AI agent for recruitment agencies that leaves the decision to the recruiter
One role pulls in a few hundred applications, and the candidate who waited a week for a reply has already taken another offer. We bring applications from every source into one list, send the confirmation within minutes and order the list against the hard requirements from the brief. Who gets invited to interview is the recruiter's call.
- 14 daysto the first working automation, counting from the end of the workshop
- EUR 3,500net per week of workshop, published up front and not negotiated
- 0 decisionsthat is how many decisions about a candidate the AI agent makes. It shortens the list to review, it never scores a person
The problem
Why does a good candidate disappear before the interview?
One vacancy can pull in 50 to 200 applications, arriving from four places at once: the form on your careers page, the job boards, messages on LinkedIn and plain email. The recruiter reads them in the evening, because the day goes on interviews. The confirmation goes out a week later, and by then the candidate has accepted another offer. You are losing on response time, not on the quality of your work.
The second loss is quieter. A candidate database built up over years sits unused, because finding anything in it means remembering the exact word somebody used to describe that role two years ago. So the next similar brief starts with a fresh job ad, while the right person has been on file for eighteen months and is changing jobs right now.
The third thing is a line nobody should cross. The GDPR, Regulation 2016/679, gives a person the right in Article 22 not to be subject to a decision based solely on automated processing where that decision produces legal effects concerning them or similarly significantly affects them. The AI Act, Regulation 2024/1689, places systems used in recruitment and selection in the high risk category, and the obligations then fall on the company deploying such a system as well, not only on the provider that built it. That is why here the agent orders the applications and a person decides about a person.
- A candidate takes another offer, because your confirmation arrived a week later
- A database built over years sits unused while the project starts with a fresh job ad
- The same work done from zero on the next similar role
- A status report for the client stitched together by hand from the ATS, the inbox and a spreadsheet
- A candidate waits on the client's decision and nobody remembers to chase it
Why now
What changed that makes this pay off today
Applicant tracking systems, job boards and mailboxes now expose interfaces that hand you an application a second after it is sent. Three years ago wiring four sources into one list was a project for a quarter. Today it is counted in working days. That is enough for the candidate's confirmation to go out in minutes instead of after the weekend.
The second change matters more. A rule will copy an application from a form into the ATS, but it will not answer the question who do we have on file with a valid forklift licence, available immediately, within 50 kilometres of Manchester. A language model reads job descriptions, interview notes and the text of applications at once and answers in a sentence, so a database built over years stops being an archive.
The third change is legal, and that one argues for starting now. Since recruitment sits in the high risk category, it is cheaper to build the process with the criteria written down before the start, with a query log and with a person at the point of decision, than to retrofit all that onto a tool somebody bought off the shelf because it promised to rank candidates.
What changes
What does the recruitment team get?
A confirmation in minutes, not in a week
Applications from the form, the job boards, LinkedIn and email land on one list. The candidate gets a confirmation naming the role and saying when a recruiter will be in touch.
One list instead of four sources reviewed separately
The list shortened against hard requirements
Experience, location, licences and availability, the things the client wrote into the brief. We write the criteria down before the project starts, so every application is measured against the same yardstick.
4 criteria written down before the start, identical for every application
Your own database answers the question
A recruiter asks in a plain sentence who is on file for a given role and gets a list of people out of your ATS. Nobody has to remember the word used to describe that role two years ago.
An answer from your own database in 5 to 10 seconds
An interview slot without ten emails
The candidate gets a link and picks a time from the recruiter's calendar, and for client interviews from the client's calendar too. The reminder goes out a day before.
Invitation, slot picking and reminder in one flow
The client report builds itself
How many applications came in, how many candidates are past the interview, who is waiting on a decision and since when. The report goes out on the agreed day, without collecting data from three places.
2 hours a week per client back with the recruiter
Nobody is left without an answer
Candidates waiting on a decision longer than the agreed threshold land on a chase list. The recruiter knows who to call today, before that person writes in themselves.
