AI in recruitment: what a recruitment agency should automate
Our co-founder, Csenge Szántó, spent years leading a recruitment and selection team. In this article she shows what she would automate at a recruitment agency today, and what she would never leave to AI.
If you had asked me back then where my week went, I would have said interviews. In reality, most of my time went on everything except interviewing.
Updating the ATS after every call. Chasing CVs. The same follow-up email for the twentieth time. Candidate reports after interviews. Searching for candidates who might fit, then for clients who might hire someone.
I no longer work in recruitment, but I help recruitment agencies make their workflows faster and more efficient. I know agency processes inside out, and I know where time can genuinely be saved.
What does AI in recruitment look like in practice?
These are the areas we automate for our partners, and the ones I would start with if I owned a recruitment agency.
1. Admin: the work around the work
Most recruitment admin is copying information from one place to another. A CV arrives by email, someone types it into the ATS. The client confirms a start date, someone updates three spreadsheets.
What I would automate:
- Interview scheduling: the candidate picks a slot, both calendars update, the reminders go out
- Notifying candidates once the recruiter or the client has decided, so nobody waits three weeks for an answer
- Weekly client reports, straight from the ATS, not pieced together by hand on a Friday afternoon
- Internal management statistics: no recruiter should have to fill in a spreadsheet by hand just so everyone can see how many candidates were approached for which project, and where each selection stands
None of this needs judgement. It just needs someone to stop doing it by hand.
2. Sourcing candidates with AI
Sourcing is where AI saves the most hours, and where it is easiest to get wrong.
Start with your own database. Most agencies are sitting on thousands of candidates they have already placed or interviewed. One condition: only approach those who agreed to be contacted about future roles, and whose data you are still lawfully keeping. If you are not sure about that for your old database, that is where the clean-up starts.
Then write down the role’s requirements properly: what is essential and what is a plus. Based on that, the AI finds the people who fit the role, and drafts a personal message for each of them.
I would also have incoming CVs read into the ATS automatically: fields filled in, duplicates flagged, sorted by relevance. That way I am not the first person who has to read every CV, and the ATS genuinely works for me.
It can also build the longlist from the job boards and sourcing tools you already pay for. The recruiter then starts not with 300 unsorted results but with a ranked list, the 30 most promising names at the top.
What I would not leave to AI is rejecting candidates. The AI ranks, but the recruiter decides who drops out, and reviews the bottom of the list to do so. A candidate may simply have left something out of their CV and ended up lower in the ranking, even though they would be right for the role. This is also the most legally sensitive of all the areas above (more on that below).
3. Finding new clients: AI in business development
Every recruitment agency runs two pipelines: one for candidates and one for clients.
AI can watch for the signals that show a company is actively hiring: job ads in your field, new investment, a new office, new leadership. It then researches each company and tells you who to approach, and why now.
“We should be doing more business development” turns into a short list every Monday morning: companies in your field that are hiring right now, each with a reason to get in touch.
4. Voice notes and meeting recordings straight into the CRM
After a promising meeting you record a two-minute voice note in the car. Or the meeting recorder joins the client briefing. The AI turns it into structured notes: role, salary band, deadline, next steps. It then writes them into the right record in your CRM or ATS and creates the follow-up tasks.
No more typing up notes from memory in the evening. The notes are better, too, because they are written five minutes after the meeting, not five hours.
One rule: before recording, say that you are recording the conversation and ask the other person’s consent. A candidate or client should never find out afterwards.
5. Cold email with AI, done properly
AI-written cold email has a bad reputation, and mostly deservedly so. Thousands of identical “personalised” messages, a blacklisted domain, zero replies.
Done properly, it looks like this:
- Small batches, to people who fit a clear profile
- A real reason to write, from the research described in point 3
- Only to recipients you have a lawful basis to contact, with an easy unsubscribe in every email. Under Hungarian law, advertising emails to private individuals require their prior consent
- From a separate, warmed-up sending domain, so your main domain stays safe
- A human reads the first batch before anything goes out
- Replies go to a human, not to another automation
The AI does the research and writes the first draft. Who is worth writing to, and what you offer them, is still your call. We wrote more about why there always has to be a human behind the AI in an earlier article.
What I would never automate: the interview
There are tools in which AI conducts the interview, or scores candidates from a video recording. I would not use them, and some of them are no longer even allowed. Since 2 February 2025 the EU AI Act has banned systems that infer a person’s emotions from their facial expressions or voice in a workplace setting, and according to the European Commission’s guidelines this extends to job interviews.
The interview is the most human part of recruitment. It is where the candidate decides whether to trust you, and where you notice what no CV shows: how they think out loud, what excites them, where they hesitate.
The candidate is assessing you just as you are assessing them. If there is a bot on the other side of the conversation, the best candidates are the first to leave.
So automate everything around the interview, so your recruiters have more time for what actually gets results: the interviews, and the questions from candidates and clients.
AI in recruitment and the law: a note from the lawyer in me
The EU AI Act classes some AI systems used in recruitment as high-risk: those used to target job ads, filter applications and evaluate candidates. On the list above, that is point 2: ranking CVs and scoring candidates fall into this category.
A common misconception is that having a human make the final decision gets you out of this. The regulation does make an exception for systems that perform a narrow, preparatory task, but a system that evaluates a candidate based on their data (in legal terms, one that profiles them) is always high-risk. Human decision-making is therefore not a way out of the rules, it is one of the obligations.
The software’s developer has obligations around documentation, transparency and conformity assessment, among others. The agency using it must provide human oversight, keep the system’s logs, and tell candidates that an AI system is involved in assessing them. These rules apply from 2 December 2027; we wrote about the timeline here.
The GDPR applies too. Under Article 22, a candidate has the right not to be subject to a decision based solely on automated processing if it significantly affects them. This matters even when a human decides on paper: if in practice the recruiter only ever looks at the first few names in the ranking and never at the rest, the ranking has really made the decision. That is why the bottom of the list needs reviewing too.
The same regulation also covers old candidate databases. A CV received for a specific application cannot be kept for years and reused for new roles; that typically requires the candidate’s separate consent.
So everything in this article rests on the same principle: people decide about people. The principle alone does not release you from any obligation, but none of them can be met without it. AI prepares, the recruiter decides.
Where to start automating recruitment
Not with the shiniest tool, and probably not with the most expensive SaaS subscription. Pick the one task your recruiters complain about most and spend the most time on. It is usually call notes or scheduling. Automate that first. Neither is in the high-risk category, so the legal groundwork is lightest there too. Once the team sees an hour a day coming back, you will not need to sell them the next step.
At Andronia we build custom AI systems and software, tailored specifically to your business. If you are curious what this would look like at your agency, see what we can help with or get in touch, and let’s talk.
This article was published on the Andronia blog. Andronia helps businesses grow with AI solutions: custom software, automation and AI compliance. Read about our services here.
