It's not about what AI can do. It's about what you can rely on AI to do.
In 2016, a Norwegian bank had a problem that sounds almost quaint now. Too many people were typing questions into its chat window, and not enough humans were available to answer them. Three developers in Sandnes - a coastal town of roughly 80,000 people on Norway's southwest edge, better known for oil services than software - took the job. They incorporated a company that May, called it Boost AI, and started building.
Ten years later, that job has become a platform running more than 600 live AI agents across roughly 450 organizations, handling over 150 million automated conversations a year. The customer list reads like the phone book of European regulated industry: DNB, Nordea, Telenor, Tryg, Vipps, Metro Bank, Santander, Vodafone, The AA. The company was named a Leader in the Gartner Magic Quadrant for Conversational AI Platforms three years running.
The strange part is that boost.ai got there by refusing to do the thing everyone else in AI was doing, which was promising more.
Go to boost.ai's homepage and the headline says: "The conversational AI platform regulated industries trust." Directly beneath it sits the line that actually explains the business - it's not about what AI can do, it's about what you can rely on AI to do.
That is a defensive sentence. In a market where every competing landing page is a list of new capabilities, boost.ai leads with a constraint. It works because of who is reading it. A compliance officer at a retail bank does not lie awake wondering whether an AI can summarise a mortgage document. She lies awake wondering what happens the one time in ten thousand that it summarises it wrong, on a recorded line, to a customer who then makes a decision based on it.
boost.ai's entire product is an answer to that second question.
Enterprises are no longer worried about whether or not they should implement AI into customer service, but rather worried about how to do it safely and at scale.
Jerry Haywood, CEO - March 2026
Strip the marketing away and boost.ai sells a no-code platform where a non-engineer can build, deploy, test and govern an AI agent - then watch what it does in production. The agent lives in two places: chat, where the company started, and voice, which it has been pushing hard since a substantial upgrade in September 2025 and the launch of Adaptive Voice in March 2026.
Around that sit the pieces that make it enterprise software rather than a chatbot builder. Pre-built industry modules ship with use cases, knowledge sources and compliance guardrails already assembled for financial services, insurance, telecom, hospitality and the public sector - which is why a customer can go live in weeks rather than quarters. An integration library lets agents do things: execute a transaction, update a record, finish a workflow, not just describe how the customer might do it themselves. And a layer of agentic capability lets those agents plan, decide and hand work between each other across systems.
The hybrid choice is the most consequential technical decision the company has made, and it has aged well. Pure rules-based bots are reliable and brittle - they break the moment a customer phrases something sideways. Pure large-language-model agents are fluent and occasionally, confidently wrong. boost.ai's position is that the interesting question was never which engine is better. It was which questions deserve which engine, and who decides.
Founder-led companies often talk about choosing customers who stretch them. boost.ai did it by geography and accident: the Nordic banking sector was simply what was nearby. It turned out to be the most demanding possible training ground. Nordic banks digitised early, run enormous chat volumes, and operate under supervision that does not accept "the model said so" as an explanation.
The results are unusually specific for this industry, which mostly trades in vague percentages. DNB's agent, named Aino, handles between 50 and 60 percent of the bank's chat traffic. Nordea's Nova runs as twelve agents across four markets, covering more than 2,300 topics and roughly 50,000 conversations a month, with in-scope resolution above 90 percent. The AA, the British motoring and insurance group, went live in 30 days with accuracy above 90 percent and containment above 45 percent. Aspire General Insurance automates 53 percent of voice interactions across about 37,000 calls a month. TourRadar reports saving more than 780 agent hours monthly.
Note the phrase that recurs: in-scope. It is the quietest and most important idea in boost.ai's approach. The agents are not asked to know everything. They are given a hard perimeter, made excellent inside it, and taught to hand over cleanly at the edge. Teams that try to automate everything usually end up automating nothing anyone trusts.
Every AI demo works. That is the whole problem with demos. boost.ai's less glamorous differentiator is a set of tools built on the assumption that the agent will eventually misbehave and someone needs to catch it first.
Persona-based testing simulates the kinds of users an agent will actually meet - the confused one, the abusive one, the one trying to talk the bot into saying something it shouldn't - and probes for where guardrails give way. A Voice Testing Studio automates test calls at scale, which matters because voice regressions are otherwise invisible until a customer complains. AI-powered analytics then review live conversations continuously rather than by sampling.
This is unglamorous engineering, and it is the answer to the only question that matters once an agent is live: how would you know if it got worse overnight?
Trust is the primary currency of the enterprise AI market. Achieving a clean SOC 2 Type II report is a clear signal.
