What does practical AI change in a business?
Six publicly documented cases, from a 30-person workshop in Jutland to Nordic enterprises. The figures are company-reported and linked under each case.
The cases are evidence to discuss - not ready-made packages. The right solution depends on your work, data and responsibilities.
Six Nordic AI cases
Open a case for its challenge, solution, company-reported results, change-leadership lesson and sources.
DenmarkNovo NordiskNovoScribe drafts 300-page clinical study reports in around 10 minutes - work that previously took a team of around 50 writers 12 to 15 weeks.Read the full case
Challenge
Clinical study reports, often up to 300 pages, took a multi-month cycle of writing, review and approval. Staff writers averaged 2.3 reports a year, while every day of delayed market entry can cost up to $15 million in potential revenue.
Solution
Novo Nordisk built NovoScribe on Claude models through Amazon Bedrock. It combines retrieval-augmented generation with text approved by domain experts, after earlier model attempts produced too many errors to be worth fixing.
Company-reported results
- Production time fell from more than 12 weeks to around 10 minutes, with 90% less writing time.
- The report-writing team went from around 50 people to three. The remaining writers were redeployed and none were fired.
- Review cycles fell by 50% as quality improved, and NovoScribe assists around 30% of reports with a target of 90%.
- Annual AI spending is less than one writer's salary. The platform also expanded to device protocols, using 95% fewer resources, and patient materials.
Change-leadership lesson
Novo Nordisk redeployed writers rather than firing them, and the constraint moved from production to review and approval. When a 12-week task takes 10 minutes, the organisation, quality process and role definitions also need redesign. That is an organisational project, not an IT rollout.
Sources
Anthropic case study MongoDB case study Excitech via The Information and DR's Prompt
DenmarkVald. BirnA traditional iron foundry trained a model on three years of pricing history. Quotes that took up to 1.5 hours now take minutes and match the manual calculation at least 90% of the time.Read the full case
Challenge
Vald. Birn in Holstebro, with around 750 employees worldwide, casts iron components for trucks and machinery. Each incoming order required a manual price calculation that could take up to 1.5 hours.
Solution
A machine-learning algorithm trained on three years of past calculations, raw-material prices and production costs creates a draft estimate for new enquiries. People make the final decision.
Company-reported results
- Price estimates fell from up to 1.5 hours to a few minutes.
- The model has at least a 90% hit rate: its price usually matches the team's manual calculation.
- Customers get a useful draft price quickly, while specialists run full calculations only for serious orders.
Change-leadership lesson
Traditional industry does not need an AI department to benefit. It needs to identify a repetitive, data-rich decision that consumes expert time. Here, the machine drafts and the person remains responsible for the final price.
Sources
DenmarkTorben Nielsen VærktøjsslibningA 30-person tool-grinding company built an AI chatbot on its product knowledge and cut onboarding time for new sales staff by around a quarter.Read the full case
Challenge
In a small specialist company, product knowledge sits with a few experienced people. New sales and service staff needed time to answer technical customer questions, and questions escalated to specialists interrupted their work.
Solution
The company built an AI chatbot and AI-supported internal system that gives sales and service staff answers to complex customer questions from the company's own knowledge.
Company-reported results
- Onboarding time for new sales staff fell by around 25%.
- Customer-facing staff can answer complicated technical questions in seconds instead of escalating them.
- DI featured the company in its AI for Alle initiative among Danish SMEs reporting major productivity improvements.
Change-leadership lesson
For SMEs, a high-value AI use case is often internal: capture specialist knowledge so it is no longer a bottleneck. DI's conclusion from more than 40 SME interviews was that willingness and structured knowledge, rather than technology, separate leaders from the rest.
Sources
DenmarkLunarLunar launched Europe's first voice-native AI assistant in banking: 24/7 phone support with no queue, expected to handle around 75% of customer calls over time.Read the full case
Challenge
Serving more than 950,000 customers across Denmark, Sweden and Norway with a lean workforce of around 450, Lunar faced growing support volume without wanting equivalent growth in costs, including for customers calling at 3 AM.
