This month’s column is about verifying your vendor’s AI, why it has become easy to buy AI but hard to know what you actually get. The summer has been quiet on the Nordic front. Globally, far more has happened. Fosway measures how few of vendors’ AI promises are actually in live use, the EU AI Act’s transparency requirements have started to apply, and Deel shows how a vendor’s pooled workforce data becomes a market asset. Among the global news are Oracle moving from promise to named AI agents in HCM and Leapsome leading with EU hosting and regulatory compliance as selling points.
🌟 And a piece of good news on a personal note: for the third year running I am on HR Executive’s Top 100 HR Tech Influencers globally. Read more here.
Upcoming events and training within HR Tech and AI can be found as always at hrdigi.se/news/event.
_____________________________________
📣 This month's invitation, together with Teamdash
AI in talent acquisition: building the TA function for 2027 and beyond
On 25 August, Anna Carlsson will co-host an invite-only breakfast roundtable at Fotografiska Stockholm together with Teamdash and Renita Käsper. The conversation will cover how AI is changing talent acquisition, what teams need to prepare for next and what should remain human in hiring.
The event is designed for a small group of senior TA and HR leaders to keep the conversation practical, focused and useful. Places are limited and participation will be confirmed individually.
👉 Read more and register your interest!
_____________________________________
Aren’t you already subscribed to the newsletter?
⇢ Register your e-mail here for the newsletter in English.
⇢ Sign up with your email here for the newsletter in Swedish.
No vendor says they lack AI any more. The features and the agents are in every proposal, every demonstration and every feature list. By AI we mean here the features and agents that HR system vendors build into their platforms, from CV screening and candidate dialogue to automated workflows and decision support. But the gap between what AI is claimed to do and what it does in practice has now been measured by the analyst firm Fosway. They find that only a good quarter of the AI features recruitment vendors say they have are actually in live use at customers, the rest sit on the vendor’s roadmap (their plan for the future) or are under development. Vendors’ feature lists are filling up faster than the features go into use, and it falls to the buyer to judge what will genuinely work and what is merely a promise, or even a wish that may never be delivered, neither with the features promised nor on the timeline given.
The summer’s launches show how easily the two are confused, that is fact and plan. When Oracle announced that Google’s Gemini models would become available in the Fusion suite, including HCM, it sounded like a finished AI feature. In reality it was a product direction, a direction for the agent platform, not something that can already be switched on in the HR module. A buyer who does not ask the question risks paying for a roadmap and believing it is a feature. The right questions are concrete: which functionality is actually deployed at customers today, where is the HR data processed, and what is required for the promised service to work in your own environment.
The difficulty rarely lies in what the vendor shows, but in what is not visible. Workday launched Agent Passport in the spring to verify and trace AI agents in HR processes. At the same time the analyst house RedThread pointed out that the questions vendors are not really talking about yet, such as accountability, change management and total cost of ownership, are now the buyer’s to ask. It is about traceability, about who carries the responsibility when an agent makes the wrong decision, and about what AI costs when it is baked into the suite instead of priced separately. That global vendors such as Leapsome now highlight regulatory compliance, data storage within the EU and the open MCP protocol as selling points is a sign that trust is built on how AI works, not only on what it does.
For HR this is both a risk and an opportunity. The risk is that our culture of caution, and sometimes our lower technical skill, leads us either to wait for the perfect solution or to buy the prettiest demonstration. The opportunity is that whoever dares to scrutinise their vendor’s AI properly, how it is trained, where the data lives, who owns it, how decisions can be traced and what it costs over time, makes better decisions than the one who settles for the feature list. Finding AI is easy. Verifying it is hard. But it is that scrutiny which sets apart the organisations that get real value from their vendor’s AI from those that only get an impressive demonstration.
More on Fosway, Oracle, Leapsome and Eightfold can be found in the news items below.
Would you like help with your AI questions? Get in touch with me directly at anna.carlsson@hrdigi.se
The news is grouped into Nordic news, regulation and policy, research and reports, and global product launches.
Norwegian Zalaris, which runs payroll and HR administration for many Nordic SAP customers, reported revenue of 344 million Norwegian kronor for the second quarter on 14 August. The business is running as usual, but the result is weighed down by costs for the takeover. Kona BidCo has taken over the company and is preparing a delisting from the Oslo exchange. If you have Zalaris in your payroll chain, now is the time to ask what happens to the product plans once the company becomes privately owned.
A common misconception this summer is that the entire EU AI regulation has been postponed to 2027. It has not. What was moved, through the Digital Omnibus, are the requirements for high-risk AI in recruitment and people management, which now apply from 2 December 2027. The rest applies. The bans on certain AI and the requirement for AI literacy among staff have applied since February 2025, and the transparency requirements started to apply on 2 August.
