The setup work is less technical than most teams expect. Modern tools connect to the ATS you already use, so applications flow in and shortlists flow back without anyone changing systems. The real work in week one is human: deciding what a good answer looks like for your roles.

This is the step that determines everything downstream. An AI interview is only as good as the criteria behind it, so your recruiters and hiring managers define the competencies that matter for each role and how heavily each one counts. A useful way to think about it: your team sets the strategy, the tool executes it at scale. Teams that rush this step get technically correct shortlists for the wrong role. Teams that give it a day of honest discussion often find the exercise valuable on its own, because it forces agreement on what "good" means, something many hiring teams have never actually written down.

For European teams there is one more piece of week-one work: naming who holds oversight. The EU AI Act treats AI in recruitment as high-risk, which means a human must be able to interpret and override the system's outputs. In practice that is a decision about roles and documentation, and it is far easier to set up now than to retrofit later.

The go-live moment is undramatic by design. Candidates apply as they always did; the difference is that an interview invitation arrives in minutes rather than weeks, and they complete it whenever suits them, mostly on their phones, often at night. Across deployments, the median time from invitation to starting the interview is 1 minute. Nobody waits for a callback slot anymore.

For recruiters, the change shows up inside the ATS: instead of a pile of unread applications, a ranked shortlist with the evidence behind each score. The pace of this can be startling. Aleris, one of the Nordic region's largest private healthcare providers, went from signing to live in 5 working days. OKQ8, which handles more than 23,000 applications a year across its Swedish petrol stations, was live within two to three weeks, and its team needed no formal training to work with the results.

Here is the honest part: the hard phase is not technical, and it starts after go-live. For the first couple of weeks, most recruiters do exactly what you would do. They double-check the tool. They read a shortlisted candidate's full answers, compare the score to their own instinct, and occasionally disagree.

Let them. This checking phase is not resistance to be managed; it is the implementation. Every score in a well-built tool traces back to specific answers against the criteria your team set, so disagreements become concrete conversations about criteria rather than arguments about a black box. Some teams discover a weighting was wrong and adjust it. That is the system working. Recruiters who have interrogated the scoring and seen it hold up use the tool with confidence; recruiters who were simply told to trust it never fully do.

The other early habit worth building is watching the candidate side. Completion rate is the number that tells you whether applicants find the interview reasonable, and it is visible from day one. If it is healthy (deployed customers report figures between 88% and 96%), your process is respecting people's time. If it drops, the interview is too long or the invitation unclear, and both are fixable in an afternoon.

By week four the novelty is gone, which is the goal. Screening happens while your team sleeps. Recruiters spend their hours with the shortlist, not the pile. Every assessment is documented, so the audit trail your compliance colleagues asked about in week one is simply accumulating on its own. The interesting question stops being whether the tool works and becomes what your team does with the reclaimed time.

At Hubert, implementation is the part we are proudest of: structured AI interviews, live in your ATS in days, with scoring your recruiters can interrogate until they trust it. If you want to see what your first month would look like, book a demo here!