Jerry Haywood, CEO - April 2026, on the company's zero-exception audit
The company added SOC 2 Type II in April 2026, covering a September-to-December 2025 assessment period, with zero exceptions. It already held ISO 27001 and ISO 27701. These are the least interesting sentences in this article and possibly the most commercially important ones. In regulated procurement, certifications are not marketing - they are the gate. A vendor without them does not reach the shortlist, regardless of how good the model is.
boost.ai is B2B enterprise SaaS in its most classical form: annual subscriptions to the platform, scoped by interaction volume and by which modules a customer takes - chat, voice, agentic capability, industry packs. Implementation, integration work and training sit alongside it. A partner channel of more than 90 technology and channel partners, including Genesys, Salesforce and Vonage, extends reach, and managed-service partners such as Natilik now deliver and operate deployments on the company's behalf.
Then there is the number that looks like a support statistic and is really a business model: 4,500 certified AI trainers. Those people mostly do not work for boost.ai. They work for its customers. The company runs certification programmes that turn client-side staff into competent agent builders, which reduces dependency on professional services and, not incidentally, makes switching platforms an organisational trauma rather than a procurement decision.
On the balance sheet, the public record is thinner. Nordic Capital's Evolution fund made boost.ai its first-ever investment in March 2021, in a deal whose terms were never disclosed; existing backers Finstart Nordic and Alliance Venture stayed in alongside the founders and management. At the time, the company stated annual recurring revenue of roughly NOK 100 million. It has not published a current figure. Its Norwegian entity, BOOST AI AS, registered 95 employees in mid-2026, though global headcount across seven offices is not disclosed.
In December 2022, Lars Ropeid Selsås stopped being CEO. Jerry Haywood, previously of LivePerson, took the job. Selsås stayed with the company as founder. His co-founder Henry Vaage Iversen became Chief Commercial Officer. The third co-founder, Hadle Ropeid Selsås, no longer appears on the executive team page.
There was no teardown blog post and no dramatic exit, which is why almost nobody outside Norway noticed. But it is the pivot point of the company's second decade. The skills that get a product from zero to Nordic banking traction are not the skills that get a vendor through Fortune 1000 procurement in Boston and London. Founders who recognise that in themselves are uncommon; ones who then stay are rarer still.
boost.ai competes with Cognigy, now part of NICE, along with Kore.ai, Ada, Yellow.ai, Aisera, Parloa, IBM's watsonx Assistant and the customer-service agents bundled into Salesforce and Microsoft's platforms. Increasingly the sharpest competitor is not a vendor at all - it is an internal team with a frontier model API and a mandate to build it themselves.
That in-house option is exactly what boost.ai is positioned against. Building a chat agent that impresses an executive committee takes a good engineer a fortnight. Building one that survives an audit, keeps a consistent answer across four markets and 2,300 topics, degrades gracefully when it doesn't know, logs everything, and can be maintained by a business analyst three years after its author leaves - that is a different project, and it is the one boost.ai has been iterating on since 2016.
The company's Gartner position has not been static. After three consecutive years as a Leader through 2025, industry press reported it placed as a Challenger in the 2026 Magic Quadrant, with analysts noting its most recent funding round was smaller than those of comparably backed rivals. That is the tension for a capital-disciplined European vendor in a market where American competitors are raising aggressively: being right about reliability does not automatically outspend being loud about capability.
For an organisation evaluating it, boost.ai is most useful in a narrow and profitable band: high-volume, repetitive, rule-bound interactions where consistency matters more than creativity, and where getting it wrong has a cost with a currency symbol in front of it.
| Sector | What the agent typically owns |
|---|---|
| Banking | Balances, card blocks, payment queries, onboarding, authenticated self-service with clean handover to advisers |
| Insurance | Policy questions, first notification of loss, renewals, claim status, secure payments via partner integration |
| Telecom | Billing, plan changes, troubleshooting scripts, SIM and device logistics across chat and phone |
| Public sector | Citizen enquiries, form guidance and eligibility questions where answers must be identical for everyone |
| Internal support | IT and HR service desks, where the same 200 questions arrive every month forever |
Where it is a poor fit is equally clear: open-ended creative work, low-volume bespoke conversations, and any organisation that wants an AI to improvise. boost.ai's guardrails are the product. If you experience them as friction, you are not the customer.
A few details that stick. The agents have names and personalities of their own - Aino at DNB, Nova at Nordea - which means the most recognisable boost.ai products in the Nordics carry someone else's brand entirely. The company maintains an active GitHub organisation publishing open chat-panel SDKs, which is unusual for an enterprise vendor of this type. And it has no working X account; its public voice lives almost entirely on LinkedIn and YouTube, which is either a strategy or an admission, and either way is consistent with everything else about it.
There is a version of this story where the interesting thing about boost.ai is the technology. There isn't one. The interesting thing is that a company in a town best known for oil services looked at the loudest market of the decade and decided the winning position was to be the one you could check.