Some of the team attended the European Chatbot and Conversational AI Summit 2025 to catch up with industry peers, join some informative sessions and get stuck into our favourite topic: conversational AI!
Alice Kerly, The CAI Company Co-Founder and Principal Consultant, shares her key takeaways from the event.
Reflections from the European Chatbot Summit: Balancing Innovation, Strategy, and Real-World Impact
Last week’s summit in Edinburgh was an exciting deep dive into the evolving world of Conversational AI. It’s fascinating to see how the landscape continues to change; but I feel a little perturbed by some of the expectations and narratives around emerging technology choices.
NLU Bots vs. LLM-Based Bots – A Misleading Distinction
One of the more surprising trends I noticed was a distinction being drawn between “NLU bots” and “LLM-based bots.” The implication seems to be that “NLU bots” associated with rule-based or machine-learning models and are now outdated, while “LLM-based bots” are newer and somehow automatically more desirable. But this framing isn’t particularly helpful.
At the core of any chatbot or virtual assistant is natural language understanding (NLU) – the ability to comprehend what a user is saying. Whether a system is built with an LLM or a traditional intent-based NLU model, it still needs to understand user intent and respond accordingly. Even in an LLM-driven system, some form of intent recognition occurs, making it, by definition, an NLU bot of a kind.
Rather than a distinction between “NLU bots” and “LLM bots” I think the line (or grey blurry boundary!) is actually between deterministic and generative bots. However, the focus shouldn’t be on choosing between them, but rather on selecting the best approach for the task at hand.
Adopting LLMs – The Right Balance
I felt a noticeable difference between two attitudes or approaches at the summit. On one side were the advocates of LLM (Large Language Model) based bots (“move fast and break things”, “don’t be scared”, “so much potential compared to “NLU bots”). On the other were those emphasising a more measured, requirements-driven approach, to ensure technology choices align with business needs, risk appetite and customer expectations.
I find myself in the second direction. Yes, LLMs hold great potential, and to avoid them out of fear is not the answer. However, adoption simply because they’re the latest trend is equally misguided.
Success in Conversational AI comes when you understand your use case, design an effective user journey, and then select the best technology to support that experience – whether that includes an LLM or not.
Rui Teimao from Lloyds Banking Group had some great words to say on this point.
Design, Customer Experience, and Brand Reputation are Vital
A major theme throughout the summit was the importance of customer experience and brand reputation. AI-driven interactions need to be more than just “natural” – they must be appropriate, controlled, and reliable.
Some LLM-based approaches risk producing unpredictable responses, which could damage a company’s reputation if left unchecked. AI must serve the end user’s needs, not just impress with its fluency.
The real power of virtual assistants (whatever technology they are built on) has always been in their ability to integrate with other systems to complete tasks. If an LLM is incorporated, it must be done responsibly and with proper safeguards in place – especially if it has access to backend APIs or personal data.
Before giving an LLM access to business-critical systems, organisations should consider the risks, establish oversight, and ensure API readiness. In fact, lack of robust API infrastructure has long been a bottleneck for many virtual assistant projects, and an LLM alone won’t solve that issue. Kane Simms has some great thoughts about why design is as important as ever, and Christoph Börner gave a memorable case study of what can go wrong if you don’t have the right robust testing in place.
Culture and Collaboration Drive Success
Another key insight was the importance of cross-functional collaboration. One department working in isolation to build a bot is unlikely to produce a successful solution. Instead, organisations can find success with:
- Tech teams that work across frontend, backend, and API integrations
- Customer experience and marketing teams that ensure the bot aligns with customer expectations and brand voice
- Data analytics teams that don’t just report numbers, but identify actionable insights to refine the bot’s performance
- User-centred design thinking, to ensure that the focus remains on the end user’s problem, not just business logic or tech constraints.
Mindset, culture, and skillsets all need to align for a virtual assistant to truly add value. This requires a structured approach that prioritises user needs and business outcomes over any one particular technology.
Jasper Klimbie and Jurgen Gravestein from Conversational Design Institute demonstrated how companies can evaluate the maturity level of their team and operation using the CDI Standards Framework.
Final Thoughts
The European Chatbot and Conversational AI Summit 2025 reinforced that the success of Conversational AI isn’t ever just about choosing the right tech; it’s about designing and building the right solution for the right problem.
Technology is an enabler, but it must be guided by clear objectives, strong design, and a commitment to delivering real user value.
As AI continues to evolve, the organisations that thrive will be those that balance
- innovation with strategy,
- new technology with business needs,
- and AI capabilities with user expectations.
I left the summit feeling optimistic but also reminded that responsible AI adoption is more important than just following the latest trends – it’s about how we deliver meaningful, reliable, and valuable experiences to customers.
And on that topic of responsible and transparent AI adoption… yes, I used an LLM to help restructure and organise my many thoughts on the summit!
Take a look at our Events page to find out what other events we’ll be at, or register to join one of our forthcoming webinars.
