Contact
ManpowerGroup chose Hubert’s AI solution because they shared the same belief: AI should make hiring faster and fairer, not compromise one for the other.
The partnership sparked three years ago when Hubert won the Ethical AI Award at VivaTech, Europe's largest technology conference, where ManpowerGroup was the headline sponsor.
ManpowerGroup handles hundreds of thousands of candidates across dozens of markets every year. At that scale, speed is non-negotiable. But so is consistency, compliance, and a good candidate experience. Those three things are usually in tension, so the brief they gave Hubert was simple: prove they don't have to be.
In short: faster screening, happier candidates, and no drop in quality, at global scale.
Across ManpowerGroup's deployment, Hubert has delivered:
"When I see 85% of candidates completing the process, it shows the candidate experience is there," said Delfosse. "Otherwise, people would stop at the first step."
Responsible AI hiring means candidates and employers alike can trust the process: every score explained, every decision defensible.
Every AI vendor claims to be responsible. What actually separates them is how the AI makes its assessments under the hood. Most AI hiring tools today are built on large language models, or LLMs, the same technology behind tools like ChatGPT. LLMs are impressive, but they work by predicting what a reasonable answer sounds like, not by following a fixed logic. This means that when an LLM scores a candidate, it often generates a convincing-sounding explanation after the fact – not the actual reason it made the call. Fredrik Östgren, CEO of Hubert, calls this a black box. "By that definition, any LLM-based assessment is a black box, and that's not legally defensible."
Hubert works differently. It uses two separate layers. The first is the interview conversation itself: natural, warm, and conversational. The second is the scoring layer, which uses deterministic models: a type of AI that follows fixed rules, so the same answer always produces the same score, with a clear explanation you can actually trace. No guesswork. No post-hoc narratives. This is what makes Hubert's approach legally defensible by design, meaning if a hiring decision is ever challenged, there's a full, honest paper trail to stand behind.
The six pillars are fairness, explainability, quality, consistency, security and human oversight. These are all required for AI hiring to hold up under scrutiny.
Fairness - Every qualified candidate gets an equal shot, regardless of gender, ethnicity, age, or any other protected characteristic.
Explainability - The AI can show its reasoning in plain terms. Not a post-hoc story, but the actual logic behind every score and decision.
Quality - The tool measures what it claims to measure, and its scores actually predict how people perform on the job.
Consistency - Give the AI the same input twice and you get the same result. No randomness, no drift, no surprises.
Security - Candidate data is handled with a privacy-first approach. Sensitive information stays protected throughout the process.
Human oversight - Hubert never makes the hire. It surfaces the strongest candidates and a human always makes the final call.
Automating one task saves minutes; transforming the whole process saves weeks, and that's the difference between AI as a tool and AI as genuine workforce intelligence.
Most early AI deployments in recruitment automated isolated steps: parsing CVs, sending reminder emails, scheduling calls. The time saved was minimal because everything around that one step stayed the same. "If you automate one isolated part of a process, there are still so many other things that person needs to do," said Östgren. "There are no real efficiency gains from that."
Real workforce intelligence means rethinking the entire recruitment process from scratch. For ManpowerGroup, embedding Hubert’s AI interviews at the screening stage didn't just speed up one step, it freed recruiters to spend their time where it actually matters: building relationships, advising clients, and making better hiring decisions. "AI is the cape and the human is the hero," said Delfosse.
Yes, when built correctly, AI removes many of the biases that make traditional hiring unfair.
Traditional hiring filters on job titles and years of experience. The problem is that those measures often reflect privilege as much as ability, favoring candidates from certain universities or companies over those with equal talent but different backgrounds. Skills-based AI interviews assess what candidates can actually do, not how their CV reads.
NSS Group, a Hubert customer in the UK, saw a 50% increase in hires from candidates who would never have passed traditional CV screening. The AI surfaced talent the old model filtered out.
Candidate experience data backs this up too. In Hubert and ManpowerGroup's live webinar poll, 0% of attendees named data privacy as their biggest concern about AI hiring. Instead, the majority named candidate experience. The 9/10 satisfaction score and 85%+ completion rate across ManpowerGroup's deployment reflect candidates who found AI screening better than the alternative: opaque online forms, zero feedback, and weeks of silence.
What makes an AI hiring tool legally defensible? It must produce explainable, consistent, and auditable results. Under the EU AI Act, the European regulation that governs how AI can be used in high-stakes situations, recruitment AI is classified as high-risk. That means transparency, human oversight, and accuracy aren't nice-to-haves; they're legal requirements.
What's the difference between deterministic AI and LLMs in recruitment? A deterministic model follows fixed rules: same input, same output, every time. An LLM is probabilistic which means it generates responses that sound right, but the same input can produce different outputs on different runs. For fair, legally defensible screening, deterministic AI is the only architecture that holds up.
How should organizations approach AI hiring implementation? Start with the problem, not the technology. Identify the real bottleneck in your process, for example slow screening, inconsistency, high drop-off, and deploy AI that solves it end to end. Ask not just what the AI does, but how it does it. Start small, build a business case, then scale.
The next phase isn't about finding newer AI, it's about getting real value from what's already available.
"Companies invested a lot in AI but don't see the results yet," said Delfosse. "The biggest trend will be how to get the full value of AI already in the process." The ManpowerGroup partnership is a working model for organizations still evaluating: globally scaled, regulatory-grade, candidate-trusted structured AI interviews that prove speed and fairness aren't trade-offs.
Hubert and ManpowerGroup will both be at VivaTech in Paris in June. Come find us and continue the conversation!