How boost.ai Taught Norwegian Banks to Answer Their Own Phones

A conversational AI network over a Nordic skyline - boost.ai was built in Stavanger, Norway, for the industries that cannot afford a wrong answer. Illustration: YesPress Newsroom.

In 2016, a small group of coders in Stavanger, Norway, took on a boring problem. A local bank was drowning in the same questions, asked thousands of times a day. What is my balance. How do I reset my card. Why was I charged twice. The team built software to answer them. That software became boost.ai, and the boring problem turned out to be one of the most valuable in enterprise software.

Nearly a decade later, boost.ai runs more than 600 live AI agents for over 450 organizations and handles upward of 150 million conversations a year. In August 2025, Gartner named it a Leader in its Magic Quadrant for Conversational AI Platforms - the same market where Salesforce, Uniphore and Five9 all compete. Not bad for a company that started by teaching one bank to answer its own phones.

The hardest customers on earth

Most AI companies chase the easy demo. Write a poem, summarize a document, generate a picture of a cat wearing sunglasses. It looks magical and nobody gets hurt if it is a little wrong. boost.ai went the other way. It chose banking, insurance, telecom and government - industries where a wrong answer is not an amusing quirk. It is a compliance incident.

That decision shaped everything. When your customer is Nordea, DNB, Telenor, Vipps, Santander or Metro Bank, you cannot ship a chatbot that confidently invents a mortgage rate. The bar is not "sounds plausible." The bar is "would survive an audit." So the product had to be built around trust from the first line of code, not bolted on later when the lawyers noticed.

The founders - Lars Ropeid Selsas, alongside Hadle Ropeid Selsas and Henry Vaage Iversen - were Norwegian, and that matters more than it sounds. The Nordics run on high trust and high regulation. Customers expect institutions to be careful, and regulators make sure they are. It is a punishing place to sell software, and a perfect place to prove that AI can be careful too. Win there, and the rest of the world looks easy.

Why "I don't know" is a feature

The counterintuitive thing about boost.ai is what it does not do. A general-purpose chatbot will always give you an answer, even when it has no idea. That is fine for trivia and disastrous for a bank. boost.ai's virtual agents are built to recognize the edge of their own knowledge, decline to guess, and hand the conversation to a human with the full context attached. The customer never restarts from zero.

That restraint is the product. In regulated industries, a system that knows when to stop is worth more than one that talks the most. It is the difference between automation you can put in front of a regulator and a party trick you keep in the lab.

Hybrid AI, or how to have it both ways

The current AI wave runs on generative models that are fluent, flexible and occasionally fictional. Rule-based systems from the previous generation were reliable but rigid. boost.ai's answer is a hybrid: pair the reach and natural language of generative AI with the predictability of controlled logic, so the agent can understand almost anything a customer types while staying inside guardrails a bank can sign off on.

The people who build these agents are not, for the most part, engineers. boost.ai ships a no-code platform, and it counts more than 4,500 certified AI trainers - the service reps, product managers and domain experts who actually know the answers. That is the quiet lesson most software companies still miss: put the tools in the hands of the people closest to the customer, not two teams away behind a ticket queue.

How a boost.ai agent gets built

That third step used to be the scary one. Shipping an AI agent into a regulated business meant crossing your fingers. In 2024 boost.ai introduced Test Studio, a built-in way to validate an agent's performance before it goes live. In an industry where "test before you ship" is not a nice-to-have but the entire sale, it is exactly the unglamorous feature that closes deals.

The ROI nobody tweets about

Enterprise AI does not go viral. It shows up as a number in a spreadsheet. Telenor, the Nordic telecom giant, built a virtual agent named Telmi on boost.ai and, in its own words, hit its business and ROI goals within the first twelve months. No demo day, no thread. Just a return in a year - which, in enterprise software, beats a viral moment every time.

The economics are simple once you see them. A large bank fields millions of repetitive questions. Each one handled by software instead of a person saves money and, done well, answers faster. The customer wins because they are not on hold. The company wins because its human agents are freed for the hard cases the AI hands up. boost.ai's whole pitch is that both can be true at once.

A founder who stepped back on purpose

In 2021, private equity firm Nordic Capital took a stake and set out to push boost.ai beyond the Nordics into the US and UK. By then the company was doing around 100 million Norwegian kroner in annual recurring revenue. The terms were not disclosed, but the direction was clear: go global.

What happened next is rarer than it should be. Founder Lars Ropeid Selsas handed the CEO seat to Jerry Haywood, a British executive brought in to drive international growth, so that Selsas could go back to the thing he cared about most - the product. Plenty of founders cling to the corner office long past the point of usefulness. Stepping aside to do the work you are best at is a quieter kind of confidence, and a healthier one.

The expansion has teeth. Over the past year boost.ai has picked up customers and partners in the US and UK including Jack Henry, SwitchThink and Sage. The Stavanger startup that once served a single local bank now sells into the Fortune 1000 - and lands on the same Gartner Magic Quadrant as Salesforce and Five9.

The steal for founders

If you are building anything, here is the transferable idea. Pick the customer who will punish you hardest for being wrong. Build for their standards, not the standards of a demo. The discipline that survives that buyer becomes a moat everywhere else, because everyone downstream of them has an easier problem. boost.ai chose banks. You can choose your own version of banks.

The unsexy version of the AI boom

There are two AI stories being told right now. One is loud, generative and full of wonder - the machine that writes, draws and improvises. The other is quiet and operational - making sure a real person gets the right balance on their mortgage at two in the morning without waiting forty minutes for a human. boost.ai chose the second story. It is less fun at parties and considerably more useful.

Ten years in, the company has kept a single obsession: automated conversations that regulated enterprises can actually trust. It stayed rooted in Stavanger, a city better known for oil than algorithms, and grew a team of more than 100 people from 18 countries. It never had the flashiest demo. It just had the customers who could least afford to be embarrassed - and it did not embarrass them.

That is the whole company in a sentence. In a market obsessed with what AI can say, boost.ai got quietly rich on teaching it when to stop.

Frequently asked