Sweco is one of Europe’s leading engineering and architecture consultancies, with more than 20,000 employees in 14 countries. Headquartered in Stockholm and listed on Nasdaq Stockholm, it plans and designs buildings, infrastructure and communities. In Sweden, Sweco runs national graduate programs that bring recently qualified engineers into consultancy roles, including a dedicated program for careers in rail. The team uses ReachMee as its ATS, into which Hubert is integrated.
The challenge: graduate hiring at volume, by hand
Sweco runs several graduate programs in Sweden, including its rail graduate program. Each spring the rail program receives 500+ applications from recently graduated civil engineers. Few have work experience, so the team has to judge two things a CV rarely shows: whether candidates have the right qualifications, and whether they suit a consultancy career in rail.
The old process was manual at every step. The team screened CVs by hand, sent cognitive tests, and then ran 80 to 90 phone interviews before candidates met with a hiring manager. The volume was so high that Sweco brought in an external recruitment consultant every spring just to support with the phone screens.
The knock-out questions in the ATS weren’t always reliable, so unqualified applicants often slipped through to the cognitive test stage. Sweco ended up paying for tests whose results it could never use.
“We paid a lot of money for tests whose results we couldn’t use later, because the candidates weren’t qualified. In those volumes, you cannot go into each CV and really thoroughly check.”
Anna Torninger, Talent Acquisition Specialist, Sweco
There was a softer problem too. Some candidates looked excellent on paper, but the final interviews revealed they weren’t suited to consultancy work. By then, the candidate had been through every earlier stage.
Building trust in AI: the compliance journey
Sweco is forward-looking on AI and was actively exploring tools that could make recruitment more efficient. As AI used to select candidates is classified as a high-risk use case under the EU AI Act, the use of tools such as Hubert is subject to Sweco’s internal processes for carefully reviewing and assessing high-risk applications.
“For a large company like Sweco, we have internal AI policies for high risk cases under the EU AI Act. So this needed to be gone through in a very thorough way, to ensure it was safe for us to use.”
Sara Stjernberg, Head of Talent Acquisition, Sweco
The review started with Sweco’s local data privacy lead in Sweden. The final decision sat at group level, with the global data privacy lead and the IT security director for Sweco Group. Most of their questions focused on how Hubert’s AI is trained and how it handles bias. They also worked through Hubert’s documentation on EU AI Act compliance, over many meetings and emails.
Hubert uses AI in two distinct layers, which made those questions easier to answer. Conversational AI runs the interview, so it feels natural and human for candidates. Hubert’s own deterministic models then do the scoring, so the same answers always get the same score, and every result can be reproduced and explained. Demographic details are never part of the scoring.
Four or five months in, the project nearly stalled.
“Our internal process was very time-consuming. There was a point when I felt we would never be able to pull this off. But then we decided to keep going and see if we could get it to the finish line.”
Sara Stjernberg, Head of Talent Acquisition, Sweco
What unblocked it was a new approach: joint meetings between Sweco’s group IT security director, its data privacy lead and Hubert, so the decision-makers could get their questions answered directly. Approval arrived before the rail program was due to go live.
Even then, approval was closely monitored. Sweco ran Hubert as a pilot first and reported the results back to legal and IT security before signing a full contract.
The solution: structured interviews, governed by people
The approval didn’t depend only on the technology. Legal and IT security also needed to see exactly how Sweco would work with it. The team set clear internal rules, and those rules are what made the exception possible.
- Ranking is never used alone. Recruiters always review Hubert’s interview notes and supporting CVs or full interview transcripts before making a decision.
- Weekly quality checks. During every recruitment, the team samples candidates from each of Hubert’s result groups (not qualified, qualified below average, qualified above average) and asks: would I have made the same call?
- Decisions owned jointly. The approval was granted by legal, IT security and HR jointly. All teams had to agree on the compliance and safety of Hubert for use at Sweco.
“They needed to really understand how it works, all the techniques, use of AI, deterministic models, the biases, and then be okay with how we would work together with Hubert. This way we ensured it’s secure to use AI in Sweco.”
