousands of candidates in the ATS that nobody has opened in two years. Bullhorn's APAC respondents spend 14.6 hours a week searching for the right candidates, and they expect AI to save 4.2 hours a week of it on search and match. That is a prediction from the agencies themselves, not a measured result, so it is the number I use.
How we estimated this
4.2 hours a week (Bullhorn GRID 2025, APAC, agencies' own prediction) x 48 working weeks = 202 hours. 202 hours x $42 an hour = about $8,400.
Slow replies that lose the candidate
The cost here is the placement, not the hours. Totaljobs found 28% of jobseekers had abandoned a hiring process because of poor communication, delays or too many rounds. Bullhorn ties speed to revenue: agencies that grew in 2024 were 34% more likely to fill roles in under 20 days, and more than half of the agencies that lost money were over that mark. An agent that answers a candidate at 9pm, when they replied, keeps them in the process.
How we estimated this
Both figures are as published. The 41% is from Cronofy's 2019 survey, the 28% from Totaljobs (UK, 2025). I have not put a dollar figure on a lost placement, because fees vary too much from desk to desk.
What admin costs a recruitment desk each year
Estimates from the figures above, at $42 an hour and 4 vacancies a month| Admin task | Per recruiter | 5-recruiter desk |
|---|---|---|
| Reading applications | 173 hrs, $7,200 | 865 hrs, $36,000 |
| Booking interviews | 120 hrs, $5,000 | 600 hrs, $25,000 |
| Call notes and ATS updates | 144 hrs, $6,000 | 720 hrs, $30,000 |
| Search time AI is expected to save | 202 hrs, $8,400 | 1,010 hrs, $42,000 |
| All four | 639 hrs, $26,600 | 3,195 hrs, $133,000 |
On those numbers a five-recruiter desk spends about 3,200 hours a year on work a machine could do or speed up. That is more than one and a half full-time recruiters, before you count a single lost candidate. Finlay, my co-founder, wrote about one client who went from sorting 600 emails a day by hand to running the tool around the clock.
The fix: four stages, admin first
The order matters more than the tool. My rule for any small business, from AI automation for small business, is to automate the annoying admin first and the core of the business last. The admin is low risk, the win is obvious, and the team gets used to AI before it touches anything that matters.
- 1
Interview booking and candidate follow-up
First buildAn agent on the recruiter's email and calendar proposes times, books replies, sends confirmations and chases silence. Contained, measurable, and no judgement calls.
Takes off the desk: pain points 2 and 5
- 2
Inbox triage and reply drafts
Second buildOur AI Inbox reads, sorts and drafts replies in your voice. It never sends and never deletes. A person approves every message. We keep a comparison of AI email assistants current if you want to see the options.
Takes off the desk: the daily email pile
- 3
Call notes into the ATS, and a company AI brain
The foundationThe company AI brain reads email, the CRM, drives and meeting notes every night and works inside the Claude or ChatGPT your team already uses, keeping each person's permissions. Most companies skip this layer and go straight to agents, and the agents break because nothing sits underneath them. More on why this is a knowledge layer, not a data warehouse.
Takes off the desk: pain point 3
- 4
Screening, database search and BD briefs
Once criteria are checkedWith the brain in place, AI ranks applicants against the brief, searches your own database for matches, and writes a research brief before a client call, the same way an AI SDR researches prospects.
Takes off the desk: pain points 1 and 4
A person makes every shortlisting and hiring decision. The AI reads and ranks, and it rejects nobody. Screening criteria get checked for proxies before go-live and at every review: age, career gaps and non-traditional careers are the usual leaks. Candidates are told an AI tool helps sort applications. And candidate data stays in your own accounts. The OAIC says "privacy obligations will apply to any personal information input into an AI system" and recommends keeping personal information out of public AI tools, so a CV pasted into a free chatbot is exactly what the regulator warns against.
The other common worry is shallow adoption: a licence bought, a few job ads written, nothing changed. The Access Group found agencies that invested significantly in training reported strong ROI at 43.8%, against 2.9% for those with minimal training. In my experience, set out in how to run an AI audit, most AI strategies fail because nobody mapped the real pain points before buying anything.
How we build it at Hourglass
AI Audit
$2,500 flat
7 to 14 days. We interview up to 10 of your people and give you a ranked list of what to build, with hours and dollars on every line. Go ahead within 30 days and the full $2,500 comes off the build. How the audit works.
Build
$5k to $25k
Per build, about two weeks each once the foundation is in. It goes live inside the tools you already use. No migration, no new platform.
Care
About $1,000 a month
Per build. Model upgrades, fixes, and a monthly review of what is being used and what to build next.
Everything runs on your servers and your accounts. We never store your data, you own the IP, and the system is not tied to one AI provider. To be blunt about what we are not: we are not the people to clean up your database or implement your CRM. We put an AI layer on top of whatever you run, JobAdder, Bullhorn or otherwise, and work with the data as it is. If you want to compare approaches first, we keep lists of AI audit providers in Australia and AI automation agencies.
Free 30-minute call
Get the admin off your recruiters' desks
30 minutes with Michael or Finlay. Tell us how your desk runs and we'll show you what we would automate first, with the hours and dollars.
Start with the $2,500 audit, credited in full to the build.
Get in touchRelated reading
Common questions
Is recruitment going to be replaced by AI?
No. AI takes over the reading, matching, chasing and logging around a placement, which is where most of a recruiter's week goes. The judgement calls, the client relationship and the conversation that persuades a candidate to move stay with a person. Agencies that hand the admin to AI free up more hours for those conversations.
Is it ethical to use AI in recruitment?
It is, if four conditions hold. A person makes every shortlisting and hiring decision, the screening criteria are checked for proxies such as age, career gaps and non-traditional careers, candidates are told an AI tool is used, and candidate data stays inside the agency's own accounts and never goes into a public chatbot.
The OAIC is clear that privacy obligations apply to any personal information put into an AI system, so the last condition is a legal one as much as an ethical one.
How is AI useful in recruitment?
It reads every application against the brief and returns a ranked list with reasons, searches the database you already own for matches, books interviews and chases replies around the clock, writes call notes into the ATS and prepares client research before a BD call. Bullhorn's APAC respondents expect it to save about 4.2 hours a week per recruiter on search and match. In Hourglass' one recruitment build so far, a Sydney CFO recruiter got 5 to 10 hours a week back from interview booking.
What does AI for a recruitment agency cost, and where should we start?
Hourglass starts with a $2,500 AI Audit over 7 to 14 days that produces a ranked list of what to build with hours and dollars on each line, credited in full against a build started within 30 days. Builds are $5k to $25k each and take about two weeks once the foundation is in, with care at about $1,000 a month per build.
Start with interview scheduling and candidate follow-up, then email, then call notes, before touching screening. Details are on [the AI Audit page](/products/ai-audit).