The list of people waiting longer than 5 days, every morning
Proof
Do you handle your own inquiries this way?
We do, and we will say up front what this proof does not cover. Since March 2026 inquiries at our founders' agency have been handled by an AI agent built with the method described above. That is inbound inquiry handling, not recruitment, so what we show is the scale of work taken over, not a recruitment result.
Built for ourselves. These are not client deployments but implementations inside the founders' own companies. We say so plainly, because proof you cannot verify is not proof.
JustBottheir agentJustIdea Agency, a performance marketing agency of around 30 people
- 23
- people from the company work with the agent
- 38
- systems wired in permanently
- 112
- processes run without a human
- 475
- thousand lines of the agent’s code
The agent is wired into the CRM, the forms on the website, Slack and the data warehouse. Inquiries arrive from forms, ads and referrals, and the agent creates the contact, checks for a duplicate and notifies the team. First commit 1 March 2026, today 23 people use it.
8 of 38 systems it works in
Google Ads
Meta Ads- HubSpot
- ActiveCollab
- Ahrefs
GA4
Search Console
BigQuery
- Before
- A person read the inquiry, created the contact and the company in the system by hand, then decided who to pass it to. Time to first response depended on who happened to have their inbox open.
- After
- The agent picks the inquiry up, creates the contact, checks for a duplicate by domain and notifies the team within seconds. A person comes in at the conversation, which is where a person is worth having.
3,432 operations on inquiries and 1,301 on forms in 30 days, with no human involved
And in your agency
What if we recruit differently?
Four sentences come up in recruitment agencies more often than any others, and all four are fair.
We do not want a machine rejecting people.
It will not. The AI agent does not reject a candidate and does not put a score on a person. It orders the applications and shortens the list to review against criteria you wrote down yourselves, and a rejection is the recruiter's decision, recorded under their name. The reason is a hard one: the GDPR protects a person from a decision based solely on automated processing, and the AI Act treats recruitment as a high risk area.
We have our own ATS and we are not buying another one.
You do not need to. We read the data through its interface and leave the system where it is: Bullhorn, Lever, Greenhouse, Workable or whatever runs at your place. Replacing an ATS is a project for a quarter, and the first automation is due 14 days after the workshop ends.
Our candidate database is a spreadsheet and an inbox, not a system.
That is more common than it sounds, and it can still be done. A spreadsheet and a mailbox are a data source like any other. During the workshop we check whether that base is worth tidying first or whether we can read from it as it stands. If there is too little data in there, we say so instead of selling you search over an empty set.
We are afraid the algorithm will start discriminating.
That is why we rule out in advance the attributes nobody may sort by: photo, age, gender, name, nationality, home address. What is left are the hard criteria from the client's brief, written down before the project starts. Every query to the database and every result stays in a log, so it is visible who asked what and what the system returned.
- Applicant tracking systems
- Bullhorn, Lever, Greenhouse, Workable, Teamtailor, a spreadsheet acting as the database
- Job ads and sourcing
- LinkedIn, Indeed, the job boards you post on, the form on your careers page
- Mail and calendars
- Google Workspace, Microsoft 365, recruiter and client calendars
- Client systems
- HubSpot, Pipedrive, client portals, per brief spreadsheets
- Reporting and communication
- Google Sheets, BigQuery, Slack, Microsoft Teams
- Automation layer
- Make, n8n, custom API integrations
What exactly we automate in a recruitment agency
A typical split of the work on a recruitment project before the rollout and after it. The last row is the most important one and it deliberately does not change.
| Area | Today | After the rollout |
|---|---|---|
| Incoming applications | four sources reviewed separately, in the evening | one list, a confirmation to the candidate in minutes |
| Hard requirements | checked from memory, a little differently on every CV | written down before the start, the same yardstick for everyone |
| Candidate database | searched by keyword, if anyone remembers it | asked in a plain sentence, answer in 5 to 10 seconds |
| Booking interviews | several emails for one slot | the candidate picks a time from the recruiter's calendar |
| Client report | stitched together by hand, around 2 hours a week | goes out by itself, on the agreed day of the week |
| Candidate waiting on a decision | chases you or disappears | the list of people waiting longer than 5 days, every morning |
| Decision about a candidate | the recruiter | the recruiter, with a log entry and a reason |
The scope of the first rollout is agreed in the workshop. We take the three items with the highest return, not all of them at once.
