general

What DNB, SR-Bank, and NAV proved about AI agents (and how a 5-person team copies it)

Three Norwegian institutions measured it. Here are the numbers — and what a small team can copy from them.

2026-05-08 · 9 min · Christian Bru

DNB automated more than half of all chat traffic with its customers. NAV handled more than 270,000 citizen inquiries in the space of a few weeks. SR-Bank increased its customer service capacity by 149 percent — without hiring a single new employee. What these three Norwegian players have in common is that they're not talking about AI agents as something they'll introduce at some point in the future. They have them in production already, and the results are measurable.

The question is no longer whether the technology works. The question is what a small team of five can actually extract from the same principles.

DNB and Aino: when Norway's largest bank automated half of its chat

In October 2018, DNB launched Aino, a conversational AI built on the platform of the Norwegian company boost.ai. From day one, all incoming chat traffic on dnb.no was routed through Aino before a human agent potentially took over.

The result stood: within the first six months, Aino handled over 51 percent of all incoming chat traffic fully automatically — without human involvement. That's more than one in two customer conversations that no longer needs an advisor on the other side of the keyboard.

That this happened in under half a year is not a technical detail — it's an implementation validation. DNB didn't take years to reach this result. The platform was designed for rapid onboarding of intents and continuous learning from actual chat traffic, which means the effect comes early and grows over time.

What DNB proved isn't that AI replaces people. What they proved is that a large portion of inquiries are sufficiently standardized that an agent can handle them fully — and that these inquiries are actually answered better by an agent that's available 24/7, responds in seconds, and never has a bad day.

boost.ai — the Norwegian-developed platform DNB used — specializes in conversational AI for Nordic banks and public sector. It combines advanced NLU (Natural Language Understanding) with enterprise security and Nordic language understanding. This is one example of how Nordic expertise in AI doesn't just exist, but is actually leading internationally.

SR-Bank and Banki: nearly two and a half times more capacity without new hires

SpareBank 1 SR-Bank is one of Norway's leading savings banks with a broad customer base in the Southwest. In 2016, they developed — in collaboration with boost.ai — the chatbot Banki.

The numbers after the rollout are explicit: Banki today handles around 23,000 conversations per month, equivalent to the work of around 20 full-time customer service employees. Capacity increased by 149.3 percent — without the bank hiring a single new person.

It's worth pausing at that last sentence.

A 149 percent capacity increase doesn't mean getting a little help during high-demand periods. It means handling almost two and a half times more work — with the existing staff — because Banki takes care of the inquiries that would otherwise fill the human advisors' calendars.

Banki can today recognize more than 1,900 different intents. That means almost all standard questions — account information, fee inquiries, mortgage calculations, opening hours, card blocking — are answered by Banki, while complex cases and relationship-building remain with the humans. According to SR-Bank's own figures, Banki resolves four out of five inquiries without a human advisor needing to be involved.

For a small team, this principle is directly transferable: don't build an agent that does everything, build an agent that handles the standardized majority of inquiries — and makes it possible for the team to concentrate its capacity where human insight actually makes a difference.

NAV and Frida: hundreds of thousands of inquiries in a national crisis

In March 2020, NAV became one of the most pressured public agencies in modern Norwegian history. The corona pandemic triggered a hurricane of citizen inquiries: furloughs, unemployment benefits, sick pay — hundreds of thousands of Norwegians needed answers immediately, and NAV's ordinary capacity was not dimensioned for what hit.

NAV's chatbot Frida, also developed on the boost.ai platform, was ready.

In the space of a few short weeks, Frida answered more than 270,000 inquiries from citizens with questions related to the pandemic arrangements. At peak, Frida handled incoming traffic equivalent to the work of 220 full-time employees — with a resolution rate of 80 percent.

Eight out of ten conversations resolved, without human help, under the highest pressure situation NAV had faced.

The most interesting thing about the NAV case isn't the pandemic scale itself. It's the structural lesson: AI agents scale instantly. It doesn't take months to recruit and train new employees. Capacity is added in hours, not quarters.

For a team of five that wants to grow without proportional growth in headcount, this is the real news. An agent handling FAQ inquiries at 2 AM on a Sunday makes it possible for the five humans on the team to spend their Monday on the work that actually requires them.

Klarna: global validation of the Nordic proof

Norway isn't alone. Klarna — the Swedish fintech company with global presence — launched its AI customer service assistant in February 2024.

The result in the first month of operation was striking: the assistant handled two-thirds of Klarna's total customer service chat, equivalent to the work of 700 full-time agents — in 23 markets and in more than 35 languages.

The Klarna case is important for a reason that goes beyond the volume itself: it shows that this isn't a local Norwegian or Nordic peculiarity. The principle applies across industries, sizes, and geographies. What DNB, SR-Bank, and NAV proved in a Norwegian context is internationally generalizable.

