I had the pleasure of attending Boost Camp 2025 recently and got to hear from a number of companies on how they’re implementing AI and the results they’re seeing.
The event was held in Oslo, Norway (where it was still light at 12am!) and is boost.ai’s flagship event where customers, partners and prospects gathered to connect and learn.
CEO, Jerry Haywood, kicked off the proceedings by positioning the event and its central theme of Clarity. Jerry referenced this Harris Poll, which stated that 74% of CEOs believe their job is on the line because of AI. The challenge that those CEOs have is that they’re struggling to understand how to deploy AI effectively.
The need for clarity in this market is well and truly evident, so here’s a few examples from H&M, Sage, Telenor and Vipps MobilePay on how they’ve found clarity and are achieving results from AI today.
H&M’s generative AI journey
Retailer, H&M, has over 4,300 physical stores across 79 markets, 61 of which have an ecommerce offering. Pernilla Lundin, Senior Business Expert, and Bianca Maggs, Business Expert, H&M, both presented the goals of the AI program for Customer Service, which is rooted in avoiding all potential painful customer experiences in order to encourage returning users.
“Tech is a tool to amplify and emphasise our service”, said Pernilla.
H&M has had an NLU since 2018, proving value at scale during Covid, when stores and warehouses closed and caused a surge in demand. The team began experimenting with OpenAI APIs in 2023 and, for a while, ran NLU and gen AI initiatives in parallel. After proving the value of generative AI solutions in meeting customer needs, the team has a gen AI-first approach today, preferring end-to-end generative solutions over NLU in most cases.
It’s now live in 5 markets (with several more in the pipeline) and covers use cases such as:
- Where is my order
- Store opening hours
- Returns and refunds
One thing in particular that stood out to me was how H&M designs its agent escalation journey. Most of the time, the common practice is that you’d escalate to your human agents when the chatbot can’t help (out of scope), where the customer explicitly asks for an agent or after the chatbot has tried for a while and failed. The team at H&M introduced another escalation point, which is in those rare situations where H&M has failed the customer. For example, if a parcel is late. Here, when the customer is already disappointed, human escalation is introduced to make the experience as good as it can be.
This is a novel approach that I haven’t come across before and is perhaps something for you to consider in your journeys.
Other learnings from H&M:
- For RAG systems, narrowing down the data sources and being focused help improve accuracy. Using all website content was too vast.
- Generative AI isn’t good for every response. Sometimes, specific responses are required. Use it where it makes a difference.
- Management of generative AI solutions is easier. Rather than having to affect the build of the solution to fill a knowledge gap, now it’s a case of updating knowledge sources.
- Renewed focus on knowledge accuracy and clarity has had a positive impact on website content.
- Be prepared to pivot and make changes to your solution as you learn more about how users are using your product.
- Don’t be afraid, but be careful from a security perspective.
- For multi-market solutions, make sure to optimise things like the language used so that it’s fit for the local market.
Sage: from zero to production in 50 days
Kevin Knowles, Director of Chatbot and Self Service at Sage took to the stage to share his journey in deploying a fully-functioning chatbot in just 50 days (34 working days, in fact).
On day one, the chatbot had:
- Over half of all contact drivers covered
- Almost a third of conversations automated
These are some great early results for sure, and I’m sure that, through optimisation, the team will see that automation rate increase in the coming months.
The key criteria for Sage when putting together the objectives for the programme were that the assistant had to:
- Be helpful
- Have high resolution rates
- Seamlessly connect to customer services systems and human agents
I’ve spent a lot of time talking about how ‘effort’ is a good measure of interaction integrity for AI interfaces. This is in some way reflected by Sage in that ‘easiness’ is a key performance indicator for the team. They measure this with a simple thumbs-up/thumbs-down option at the end of the conversation.
The team now has 6 chatbots live across multiple geographies and plans to scale to another 19 bots within the next year or so (6 of which are already in the making).