How we work
Five steps from the inbox to a process you can defend
Every step ends with something you can judge before you decide on the next one.
Workshop
One week. We walk one recruitment project from the client brief to the close and look for where applications sit waiting and where the same work is done twice. We come out with a process map and a list of the places worth fixing.
Writing the criteria down
The step nobody gets to skip. We write down the hard requirements that applications may be ordered by, and separately the attributes that must never be used. This is a document that stands up in front of a client and in front of a candidate.
Building the first automation
14 days to a working solution. In recruitment we usually start with one inbox for applications plus the confirmation to the candidate, because that is the piece that stops costing you candidates fastest.
Pilot on real applications
The automation runs alongside your current process, with nothing switched off. The recruiter compares the shortened list against what they would have picked and says where the criteria are too narrow or too wide. We correct them before anything is switched over.
Measurement and the next area
After 30 and 90 days we check the indicators agreed in the workshop: time to first response, how many candidates came out of your own database, how long a client report takes to build. Only then do we decide on the next area.
Before and after
How a recruiter's week changes
Before
- The evening goes on reviewing applications from four sources
- A candidate from a year ago is a perfect fit, and nobody remembers them
- One interview slot settled across eight messages
- Friday means stitching client reports out of the system and a spreadsheet
After
- Applications sit on one list, ordered against the requirements from the brief
- The database answers a question asked in a plain sentence, in seconds
- The candidate picks a time from the calendar, the reminder goes by itself
- The report reaches the client on the agreed day, with no stitching
What does not change is who decides about a candidate. What changes is how much time the recruiter has to talk to them.
For the owner of the agency
What the person signing this off gets out of it
The first thing shows up immediately: time from application to first response stops depending on who happens to have their inbox open. A recruitment agency earns on closing a role, and a candidate lost to silence costs exactly as much as one who fell out after the client interview.
The second is quieter, and it matters during an audit and in the hard conversation. The selection criteria are written down and every query to the database stays in a log. When a client or a candidate asks on what basis somebody did not make the shortlist, the answer is in the system rather than in somebody's memory.
- Hard criteria written down and versioned, independent of one person
- A trail in the log: who asked what and what the system returned
- The decision about a candidate stays with the recruiter, under their name
- A new recruiter learns from a written process instead of watching over a shoulder
What it costs
The workshop price is published, the rollout is quoted after it
The workshop costs EUR 3,500 net per week and we give that number before the first call, because an agency has a right to know what it is agreeing to. In recruitment one to two weeks is usually enough. The rollout is quoted after the workshop, once we know how many systems we are connecting to and what state the candidate database is in.
The cost of the language models and the automation tools you pay directly to the provider, or you pay nothing at all when everything runs on your own infrastructure. Expect roughly EUR 0.005 to 0.07 per query on the provider's side. No commission from us and no margin added along the way.
Process workshop
EUR 3,500 net
per week, one to two weeks in a typical recruitment agency
A self-contained piece of work. It ends with a map of the recruitment process and a 30 day plan you can execute without us.
- A week of work inside the agency and inside your systems
- A map of the project from client brief to close
- Hard criteria written down, and the attributes excluded from selection
- Three automations with the highest return, counted before the start
- The indicators to measure after 30 and 90 days
- Most popular
Workshop and first rollout
Quoted after the workshop
the scope follows the number of application sources and the state of the database
The variant most agencies pick. For a team that wants one inbox for applications within two weeks of the workshop ending.
