With a tech stack that currently performs the work equivalent of 4,000 full-time employees, we are proving that the secret to AI leadership isn't just about using the latest models; it’s about owning the specialised engine behind them.

In the competitive landscape of SaaS, many companies are currently scrambling to integrate generic AI layers into their products. However, our winning recipe is fundamentally different. Inside the Visma engine, our secret sauce for automation isn't something we buy off the shelf; it is engineered at our core by Visma Machine Learning Assets (VMLA).

A decade of AI-native innovation

Our leadership in AI is no overnight success; it is the result of a decade of focused development. What began ten years ago as a modest assistant has evolved into a cornerstone of our AI-Native Products initiative, a technical powerhouse that develops and maintains industry-leading APIs for our entire Group. By integrating VMLA directly into our broader Product Development ecosystem, we ensure that our machine learning assets are not only world-class but also fully compliant with our rigorous security and frameworks. We haven't just joined the AI race; we’ve spent ten years building the track.

Building the moat: Why our specialised AI wins

This history is what allows us to build a technological moat that generic models cannot cross. We achieve this through domain-specific precision and our decade of proprietary data.

"We have successfully integrated foundational models like Gemini Pro into our stack, but LLMs alone aren't enough for business-critical data," says Claus Dahl, Director of VMLA.

"By blending our ten years of specialised AI expertise with the brainpower of modern LLMs, we outperform generic models by over 30% in extraction accuracy. This means we hit the 100% correct mark for specific data fields far more consistently than any generic model could."

This technical superiority has led to the rollout of Smartscan ULTRA and VERIFIED, a consistency-checking layer that ensures the data delivered to our systems is 100% reliable. While we automate 93% of the overall document volume, these features ensure that for every document processed, the accuracy of the extracted data meets the rigorous standards required for financial applications.

Trust as a competitive advantage

In an era where data usage is under intense scrutiny, we view compliance not as a hurdle, but as a fundamental part of our product quality. Using customer data to train AI requires a level of trust and transparency that is central to how we operate.

"Compliance and data security are the foundation of everything we build. We operate under our VCDM process (Visma Cloud Delivery Model), which ensures that all regulatory requirements, including GDPR and the upcoming EU AI Act, are integrated into our solutions from day one. Our customers need to know that their data is handled with the highest integrity."

By being transparent about how data is used to improve our services within the framework of our data processing agreements, we enable our companies to scale AI solutions securely across different countries and highly regulated industries.

The advantage of scale: Innovation at record speed

For a company within the Visma family, our winning recipe lies in the ability to tap into a central engine that is both powerful and incredibly agile. One of our most impressive milestones in 2025 was reducing the deployment time for new AI features to just one day.

"The speed at which we can move is a massive differentiator," says Claus Dahl. "In one instance, we received a request to extract specific hotel check-out dates from invoices. Because of our configuration-managed development, we were able to build, test, and deploy this as a full platform feature across our infrastructure in a single day. That is the power of having a dedicated, internal ML powerhouse."

This speed, combined with an NPS of +69 for our developer onboarding, means that our products can innovate at a pace that standalone competitors simply cannot match.

From accounting to autonomous agents

The impact of VMLA is already expanding far beyond traditional bookkeeping. From instant ID verification for customer onboarding to document management and ESG reporting. 

As we move toward 2026, our focus is on Reinforcement Learning and Reasoning Capability. Our objective is to move from software that merely "suggests" to autonomous agents that can execute complex business processes with total autonomy and reliability. Claus Dahl concludes: 

"We are moving from systems that store data to systems that act on it. By providing the specialised machine learning assets that power these agents, we ensure our companies deliver a level of proactive value that a standalone startup simply cannot replicate. We aren't just riding the AI wave, we have built the machine that drives it."