Companies No Longer Need Another AI Pilot. They Need an Operational Layer for AI

June 10, 2026

Today, most companies begin their AI journey in a similar way, a few pilot projects, the first chatbot, an internal assistant, automation of individual tasks. Yet based on our experience from the Czech Republic and abroad, this is exactly where the biggest risk emerges. AI starts spreading faster than the organization’s ability to govern it.

“The biggest mistake is not starting too late. The biggest mistake is starting chaotically,” says Petr Zelenka, Chief AI Officer, Adastra Group.

The banking sector, in particular, is already showing what AI adoption will likely look like across other industries as well. Fewer experiments, more governance. Fewer isolated tools, more systematic approaches. And most importantly, AI is no longer just an IT topic.

AI Is Moving from a Tool to an Operational Layer of the Company

Just two years ago, most organizations viewed generative AI as a new type of software tool. Today, that perspective is changing.

According to Petr Zelenka, AI is gradually ceasing to be merely an “assistant” and is becoming an operational layer of the organization, much like cloud or data platforms once evolved into invisible infrastructure. “AI is shifting from being a tool to becoming something through which most work, decision-making, and knowledge sharing will eventually flow,” he explains.

This is precisely why AI adoption is no longer just a technology project. Once AI begins influencing decision-making, customer communication, and internal processes, and works with sensitive data, it stops being a purely technological initiative and becomes a strategic topic for executive leadership. Companies are discovering that the technology itself is paradoxically the easier part.

Why Most AI Projects Run into the Same Problem

Today, companies often start with quick experiments. One team deploys a copilot, another builds its own AI assistant, while others use external models without broader coordination. In the short term, this works. In the long term, however, three major problems emerge:

1. Duplicated work – every team solves the same issues repeatedly, such as data access, security, monitoring, model management, and compliance
2. Shadow AI – employees use AI tools beyond the organization’s control, often involving company or customer data
3. Unmanageable architecture – every new use case adds another technology layer that becomes increasingly difficult to manage and scale

That is why large organizations are starting to build a shared foundation instead of isolated use cases – an AI platform.

NLB: Why the Bank Did Not Start with a Chatbot but with a Platform

When NLB Group, the largest banking group in Slovenia and a leader in retail, corporate, and investment banking across Southeast Europe, approached us to discuss its AI strategy, the organization already had, in its own words, a list of approximately 500 potential use cases. Instead of trying to implement as many of them as possible as quickly as possible, NLB chose a different path. They first created an architecture capable of supporting AI operations securely and sustainably over the long term.

This meant investing in areas that do not deliver immediate business impact at first glance, such as governance, security, observability, access management, integration with core systems, and control over data handling.

“The value of the first use cases was not only in ROI. Each of them created reusable components we could leverage again,” explained Group CIO Dejan Pust.

That is the key difference between an AI experiment and AI transformation.

AI Adoption Today Is Not a Sprint Toward the First Use Case

A large part of the market is still searching for a “killer use case.” Yet enterprise experience shows that more important than the first use case is the ability to govern AI, scale it securely, and deploy it repeatedly.

In regulated industries, this becomes even more critical.

“As a bank, we cannot afford to experiment with a ‘let’s try and see’ approach. Our brand is built on trust,” emphasizes Dejan Pust.

That does not mean moving slowly. NLB managed to build its first functional AI platform and deploy initial use cases within five months. The key was not building everything from scratch, but leveraging the existing cloud ecosystem, choosing the right partners, and proceeding systematically.

AI Reveals Problems That Already Exist Within the Organization

AI itself is usually not the biggest problem. What AI does very quickly is expose poor-quality data, missing integrations, unclear processes, weak governance models, etc.

“During development, we often discovered that we first needed to fix a data pipeline or improve data quality,” says Dejan Pust.

This is an important message for CIOs and executive leadership as well. AI projects are not separate from the organization’s existing IT reality. On the contrary, they expose it very quickly.

Governance Is No Longer a Barrier; It Is a Prerequisite for Scaling

Until recently, governance was often perceived as something that slows innovation down. With AI, the situation is reversing. Without clear rules, organizations quickly lose control over costs, visibility into data, the ability to audit decisions, and regulatory certainty.

This is why NLB restricts work with sensitive data, uses only approved models, keeps data within the EU, and keeps humans in the decision-making loop. “The AI agent only proposes an action. Final decisions and changes are still approved by humans,” explains Dejan Pust.

The Real Value of the Platform Comes Later

From a financial perspective, the first year of an AI platform is often difficult to measure. Organizations invest in foundational capabilities that are not directly visible. The real impact only becomes evident later:

• faster deployment of new use cases
• lower development costs
• reduced duplication
• simpler compliance

“New use cases will be developed approximately ten times faster than the initial implementation,” says Ales Gorisek, AI Architect, NLB Group.

And this is where the true economics of AI platforms begin.

AI Strategy Is No Longer a Question of “If,” but “How”

There are still companies waiting to see “how AI evolves.” Today, this is becoming an increasingly risky strategy; not because of the technology itself, but because of the experience gap. “A year lost by delaying AI adoption cannot later be recovered simply by purchasing technology,” says Petr Zelenka.

Organizations are not only learning how to work with models, create use cases, or select tools. They are learning how to redesign processes, establish governance, work with the evolving role of employees, and integrate AI into the company’s day-to-day operations.

And according to many, this experience will ultimately become one of the main differences between companies that truly leverage AI and those that remain stuck with isolated experiments.

This article was published by CIOtrends.