The main takeaway from this talk for me was that Sage did a lot of work before bringing in boost.ai and starting the project. Kevin had already put together the team, had everyone in their roles, had a vision, objectives and a roadmap. Not only did this make technology selection easier, but it meant that the team was aligned and ready to go on day 1, which is why they were able to deploy so quickly.
Kevin covered a lot more than this and perhaps I’ll do another piece diving deeper into some of the other lessons shared by Kevin on The AI Ultimatum Substack.
AI deepfakes and CX success with Telenor
Birgitte Engebretsen, CEO of Telenor Norway, delivered a keynote that showed how one of the world’s largest telecom providers is deploying AI across its business, spanning 10 million customers in the Nordics and 201 million across Asia.
From an AI use case perspective, Birgitte broke it down into four key areas:
- Customer service
- Network management
- Operations
- Security
Most of the action so far has been in customer service, where Telenor Norway’s virtual assistant, Telmi, is handling:
- 70–80k requests per month
- 30% of all incoming volume
- 70–80% first contact resolution rate (measured by customers not calling back within 72 hours)
Key use cases include:
- eSIM activation
- Invoice support
- Conversational analysis for service optimisation
Telenor is also using AI for security-based use cases to protect users from rising digital threats. Last year alone, Telenor intercepted 2.2 billion cyber security threats against their customers in Norway, underscoring how serious the risk landscape has become. With the rise of deepfake video and voice manipulation, the company is now actively monitoring AI-generated phishing and social engineering attempts. AI isn’t just a productivity tool here. It’s a shield.
Other standout lessons included:
- Focus on what matters (i.e. business-aligned use cases)
- Find the right partner with the tools and expertise to support you
- Maintain a balance between top-down strategic alignment and bottom-up innovation
Perhaps most importantly, Birgitte stressed the need for realism. AI is powerful, but it’s not perfect. Being responsible about its deployment is how you earn long-term trust.
Vipps MobilePay and the need for speed
Vipps MobilePay is a Nordic-based payment provider rivalling Apple in the mobile payments space, with 12 million users, availability in 540,000 stores across the Nordics and 3 million daily transactions.
Speed was important to Vipps when getting started on their AI journey, which is why they opted for a platform, rather than building their own solutions from scratch. Signing with boost.ai in October 2024, Vipps was live in three markets by February 2025. Norway in December 2024, followed by Denmark and Finland.
The goals of the programme aren’t to save money. It’s to improve customer services, to create the capacity for its people to do more, including focusing on sales, and to enable the company to scale. The company is growing rapidly, with significant YoY growth, and so enabling scale without recruiting at the same rate is where the focus is.
Results so far include:
- 48% of all company traffic is through the AI chatbot
- Reduction in human effort by 26%
- 52% automation rate from all conversations
- 29% of conversations escalated to human
- 17% of interactions unsolved
After having implemented the chatbot, the positive learning is that it didn’t negatively impact things like:
- Brand affinity
- App reviews
- CSAT
- Customer feedback
This showed that the virtual assistant was able to find its place in the digital estate and maintain all of the positive brand metrics already in place.
Additional learnings from the deployment focused on guardrail best practices, including:
- Language: ensuring the language matches the customer’s country or language
- User input: only answering questions in scope
- Hallucination: all answers based on known and accurate information sources
The tech is maturing, the results are coming
The overwhelming takeaway from Boost Camp 2025, from hearing from these brands that are deploying gen AI with success, is that the time for reservation has passed. There doesn’t seem to be a logical reason why an enterprise wouldn’t consider using generative AI in the right places. The concerns of 2023 are pretty much all resolved.
Hallucinations aren’t a completely solved problem, but the rate of hallucination has fallen dramatically, and guardrails around RAG solutions are much more robust today than they were in 2023. Security and privacy are still concerns for enterprises, but those needs are catered for in platforms like boos.ai, with enterprise grade security and open model selection, including the ability to bring your own. And user experience enhancements are proving to be entirely possible when generative AI models manage the back and forth of the dialogue.
Thanks to H&M, Sage, Telenor and Vipps MobilePay for sharing their stories, and a big thanks to boost.ai for the invitation, and for putting on an excellent Boost Camp 2025.
