Every ATS shows "AI" on its homepage. Behind that, there are three very different levels: AI that writes, AI that sorts, AI that acts. Here's what each one brings to an agency, along with the limitations we see in production, and the legal framework.
A recruiter at an agency spends part of their day on tasks that don't require judgment: reading forty CVs to shortlist six, writing the same follow-up email, logging a call report, scheduling an interview, updating a status. That's where AI delivers a measurable return. It delivers much less on what makes an agency valuable: understanding a poorly expressed need, convincing a passive candidate, negotiating a counter-offer. This article distinguishes three levels of AI, explains what each one brings, and what we see at agencies running Profilio in production.
It's the one everyone has: job descriptions, emails, LinkedIn messages, CV summaries. It saves a few minutes per piece of text and above all removes the blank page. Its limits are well known: without context, it produces a generic listing; without the real brief, it makes things up. It's only useful when connected to the job's data (the brief taken over the phone, the CV, the exchange history) and to the agency's tone (one template per recruiter or per agency). The gain becomes real once writing disappears from the task list: the call summary is written during the call, the listing at brief-taking, the follow-up at the scheduled time.
Common limitation: this AI is only as good as the data. A talent pool imported with free-text locations ("Brussels", "Ghent area") won't geocode itself; a scanned CV with no OCR won't get parsed. Setup starts with cleanup.
This is what's called autopilot or an agent: the AI no longer just suggests text, it runs a sequence under the recruiter's control.
The payoff at this level is the biggest and the most conditional: it assumes the recruiter accepts letting the tool act, with a log of everything it did, and spends the freed-up time on what AI doesn't do. An agency that turns on autopilot but keeps its recruiters on data entry has gained nothing.
Three pieces of feedback keep coming up at agencies using Profilio, and they align with reviews left from within the app : drafting time (call reports, interview reports, emails) is what disappears fastest; matching changes the way you search the talent pool, provided it's clean; automated follow-ups and calendar reminders surface applications that would otherwise have gone nowhere. The friction point is always the same: the first few weeks are spent tuning the templates to the agency's tone and cleaning up the data. We don't publish a "time saved" percentage: it depends on volume, the type of roles, and what the agency does with the time freed up.
The European AI regulation (AI Act, Regulation 2024/1689) classifies systems used for recruitment and selection as high-risk: targeted job ad distribution, screening and filtering of applications, candidate evaluation. The main obligations fall on the system provider (risk management, data quality, documentation, transparency, human oversight, robustness), and the professional user must use it in line with the instructions, with genuine human oversight, and inform the people concerned. In practice, for an agency: AI proposes, the recruiter decides; candidates are informed that an automated assistant is calling them or screening applications; GDPR applies to all data processed (legal basis, retention period, right of access). Ask your vendor where the data is hosted, which models are used, and whether your data is used to train them.
See also: the AI features of each ATS, compared · choose an ATS for a Belgian agency · all resources.
Parsing, matching with a real radius, call transcription, multichannel autopilot, AI calls during business hours, EU hosting. All of it in the Autopilot plan at €129 per user per month, free migration.
Book a demoUpdated September 21, 2026.
No, and that's not the goal. It handles the repetitive tasks: reading CVs, writing, following up, screening objective criteria by phone, scheduling, summarizing. Evaluation, negotiation, and the client and candidate relationship stay with the recruiter, who spends the freed-up time on that. The agencies getting results are the ones that redefined the recruiter's role, not the ones that bought a button.
Yes, as long as you disclose that it's an automated assistant, respect reasonable hours, let the candidate decline and speak to a person, and handle data in line with GDPR. An AI system that influences recruitment decisions falls under the EU AI Act's high-risk systems, which requires its provider to ensure transparency, human oversight and documentation.
It complements it. Semantic matching finds profiles whose title doesn't match word for word (“projectingenieur”, “chef de projet”, “project engineer”) and ranks them by closeness to the job. Boolean search is still useful for a strict criterion (a certification, a language). The real gain comes from combining both with the geographic radius.
With the metrics the ATS already produces: time between taking the brief and the first shortlist, number of candidates contacted per recruiter per week, follow-up reply rate, rate of qualified candidates before interview, placements per recruiter. Measure one month before, one month after; if the tool doesn't produce these numbers, that's already an answer.