A short Q&A on the two parts that raise the most questions, transparency and AI literacy:
What do the transparency requirements mean? That people should understand when they are meeting AI. If the customer service chat is a bot it must be said, and images, audio or video manipulated by AI must be labelled.
Does all text that AI helped write need to be labelled? No. Only text published to inform the public about matters of public interest, and not even then if a human has reviewed it and someone is accountable.
What does the AI literacy requirement mean? That those who use AI at work understand the tools well enough to use them wisely. It has applied to all employers since February 2025. No certification is required, but you must be able to show how the competence was built.
We only use ChatGPT, does it apply to us? Yes. The requirement applies as soon as AI is used at work, whatever the tool and whether or not you procured it.
Thanks to Josefin at Clarity Works for a clear summary of the situation, which this walkthrough builds on.
Fosway, one of Europe’s leading HR analysts, examined what recruitment vendors actually deliver behind their AI promises. Only 27 percent of the promised AI features are in live use at customers, the rest sit on the roadmap or are under development. Ahead of a procurement, the lesson is to separate a demonstration from a deployed feature.
Atlassian’s HR lead Alicia Lenart describes a new HR role that plans capacity and allocates responsibility between employees and AI agents, rather than strict AI mandates from above. It is one of the more concrete examples of how a larger technology company organises the collaboration between human and agent, an area where many HR functions still lack ready answers.
Johannes Sundlo went through 3,350 HR job ads in the United States and 20 European countries over the summer. Only 5.4 percent set AI as a hard requirement and a further 4.1 percent as a merit. While HR urges everyone else to build AI skills, it is rarely valued in HR’s own ads, and the requirement profile is a clear way to show that the knowledge is wanted.
The Josh Bersin Company and RPO firm AMS published a playbook on 7 July showing that AI in talent acquisition is now being scaled across whole organisations rather than in isolated pilots. The recruiter becomes an orchestrator of AI-driven processes, and success is measured in business results rather than time to hire. For Nordic TA teams still experimenting on a small scale it gives a clear direction.
The global HR platform Deel showed in a press release on 7 July that Swedish employees more often than any other country take at least 15 consecutive days of summer holiday, based on actual leave taken at customer companies. What is interesting for HR is that a vendor with 40,000 customer companies can turn its pooled workforce data into a market asset. Turn the question around in your next procurement: what patterns can the vendor see in your data and everyone else’s, who owns them, and how does that weigh in when you evaluate the vendor?
Workforce Skills Forecast 2026, a collaboration between ServiceNow and Pearson presented on 14 August, places project management at the top of the list of Sweden’s future skills needs through to 2031, ahead of individual technical skills such as Python. The top ten also includes leadership, communication, problem solving and collaboration. Technical skills are still growing fastest in percentage terms, but the forecast points to the ability to execute as the Swedish bottleneck in the AI transition.
Oracle made two announcements in a row, and the difference between them is exactly what this month’s column is about. On 30 July Google’s Gemini models became available in the Fusion suite, including HCM. It sounded like a finished feature but was a product direction. On 11 August came the concrete part, Fusion Agentic Applications for HR with named agents for role design, skills profiles, manager coaching, internal mobility and workforce planning, plus the option to build your own automations in AI Agent Studio. If you are an Oracle customer, the question is which of the two your licence actually gives you access to today.
👉 oracle.com, Fusion Agentic Applications
Eightfold AI launched Candidate Agent on 15 July, answering candidate questions about jobs, pay and benefits in more than 24 languages and handing over to the company’s AI interview agent. Agent-based candidate experience is becoming a recurring theme ahead of coming procurements, and the question is how the transparency requirement in the EU AI Act is met when the candidate first meets an agent.
Leapsome, with more than 2,000 customer organisations, launched a built-in ATS in early access, AI-generated workflows and an MCP connection, described as adapted to the EU AI regulation and EU hosted. That a player with a strong position in the Nordics highlights regulatory compliance and data storage as selling points, not just features, is worth noting.
The newsletter covers developments in the HR Tech market and explores how digitalisation, AI and innovation are shaping the future of HR and business.
About me, the author
My name is Anna Carlsson and I'm a strategic advisor and HR Tech analyst. I help HR and management make better decisions around HR Tech, AI, and digitalization – through strategies that are clear, sustainable over time, and implementable. My work ranges from one-hour advisory sessions to larger projects, supporting organisations from initial assessment through to implementation and results.
I run HR Digi, which supports HR practitioners and HR Tech companies to navigate, understand, and lead in a digital age. And it's not always a big change that's needed – sometimes it's enough to clarify, prioritize, and communicate strategically to achieve success.
Read more at hrdigi.se or Book a virtual coffee with me.
Also a proud member of the group Top 100 HR Tech Influencers globally 2024, 2025 & 2026🙏