For a small team, this means the knowledge exists — and the methodology doesn't require an enterprise budget to get started.

What all the cases have in common

Looking at DNB, SR-Bank, NAV, and Klarna together, a clear pattern emerges.

They started with the standardized. None of them tried to automate complex advisory work or discretionary decisions from day one. They identified the inquiries that repeat, are well-formulated, and have a clear answer — and they built the agent around exactly those.

They used platforms, not custom AI. DNB, SR-Bank, and NAV all used boost.ai — a Norwegian-developed platform for conversational AI specializing in NLU for Nordic languages. None of them built their own language models. They assembled a solution on top of a mature platform.

They let humans do what humans do best. Advisors at DNB freed up time for complex client dialogues. Case workers at NAV were spared the thousands of standard questions about furlough rules. The principle is simple: the agent solves the simple and repetitive, humans own the complex and relational.

Volume scales without proportional cost growth. A human customer service agent costs roughly the same regardless of whether they handle a low or very high number of inquiries per day. An AI agent has a marginal cost per conversation approaching zero once it's set up.

They defined clear escalation points. None of the cases tried to make the AI agent an endpoint for all inquiries. A well-defined escalation point — where the agent knows a case needs a human and hands over in the right way — is what makes users actually trust the system. Trust isn't built by the agent never failing, but by it being honest about what it can't do.

What a 5-person team can actually do

Let's make this concrete for a small team delivering professional services — advisory, marketing, audit, law, real estate.

Such a team typically has a recognizable pattern of incoming inquiries:

  • "What does it cost?"
  • "What's included in that package?"
  • "Are you available in week X?"
  • "Can I see examples of previous work?"
  • "What's your process?"

None of these require a human expert. They require the right answer, delivered quickly. An AI agent answering these 24/7 does two things at once: the client gets an answer immediately — not the next morning after coffee is made — and the team doesn't have to use its capacity on inquiries where they don't add unique value.

Another dimension involves competitive availability. A small team often competes against larger players with more staffing. But an AI agent that responds instantly at 11 PM eliminates one of the most concrete advantages large players have: faster response times during business hours. For many clients, response time — not size — is the first signal of whether a provider is serious.

The next step is qualification. An agent that doesn't just answer, but asks follow-up questions — "What's the most important challenge you want to solve?", "What's your timeframe?", "What have you tried previously?" — delivers a fully qualified and contextually enriched lead to the team the day the conversation escalates to a human expert.

A third opportunity is automated follow-up: the agent sends a summary of the conversation to the right person on the team, and starts a nurturing sequence for clients who aren't ready to buy yet. A small team can thus work with a far larger portfolio of potential clients than would be possible with exclusively human follow-up.

That's exactly what DNB, SR-Bank, and NAV showed: it's not about replacing the human. It's about using the human where they have the most value.

For a small team, the asymmetric advantage is even clearer. A large player with many employees and an AI agent doesn't necessarily gain more from automation than a team of five with the same agent — because the relative contribution is so much larger for a small team. The agent doesn't just scale volume, it fundamentally changes what the team can accomplish in a week.

How to start

The cases illustrate a process that is directly transferable to a small team:

  1. Map the repetitive inquiries. What are the ten questions the team answers weekly? That's where the agent starts — not with everything.
  2. Choose the right entry point for your size. A team of five doesn't need enterprise contracts. There are platforms and APIs scaled for all sizes, with high quality and predictable costs.
  3. Define the escalation point. The agent handles the simple. The team takes over at complexity, deep questions, or where the client explicitly wants human contact.
  4. Measure the impact on what matters. Not just time saved, but response time on new inquiries, conversion rate from inquiry to engagement, and client feedback.
  5. Iterate on the intent library. An agent can start with a limited intent library and grow significantly over time — because you learn from the inquiries that weren't handled. SR-Bank's Banki reached 1,900 intents through continuous expansion, not through a perfect design from day one.

Crunchtime offers a Discovery Sprint that maps this for your business: which inquiries are automation candidates, which platform fits your size and budget, and what a first agent can actually cost and deliver.

For a team that has never set up an AI agent, the first project is often the hardest — not because the technology is complex, but because it requires defining what the agent should and should not do. A Discovery Sprint compresses that decision-making process to two days, so that implementation can start with a clear mandate.

No commitment. Two days. A concrete roadmap.

Book a Discovery Sprint

Sources

Sources

  1. boost.ai / DNB (Aino)
  2. boost.ai / SpareBank 1 SR-Bank (Banki)
  3. boost.ai / NAV (Frida)
  4. Klarna International

Want to test a pattern like this in your own business?

30 minutes on a call. We sketch how the workflow could look for you — and tell you honestly whether an agent fits.

Book a free call