800,000 candidate interviews in 18 months, none of them conducted by a human.

That is the dataset Paul Lagrange, CEO of Omogen, has built in the time it takes most agencies to pick an ATS. And the picture it paints is the opposite of what the "AI will eat recruitment" crowd would tell you. The more you industrialise pre-qualification, the more the recruiter's job becomes about things no software can do.

That was the throughline of the second French edition of Recruitment Tech Talks- Watch the full webinar replay here. Vincent Baron (StudioDino) moderated a conversation between Paul and David Kieffer, co-founder of Achil, a network of independent recruiters whose near-daily LinkedIn market analysis has become a reference point for agency operators watching the sector shift. Paul came to recruitment from deep tech, having previously run operations at a company that hired more than 50 people through headhunters whose CVs he had to re-qualify himself. David came from finance and M&A.

Different routes in. Same conviction: industrialisation isn't the opposite of humanity. Used well, it's what gives recruiters back the time, the data, and the focus to do the human part of their job.

The data you only see once AI starts asking for it

The most original insight of the evening wasn't about time savings. It was about what systematic pre-qualification reveals about your acquisition strategy.

Paul gave a concrete example. A job spec asks for a driving licence and a specific certification. You call the candidates. It turns out 99% have neither. The conclusion isn't that the candidates are bad. It's that you are not buying the right candidates, or you are buying them from the wrong source.

A human recruiter would never catch this. The volumes are too high and the friction of manual pre-qualification too heavy. Once an AI agent pre-qualifies at scale, the pattern becomes unmissable.

The job-board economics behind this are part of why it happens. Job boards are paid on volume. Indeed launched sponsored flux, then started selling a filtering product on top of the flux it had sent.

"The flow isn't good, so now they sell you the filter for the flow they just sold you." — David Kieffer

Paul's number: 1 to 5% of candidates in a typical job-board batch actually match the criteria. That ratio will keep falling as mass-applying gets easier. The shared conclusion: stop reasoning in volume, start reasoning in quality.

TLDR

Paul Lagrange's team at Omogen has run 800,000 AI-led candidate interviews in 18 months. The time saved is real: a home-care staffing operator has compressed application-to-first-contact to under four hours, and candidate NPS on six-figure tech-sales roles sits at 4.6 out of 5. But the time saved is not automatic revenue. Without a deliberate rewrite of the recruiter's job description, those hours disappear back into the noise. The next 18 months won't reward the agencies with the most AI. They'll reward the ones that use it to do less sourcing and more consulting.

What it looks like in practice

Take Vitalliance, a home-care staffing operator. A candidate who applies at 10am can be in-agency by 2pm. Application-to-first-contact under four hours, from a process that used to run in days.

Paul's broader point: shaving one, two, three, even six days off time-to-staff doesn't just make the recruiter's life easier. For a candidate who needs income, that speed is the difference between paying rent this week and waiting for a callback that may never come.

And the white-collar side of the story runs against the grain. Paul pushed back on the assumption that AI pre-qual only works at volume, for blue-collar roles. Omogen's strongest candidate NPS, 4.6 out of 5, actually comes from the 100-300k€ tech-sales segment. Why? Because white-collar candidates prefer to do pre-qualification at 10pm from home, don't want colleagues to notice they're interviewing during working hours, and often find a context-aware AI better prepared than a human recruiter who picked up the brief five minutes before the call.

What none of this automatically changes is placement volume. Saved hours don't magically convert into more deals. The time has to be reallocated, and the reallocation has to be tracked and measured. Same as any organisational change. It doesn't happen overnight.

Your ATS is the most under-used asset you own

David pointed at a second data source that gets less airtime: duplication. Staffing clients rebuy the same candidates 30 to 60% of the time in interim. A typical ATS holds around 150,000 profiles that never get reactivated.

"You're filling a leaky basket with the same candidates while the candidate is already in your database. You just haven't spoken to them in too long." — David Kieffer

The agencies winning here aren't the ones buying more tools. They're the ones re-activating what they already own.

