At Boost Camp 2025, I had the honour and pleasure of hosting a live VUX World podcast from the main stage. I was joined by Ben Maxim, CTO at Michigan State University Federal Credit Union (MSUFCU), Åse Marthinsen, Head of Generative AI at Norway’s DNB Bank, and Nick Mitchell, CRO at boost.ai, to explore how leading financial services organisations are finding success with AI.
DNB’s AI use cases
DNB, Norway’s largest bank, began its AI journey in 2018 by launching Aino, a customer-facing chatbot developed in collaboration with boost.ai. By 2019, they had also deployed Juno, an internal chatbot to support human advisors. These initiatives laid the foundation for integrating conversational AI across customer service and internal operations.
More recently, DNB began incorporating large language models (LLMs) into their chatbot systems starting in 2023, enhancing both the scope and quality of responses.
MSU Federal Credit Union use cases
MSUFCU started its AI journey around the same time, conceptualising its member-facing chatbot, Fran, in 2018 and launching it in 2019. An internal bot for employees followed in 2020. These tools were designed to streamline support, unify knowledge and enhance both customer and employee experiences.
Initially, MSUFCU relied on traditional NLU models trained on banking data, and migrated Fran to the boost.ai platform in 2021. Today, they’re expanding into LLM-based capabilities while maintaining a hybrid approach for better control and scalability.
AI as the interface to the brand
One of the biggest discussion points centred around how, given the maturity of the AI programme at both organisations, the customer-facing AI assistants are becoming the primary interface to the brand.
DNB’s chatbot handles nearly a quarter of all customer service traffic. Ben Maxim, MSUFCU, shared that the credit union now sees Fran, their virtual assistant, as a “gateway into the experience of the organisation”.
Inspired by how users start their internet journey with Google, MSUFCU envisions members beginning their entire digital banking experience by engaging with Fran, whether it’s via chatbot, voice or other future interfaces. Ben emphasised that Fran isn’t just answering questions; she is becoming the digital front door to the credit union’s ecosystem.
Capital One were early proponents of this approach back in 2018 when its assistant, ENO, was available through Amazon Alexa, Google Assistant, through the call centre, email and even as a Google Chrome shortcut. It’s safe to say the tech wasn’t quite ready then, but now, we might be at a place where that single entry point for all services is feasible.
Nick Mitchell (boost.ai) echoed this trend, explaining that conversational assistants are evolving into core brand interfaces. In his view, the first brand to fully embed a conversational AI as its main customer interface will set a major precedent for the future of customer experience.
Left to right: Kane Simms, Founder and CEO at VUX World; Ben Maxim, CTO at Michigan State University Federal Credit Union (MSUFCU); Åse Marthinsen, Head of Generative AI at DNB Bank; and Nick Mitchell, CRO at boost.ai.
Generative AI from end-to-end or hybrid?
To get to a place where you’re able to have conversations with customers about all banking use cases, triage to the best channel of resolution, the best live agent to help or automate entirely, you need to have a centralised ‘brain’. That’s the AI capability that’s distributed across channels. That requires a blend of orchestrated technologies. From data, integrations, knowledge, workflows, channels and much more. Inevitably, to create the best possible solution, you need the best tools for the job.
We’re now starting to see generative AI being viewed as that: a tool. An enabler. Something that can help improve the efficacy of a specific job or task. This is opposed to the perspectives from 2023, which was that everything has to be generative AI or else!
Blending generative AI and NLU
There’s growing consensus around the importance of blending traditional natural language understanding (NLU) with generative AI. While the excitement around LLMs continues to build, and rightly so, none of the speakers viewed generative AI as a standalone solution. Instead, they advocated for a hybrid approach that combines the control and precision of NLU with the creativity and coverage of generative models.
Nick Mitchell, Chief Revenue Officer at boost.ai, highlighted how most clients now seek a balance between the two. While some early RFPs requested ‘pure generative AI’, the reality is that most organisations need tools that allow them to blend approaches.
For MSUFCU, NLU still plays a key role in handling high-stakes financial inquiries with consistent, scripted accuracy. At the same time, they’ve integrated generative capabilities to expand coverage and handle broader, more dynamic customer needs.
The benefits of generative AI are clear: it offers richer interactions, greater flexibility, and faster scalability. But Åse was quick to point out that this power comes with a need for strong governance, data quality and intentional experience design.
Even from a broader industry lens, there’s growing recognition that generative AI excels in some areas, but not all. As I shared during the session, regulated industries in particular require more controlled, auditable systems, which NLU is uniquely well-suited to deliver. There’s a place for both technologies and the brands that win will be the ones that know when to use which.
Results and outcomes
Both MSUFCU and DNB have achieved some notable results from their AI programs that include:
1. Workload automation
- Fran handles the equivalent of 65 full-time employees’ (FTEs) worth of volume monthly, including 24/7 coverage, particularly outside regular contact centre hours. This has allowed the organisation to understand customer needs during off-peak hours (e.g., failed transactions at 2–3 a.m.).
- Aino is handling around 60% of all incoming chat traffic and ~22% of the bank’s total customer service traffic.
2. High success rate
- 79% of conversations initiated with Fran are fully contained with no need for escalation to live agents. While I have a gripe with the ‘containment’ metric, surely a good portion of that user base has had their issue successfully resolved? Ultimately, this translates to more efficient service, reduced costs and better use of human support staff.
- DNB automated 20% of all customer service traffic across the business within the first six months.
3. Improved customer experience and access
- Members benefit from faster, more reliable self-service, while staff gain guided support via the internal assistant.
- Aino CSAT scores are at an all time high at 68%
You can listen to the full discussion on the VUX World podcast. Listen on Apple Podcasts, Spotify and YouTube.
