Europe's first enterprise benchmark on integration and AI reveals why three in four AI projects fail to deliver and what IT leaders say needs to change.
"No one is talking about AI projects anymore in three years, ideally, because AI has become boring and so quietly embedded in everything we do."
Those were Prof. Dr. Moritz Behm's closing words at the State of Integration & AI 2026 broadcast: a vision of what success actually looks like when European enterprises get the infrastructure right.
Europe’s AI ambition is real. The investment is real, too. But according to the newly launched State of Integration & AI 2026 report, most organizations are still struggling to turn that ambition into measurable business value.
During the live broadcast hosted by Frends, leaders from Microsoft Germany, Sifted, Funnel, Österåker Municipality and academia came together to discuss the findings from the report, what is holding European enterprises back and what needs to change next.
The State of Integration & AI 2026 report is a new European benchmark on automation maturity, based on insights from more than 600 IT and business decision-makers across Europe. "This is probably the first work that I've seen which is really focusing on a benchmark that hits the spot with Europe," said Anna Kopp, CIO at Microsoft Germany.
→ Download the State of Integration & AI report
The report paints a picture of organizations eager to adopt AI, but constrained by fragmented systems, manual processes, governance concerns and decades-old infrastructure.
As Frends CTO Asmo Urpilainen summarized during the closing keynote:
“Three out of four projects relating to AI failed to deliver meaningful business value. And the key finding here is that the problem was not AI, it was integrations.”
Building a house
Opening the discussion, Prof. Dr. Moritz Behm used a metaphor that would shape the entire conversation: the difference between a well-built house and a fragile DIY construction.
He warned that many enterprises are attempting to layer AI on top of disconnected legacy environments that were never designed for modern automation or agentic AI.
Behm pointed to enterprise systems that are still 30–40 years old, with integration gaps that continue to create friction across organizations. He also highlighted one of the report’s most striking findings: “44 working days, this is the amount of work that has been completely wasted by the average knowledge worker in Europe.”
That lost time, he explained, is often spent on work that could already be automated today.
The argument was echoed by Frends CEO, Jukka Rautio. He argued that many enterprises are still operating from an architectural standpoint that resembles “roughly a 2010 level,” despite the rapid acceleration of AI capabilities. According to him, integration and centralized access to data are prerequisites for meaningful AI adoption.
“You need to have integration, you need to centralize things. Otherwise, you just inscrease the bad foundation and your operational cost,” he explained.
The report showed that organizations taking an “integration-first” approach achieved significantly faster project delivery and stronger AI ROI, reinforcing the idea that infrastructure maturity is becoming a competitive differentiator.
Governance, sovereignty and the European AI model
“There’s a lot of interest and want, but companies aren’t managing to actually put AI where it should be,” said journalist Mimi Billing, Europe Editor at Sifted. She was joined by Anna Kopp and Prof. Dr. Behm in one of the discussions.
Billing noted that while nearly all organizations are discussing AI governance and strategy, only a small minority have actually deployed it broadly: “We shouldn’t be too afraid of the regulation because it’s there for meaning, right?” The theme would become central throughout the discussions.
Anna Kopp emphasized that Europe’s AI reality differs fundamentally from that of the US or China. From data sovereignty and GDPR to labor protections and works councils, European organizations operate within a more complex governance environment, one that cannot simply be bypassed.
“We can’t have good AI if we don’t have good data,” she said. Kopp described manufacturers still operating with on-premise legacy systems that are decades old, making large-scale AI deployment difficult without significant modernization efforts.
AI agents put pressure on enterprise architecture
One of the clearest themes across the broadcast was the transition from AI assistants to AI agents, systems capable not only of answering questions, but performing autonomous actions across enterprise systems.
According to Asmo Urpilainen, this shift fundamentally changes the infrastructure requirements organizations must prepare for.
“We are moving away from assistants towards agents, where assistants, of course, answer questions and agents perform actions.”
Urpilainen explained that the value creation enabled by agentic AI could become exponential, but it requires meeting the integration demands beneath it.
The report found that while 97% of organizations are planning to implement AI agents, only 12% currently have the integrations or infrastructure necessary to support them.
Without those foundations, organizations risk introducing new operational vulnerabilities rather than efficiency gains.
“Without integration, AI is isolated. It can think, it can reason, it can answer questions, but it can’t act,” said Asmo Urpilianen, CTO at Frends.
Public sector and enterprise realities expose the same problem
Although industries differed, many speakers described remarkably similar obstacles.
Pontus Gustafsson, Data-innovation Manager at Österåker Municipality, explained that municipalities continue to struggle with basic digital maturity and fragmented communication between departments: “If we don’t talk to each other, we can’t integrate and automate.”
Meanwhile, Armin Catovic, Director of Data and AI at Funnel and Vice Chair of Stockholm AI, described how disconnected systems create conflicting versions of reality across organizations.
“Everyone’s going to have a different picture of the reality,” said Catovic.
The challenge, he argued, is making enterprise environments accessible, connected and governable enough for AI to operate effectively. The solution starts with consolidating and structuring enterprise data correctly before layering AI on top.
The organizations that close the gap first will pull ahead
Despite the challenges discussed throughout the broadcast, the tone remained optimistic.
The speakers repeatedly emphasized that Europe still has an opportunity to turn governance, trust and integration maturity into long-term competitive advantages. When asked what IT leaders should prioritize in the next months, Armin Catovic said:
"Integrate all the data in one place. Once you have that, it's incredible what you can do."
As the broadcast approached its end, host Sabinije von Gaffke summarized the core takeaway: “AI projects fail when systems don’t talk to each other. Fix the connections and the results will follow.”
In other words, as Prof. Dr. Behm put it, the organizations that get infrastructure, integration and governance right today may eventually stop talking about AI altogether, because it will simply become embedded into how the business operates.