The macro picture, in one paragraph

David opened with a caveat on the headlines. Yes, the latest France Travail labour-needs survey points to levels below COVID. Yes, tech layoffs keep making news. But a meaningful share of what gets branded as "AI restructuring" is quiet cost-cutting dressed up in better language. When Amazon announces a 30,000-person layoff, the stock gains 2%. Companies hold revenue flat, cut headcount, and report the result as a productivity gain. Paul added the technical counterweight: the job category actually being reshaped right now is software engineering, where industry estimates suggest a large share of code is now written with AI-assisted agents. Beyond software, the impact is much more localised.

"No recruiter woke up AI-native." — Paul Lagrange

Why adoption is still at 11%

A LinkedIn study Vincent put on screen sets generative-AI integration at just 11% of organisations. Systems that are actually running, measured, and producing impact. Paul gave three reasons for the gap.

Recruitment is not a change-hungry industry. ATS contracts are typically signed for three or four years; the DNA of the sector doesn't push toward permanent reinvention.

Hardcore AI users, people actually paying for agentic solutions, are still around 0.1% of the world population. Globally, we are in an evangelisation phase, not a mass-adoption phase.

Plugging in a tool isn't magic. ROI only shows up when the organisation around the tool is rewired, and that part takes real time.

David added a fourth reason. There is a real gap between what AI promises and the proof points leaders can actually point to. A lot of deployments don't produce results, and people talk.

What the big players are actually doing

Randstad partnered with Salesforce and has reported a 15% time saving on placement, with better conversion rates. Korn Ferry went further and built a data business unit from the 10 billion data points sitting in its system, candidates, job movements, industry matches, monetised as a subscription product now close to 400M€ in revenue at around 30% margin. Newer AI-native plays like Jack & Jill internationally are chasing similar angles.

But don't read these as templates. Most agencies don't have ten billion data points. What they have is an ATS, a CRM, a team of recruiters, and a lot of manual work. The question is what to do with that.

The new recruiter job description

If industrialisation absorbs pre-qualification, what does a recruiter do with the time?

Both panelists had the same answer, but they came at it from different angles.

David's version: consulting.
The recruiter who spends freed-up hours with the client: understanding why the role opened in the first place, whether it's a retention issue, an employer-brand issue, a missed workforce plan. That creates value the commodity pipeline can't.

"We often arrive as firefighters. The consulting work is about making sure the client doesn't need to call us in the first place." — David Kieffer

Paul's version: seduction.
Once AI handles qualification, recruiter time goes to relationship: understanding the client's business well enough to sell the role authentically, and keeping candidates warm through a longer, more tailored process.

Both landed on the same underlying move. Rewrite the job description. If a consultant has 10 extra hours a week, what are they actually going to do with them? That question has to be answered deliberately, and the screening when hiring recruiters needs to update too.

David noted that Achil now explicitly tests comfort with tools, not just AI, basic computer literacy too, before bringing a recruiter into the network.

Key takeaways

The layoff headlines are real but not the full picture. A meaningful share of what is called "AI efficiency" is restructuring in better packaging.

The real AI tailwind for the industry sits in two places, sourcing and pre-qualification, where ROI is most measurable.

Adoption is still around 11%, not because the tools don't work, but because change management is where the work actually lives. A weekly time saving per agency only becomes revenue when the organisation around it is rewritten.

Your own ATS is probably the most under-used asset you have: 150,000 profiles, 30 to 60% candidate reuse rates across the industry, and a habit of re-advertising instead of re-activating.

The job-board business model is misaligned with your outcomes. Reason in quality, not volume.

The recruiter job description is now part consulting, part seduction, part tool operator.

The agencies that will outperform in 2026 won't be the ones with the most AI. They'll be the ones who have rewritten their own organisation around the time AI frees up, and who use that time to do the part of the job no software will ever do.

Watch the full webinar replay here.

This article is based on Recruitment Tech Talks, a webinar series by Talent Business Club bringing together recruitment industry leaders to discuss the strategies, technologies, and practices that drive agency success.