Sara Stjernberg, Head of Talent Acquisition, Sweco
In the new process, every applicant completes a structured, competency-based Hubert interview, with scores delivered straight into their ATS, ReachMee. Recruiters can click through to the full interview for the most promising candidates. Cognitive tests now come after the Hubert interview, so they only go to qualified applicants.
The team built the interview with Hubert’s customer success team and Sweco’s hiring managers. Alongside qualifications, the interview questions test what matters most for this program: genuine motivation for a long-term career in rail, and fit with consultancy work.
“Hubert is a way to see where we find a candidate with drive and motivation for this specific career path, and who is just applying to get a job.”
Anna Torninger, Talent Acquisition Specialist, Sweco
The impact: time back, money saved, fairness evidenced
No more phone screening, or seasonal consultant
Hubert has replaced phone screening entirely in the recruitment process for the graduate programs. Sweco no longer brings in an external recruitment consultant each spring to handle them, and recruiter time on manual screening is down 80%, or 606 hours a year.
“We have saved both money and time using Hubert, and it’s really easy to work with. Previously we needed to take in an external recruitment consultant to manage all the screening interviews manually. Now we can hand all that to Hubert.”
Sara Stjernberg, Head of Talent Acquisition, Sweco
Unqualified applicants are identified early, so tests now go only to candidates who have passed the Hubert interview. Across the efficiency gain in staff time and cognitive tests now being used on qualified candidates only, Sweco estimates annual savings of SEK 435,000, a 4.7x return on investment. Average time to hire has fallen from 52 to 30 days.
Fewer, shorter manager interviews
Hiring managers have already read candidates’ answers, so their human interviews can be more focused: presenting the role to candidates, checking it’s someone they want to work with, and making sure the feeling is mutual for a better candidate experience. More manager interviews now end in a hire, with interview-to-hire conversion estimated to rise from 27% to 33%.
Stronger candidates at the final stage
Managers now meet candidates who have the right qualifications and suit consultancy work, a fit that used to show up only at the final interview.
“When you have built an excellent template and fair process, it makes a significant difference in how efficiently you can find the best candidates.”
Anna Torninger, Talent Acquisition Specialist, Sweco
A candidate experience graduates value
94% of candidates complete the interview, rating it 8.2/10 on average.
“For the newly graduated engineers who took part in the digital interview, it wasn’t strange or something that made them feel disconnected from us. Instead, they saw it as a great way to demonstrate their skills early in the recruitment process.”
Anna Torninger, Talent Acquisition Specialist, Sweco
Fairness, evidenced twice
Hubert’s bias analysis of Sweco interviews found no bias. Average scores for gender bias showed women and men differed by no more than 4% in any job category, well within the 20% threshold set by the EEOC’s four-fifths rule. Sweco’s own weekly checks support the result.
Advice for TA leaders facing the same compliance review
Sweco’s experience offers a practical playbook for any organization bringing AI into hiring.
- Know your internal AI policy. If it restricts high-risk AI, expect a deep review and plan for it from day one.
- Make it a joint decision. Bring in legal, data privacy and IT security early. The decision shouldn’t sit with HR alone. Bring all stakeholders into one room.
- Define how people stay in control. Decide up front how recruiters will review AI output, and how you will monitor quality and bias.
- Start with a pilot. A pilot with a clear report gives decision-makers real evidence.
- Get decision-makers talking to the vendor. When progress stalls, direct meetings with the vendor answer questions faster than email chains.
“An important learning for us was that HR should not make this kind of decision alone. We worked closely with Legal and IT security, and being involved together in the decision, and that made us feel confident.”
Sara Stjernberg, Head of Talent Acquisition, Sweco
What’s next
Hubert now supports other Sweco graduate programs alongside rail, including the T&D Academy, as well as standalone high-volume, entry-level roles that can attract hundreds of applicants, where scale makes human review alone a challenge. Together with legal, Sweco plans to explore how future AI reviews could move faster.
