This Belgian Startup's Bet on Recruitment's First 30 Seconds

TLDR: A small cluster of European startups is betting that the first phase of candidate contact, screening, calling, scheduling, matching, is about to become an AI job. Talio, founded in Belgium by Jarne Verstraete and Stijn Bruggeman, is one of the more focused entrants. The sharper argument from Verstraete isn't about his product. It's that building your own version of this is harder than it looks, and the agencies that try will lose the race they're trying to win.

A recruiter's day still looks a lot like it did in 2015. Screening calls, voicemails, WhatsApp follow-ups, re-scheduling. Most of it is repetitive. Almost none of it is where a recruiter actually earns their fee.

A small but visible group of European startups is betting that the first phase of candidate contact, the part before any meaningful conversation, is the next domain to be automated. Talio, founded in Belgium by Jarne Verstraete and Stijn Bruggeman, is one of the more recent entrants. They ran close to 200 customer-discovery conversations before locking in the product direction, and they are now backed by VLAIO, KBC Start it X, and Bryo.

What sets them apart is not technology, it is scope. Carv, the Dutch player most often named in the same breath, runs end-to-end across the recruitment workflow. Ringtime, the Ghent-based startup, is the one most often cited for outbound calls into blue-collar candidates. Talio has positioned itself narrowly on the first phase only: voice or chat interviews at scale, outbound screening on phone and WhatsApp, conversational job matching on the agency's own website, and the scheduling and feedback layer that sits after the screen.

"The things we focus on are the things that ideally shouldn't be done by the recruiter, so the recruiter can focus on the tasks where they add unique value."

Everything is white-labelled, so candidates see the agency's brand and not Talio's. Everything is built integration-first, sitting on top of Bullhorn, Salesforce, Workday or Carerix rather than asking the team to log in to yet another tool. The voice agent works across more than two dozen languages and can switch mid-call without breaking the voice.

The 80% trap

The sharpest part of the conversation isn't the product. It's Verstraete's view on what every agency CEO is now quietly considering: building it themselves.

"It looks really easy to build something, and you can get to something that's 80% good very quickly. But that last 20%, the fine-tuning, the compliance, the security, getting it to actually work at scale, takes much longer than the 80% before it. And even when you hit 100%, within a week the model has moved on and you're outdated again."

That last sentence is the one to underline. The pace of model improvement means a vibe-coded screening agent isn't a one-time build. It's a permanent maintenance project, against a benchmark that shifts every quarter. For a recruitment business whose actual job is placing people, that's a category of work most teams shouldn't be taking on.

Verstraete isn't arguing every internal build is wrong. He's arguing that the line between "tool we can build in a weekend" and "tool that genuinely behaves well at scale" is wider than founders realise. The cost of being on the wrong side of that line accumulates in candidate experience, compliance exposure, and employer brand.

The startup paradox

The other admission worth surfacing is rare to hear so plainly from a founder selling into the enterprise.

"Our use cases are more relevant the bigger the company gets. But the bigger the company, the more often the question becomes: you're a startup, are you still going to be around in a year?"

This is the recruitment-tech buying decision in one sentence. The agencies who most need this work done well are also the most exposed to vendor risk. Talio's answer, like most early-stage answers, is to run paid pilots and let the outcomes do the convincing. That works, but slowly.

Why this matters now

Voice models have crossed a threshold. The team uses OpenAI's real-time voice-to-voice model as the core, with ElevenLabs as a fallback for cases where a chained architecture is preferred. The candidate feedback they hear most often isn't that the agent sounds robotic. It's that the agent sometimes jumps in on a thinking pause too quickly.

That detail matters more than it sounds. The threshold for an automated first call to be tolerable to a candidate has been crossed. The remaining work is calibration, not capability.

Which leaves a single question for any agency leader. The first phase of candidate contact is going to be automated. The only real decision is whether your agency runs that automation, buys it, or watches a competitor do both before you've made up your mind.

The race isn't whether voice AI gets to the front door of recruitment. It's already there. The race is who notices first.