Solution
In October 2024, Lunar launched a genAI-native voice assistant that handles interruptions and natural dialogue without a conventional voice-to-text-to-voice pipeline. Staff also use LunarGPT, trained on internal material, and engineering uses AI coding assistance.
Company-reported results
- Lunar describes itself as the first European bank with genAI-native voice technology in customer service.
- The assistant is available 24/7/365 with no queue time and is expected to handle around 75% of calls over time.
- Lunar states there were no job cuts. It slowed hiring instead, and customers can always choose a person.
Change-leadership lesson
Lunar paired a customer-facing agent with internal enablement through LunarGPT for all staff. Adoption inside the company came before automation outside it, while keeping a human option visible as a trust decision.
Sources
Lunar press release Tech.eu on the launch Tech.eu on internal AI
NorwayreMarkableThe Norwegian scale-up put its AI support agent Mark live in three weeks. It now resolves more than a third of support cases autonomously while matching human customer satisfaction.Read the full case
Challenge
With more than three million paper tablets sold, around $500 million in revenue and sharp seasonal spikes, reMarkable's support and IT teams were stretched. Outdated knowledge articles also made answers difficult to find for people and machines.
Solution
reMarkable built two Agentforce agents: Mark for customer support, grounded in real-time customer and product data, and Saga, an internal IT helpdesk agent in Slack for password resets and routine tickets.
Company-reported results
- The first agent went live in three weeks and handled thousands of conversations through the company's busiest Black Friday.
- It resolves 35 to 37% of inbound support cases autonomously, equivalent to roughly 20% of the 115-person support team.
- Customer satisfaction is on par with human agents, with more than 25,000 conversations handled.
- Preparing the agent led to a cleanup of outdated knowledge articles, improving answers for people too.
Change-leadership lesson
reMarkable treated the agent as a new colleague: training, feedback loops and better resources. The essential but unglamorous prerequisite was knowledge hygiene, which the agent made impossible to ignore.
Sources
Salesforce customer story Salesforce newsroom Agentforce metrics page
SwedenIKEA / Ingka GroupIKEA's Billie chatbot took over 47% of customer-service enquiries. Ingka retrained 8,500 call-centre workers as remote interior design advisers instead of laying them off.Read the full case
Challenge
Ingka Group, which operates most IKEA stores, handled millions of enquiries each year. Many were necessary but transactional contacts about orders, deliveries and returns.
Solution
The Billie chatbot, launched in 2021, absorbed routine volume. Ingka analysed the remaining contacts and found demand for design advice that agents had not had capacity to serve, then launched a structured reskilling programme.
Company-reported results
- Billie handled around 47% of enquiries, or 3.2 million conversations from 2021 to 2023, later rising towards 57%.
- Ingka reports around €13 million in operating savings from automation.
- Around 8,500 call-centre workers were retrained as remote interior design advisers, with zero layoffs.
- The remote design channel generated around €1.3 billion in FY2022, around 3.3% of sales, with a stated 10% target by 2028. This is the full channel's revenue, not revenue solely attributable to the chatbot.
Change-leadership lesson
The chatbot's larger contribution was revealing what customers wanted from people. Ingka asked what these people could do that a bot could not, rather than how many people it could remove.
Sources
What the cases have in common
- Size does not gate ROI. A 30-person tool grinder and a 750-person foundry show the same kind of win as Novo Nordisk: a narrow workflow with measurable time saved.
- Nobody leads with layoffs. Novo Nordisk redeployed writers, Lunar slowed hiring and IKEA reskilled 8,500 people. In these examples, AI is capacity rather than headcount reduction.
- Knowledge hygiene comes first. reMarkable rewrote its knowledge base and Torben Nielsen structured specialist knowledge. An agent is only as good as what the organisation has written down.
- Speed is real. reMarkable put an agent live in three weeks, while Vald. Birn cut quoting from up to 1.5 hours to minutes. Organisational readiness is the longer pole.
- More local evidence is available. DI's AI case archive documents more Danish SME examples and continues to grow.