- Everything in the workshop
- The first automation live 14 days after the workshop
- Connection to the ATS, the job boards and the mailbox
- A parallel pilot: the agent's list next to the recruiter's own picks
- Training for the person who runs it on your side
- Measurement after 30 and 90 days
AI agent for the recruitment agency
Quoted after the workshop
for teams that query the database a dozen times a day
For an agency where questions about candidates, statuses and dates come up all day and the answers live in three different places.
- Everything in the rollout
- An agent answering questions about the candidate database and project status
- Client reports built and sent automatically
- Chase lists for candidates waiting on a decision
- Access for recruiters and management, with a query log
If after the workshop we conclude that tidying up your message templates and job descriptions is enough, we say so plainly and propose no rollout. The workshop ends with a document you can execute yourselves or show to a client without us.
Frequently asked questions
What does automation mean in a recruitment agency?
It means a system taking over the work a recruiter does by hand today: collecting applications from four sources, sending confirmations, ordering the list against hard requirements, booking interviews and building the client report. The data for all that usually sits in the ATS, the inbox and a spreadsheet already, it just sits there separately. Automation is about pulling it into one flow, not about buying another platform.
Does the AI agent reject candidates?
No. The agent orders the applications and shortens the list to review against the hard criteria you wrote down before the project started: experience, location, licences, availability. It does not put a score on a person and it does not close the door on anyone. The decision to invite, to reject and to recommend to the client is the recruiter's, and it is their name that stays on it.
Is this compliant with the GDPR and the AI Act?
We build the process so that the decision about a candidate is not automated: the agent prepares the list, a person decides, and queries and results stay in a log. Article 22 of the GDPR protects a person from a decision based solely on automated processing, and the AI Act treats systems used in recruitment as high risk and puts obligations on the company deploying them as well. This is not legal advice, so the details get written down during the workshop together with your lawyer, and the practical side is covered in our <a href="/en/guides/ai-and-gdpr/">guide to AI and GDPR</a>.
Do we have to replace our ATS?
No. We read the data through its interface and leave the system where it is, whether that is Bullhorn, Lever, Greenhouse, Workable or a spreadsheet doing the job of a database. Replacing an ATS is a project for a quarter, while the first working automation is due 14 days after the workshop ends. If your system exposes no interface, we say so in the workshop and show what can be done on the mailbox and the spreadsheet alone.
How does the agent search our candidate database?
You ask in a plain sentence, for example who has a valid forklift licence, production experience and is available immediately around Manchester. The agent reads experience descriptions, interview notes and application data, then returns a list of people from your own database together with the reason each one is on it. The answer comes back in 5 to 10 seconds and the recruiter checks it before making contact.
What about candidates whose data we have held for years?
During the workshop we establish how long you keep candidate data, on what basis, and when consent has to be refreshed. Searching your own database only makes sense where you are allowed to process that data, so sorting out retention periods is part of the work rather than an add-on to it. Records past their date drop out of the base instead of coming back in results.
How long does it take from signing to the first effect?
A week of workshop and 14 days to the first rollout, counting from the end of it. Then the pilot runs alongside your current process, so the recruiter can see the agent's list next to their own picks. We measure the effect after 30 and 90 days, against indicators agreed before the start.
What does the rollout cost after the workshop?
It depends on the number of application sources, on the ATS and on the state of the candidate database. We quote after the workshop, because that is the first point at which we know the answer, and a number given earlier would be guesswork. The workshop costs EUR 3,500 net per week and you know that price before the first call.
Let us start with one recruitment project
The free consultation takes 30 minutes. You tell us where your applications come from and how long a candidate waits today for a first response. We tell you what can be tidied up straight away and what needs work on the database first. You do not have to give us access to anything at this stage.
We run at most two rollouts at a time, so that somebody is genuinely sitting on each of them.
Book a free consultationPS. We are not promising you better candidates or a process shorter by some percentage. We are promising a faster answer to the people who already applied, and use of the database you have been building for years. The decision about a person is still made by a person.