How to use AI in talent acquisition

Most organisations don't have an AI problem in talent acquisition. They have an adoption problem. The tooling has matured faster than the operating model around it, and the result — across the EMEA functions I've spent the last decade inside — is a familiar pattern: pilots that never scale, sourcing teams running two parallel workflows, and Partners or executive sponsors who can't articulate what good looks like.

This guide sets out a practical view of how to use AI in talent acquisition at scale: where it genuinely changes the work, where it doesn't, and what a sensible adoption path looks like for a function operating across 10, 20 or 30+ countries.

Where AI genuinely changes talent acquisition

Strip away the noise and there are four places AI changes the day-to-day of a hiring function in a meaningful, defensible way:

• Sourcing and market mapping. Generative tools, when paired with a structured data source, materially speed up the construction of long-lists and competitor maps. The cost of a first-pass market view has collapsed.
• Screening and shortlisting. Used carefully — and within EU AI Act constraints for high-risk use — AI can reduce manual CV triage time substantially. Used carelessly, it embeds bias at industrial scale.
• Interviewer and hiring manager support. Structured interview-guide generation, summarisation of feedback, and consistency checks across panels are quiet but compounding wins.
• Reporting and TA analytics. Natural-language queries over ATS data make TA leaders less dependent on a single analyst and shorten the loop between question and answer.

Note what is not on that list: candidate experience as a headline benefit, fully automated hiring decisions, or replacement of recruiters. Those are the narratives that get pilots cancelled when reality lands.

The EMEA reality: why a single playbook doesn't work

A talent operating model that works in the UK and Ireland will not, without adjustment, work across France, Germany, Spain, Italy, the Nordics, the Middle East and Türkiye. The EU AI Act treats most recruitment use cases as high-risk; Germany's works councils have a meaningful say in what monitoring or screening tools may be introduced; data protection authorities across the Nordics interpret transparency obligations more strictly than other jurisdictions.

The practical implication is that an AI adoption plan owned centrally has to be designed for local accountability from the start. The model I've seen work — across 30+ EMEA offices at Baker McKenzie and 9 at Fragomen — is a small central spine that sets standards, evaluates vendors and runs the legal and risk conversation, with local TA leadership owning what is actually switched on, on what timeline, and with what disclosure to candidates.

A six-dimension diagnostic before you choose a tool

Before evaluating vendors, run the function through six dimensions. This is the diagnostic I use in Newbery Advisory engagements, and it tends to surface the real blockers in two to three weeks rather than two to three quarters.

1. Data foundations. Is your ATS data clean, structured, and complete enough that any AI layer is working from signal rather than noise?
2. Process maturity. Are your hiring processes standardised enough that automation reinforces good practice rather than scaling bad practice?
3. Capability and confidence. Do your recruiters and hiring managers have the judgement and the language to use AI outputs critically?
4. Governance and risk. Are roles, accountability, audit logging and the EU AI Act compliance picture clear?
5. Vendor and architecture. Will the tools you're considering sit cleanly inside your existing ATS and identity stack, or create a parallel system of record?
6. Change and adoption. Is there a credible plan for the human side — communications, training, feedback loops — or just a rollout email?

What a 12-month adoption path looks like

A realistic adoption path for a mid-sized EMEA TA function — 200 to 1,000 hires a year, in-house team of 10 to 40 — sequences in three roughly four-month phases:

• Months 1–4: Foundations and one defensible pilot. Run the six-dimension diagnostic, fix the worst of the data issues, and pick a single, low-risk use case (sourcing assistance, interviewer summary support) to pilot in two or three markets.
• Months 5–8: Operating model and governance. Codify what central owns, what local owns, what is logged and reviewed, and how candidates are informed. This is where most programmes that have skipped the governance work begin to stall.
• Months 9–12: Measured expansion. Roll the validated use case into additional markets, layer a second use case behind it, and start the conversation about TA analytics maturity.

The questions to ask before you write the business case

• Which hiring decisions are we comfortable letting AI inform, and which must remain entirely human?
• What does the EU AI Act classification look like for each use case we're considering?
• Where does accountability sit when an AI-supported decision is challenged by a candidate?
• What is the smallest version of this we could honestly call a success in six months?
• Who, by name, owns each of the six diagnostic dimensions above?

If you can answer those five questions clearly, you are unusually far ahead. If you can't, that is the work to do before tooling — and it is exactly the work I help People & Talent leaders work through.