← Back to news

When IT Legacy becomes a barrier to scaling AI

September 21, 2026
When IT Legacy becomes a barrier to scaling AI

Two Different Technology Journeys

AI-native companies have often built their operating models around platforms, data, and reusable services. Their architectures make it easier to integrate new use cases and rapidly expand their capabilities.

Established companies have typically followed a different path. Their IT landscapes have evolved incrementally, with each application addressing a specific business need. Acquisitions may have brought their own ERP or CRM. Each transformation added new interfaces, data flows, and layers of complexity.

The result is an IT landscape that works  but where the same business reality may be represented in several different ways.

A customer or site may not have a unique identifier. A product may be defined differently from one application to another. Data may have been moved into data lakes or warehouses whose quality, provenance, and underlying logic have become increasingly difficult to control.

The company may therefore have plenty of data, without necessarily having a consistent view of it.

For AI, this distinction is critical. AI models rely on enterprise data to generate answers, insights, and recommendations. They can process huge volumes of information, but they cannot decide on their own which conflicting source represents the actual business reality.

When source data is unreliable, inaccessible, or inconsistent, the ability of AI to create value is fundamentally constrained.

Don’t Start from Scratch

Faced with this situation, the obvious temptation is to rebuild the IT landscape on a clean, modern foundation.

For a large organization, this is rarely realistic or even desirable. It is too costly, too disruptive, and too risky.

The real question is what needs to change to make AI scalable, without trying to modernize the entire IT landscape at once.

Depending on the context, this could mean evolving the data platform, clarifying data domains and ownership, modernizing specific interfaces, or strengthening data access and governance.

But there is no universal answer.

A new data platform will not resolve contradictory data. Defining data domains will not help if ownership remains unclear. Modernizing an interface will have limited impact if the real problem lies in the underlying business model.

The starting point should therefore be the business problem to solve not the technology to deploy.

AI use cases can be particularly useful in exposing these underlying issues. A sales assistant, for example, may quickly reveal inconsistencies between CRM data, contracts, and product information.

These issues are not necessarily failures. They are signals. They help identify where the real bottlenecks are, prioritize investments, and progressively build the capabilities required to scale AI.

From POCs to an Enterprise AI Capability

This is where the real challenge lies.

A successful POC does not demonstrate that an organization is ready to deploy AI at scale. Scaling requires data, services, and capabilities to be reused across multiple applications and teams.

The objective is therefore not to suddenly become “AI-native”, but to progressively make the existing IT landscape AI-ready.

That means aligning technology decisions with business priorities: Which data is truly critical? Who owns it? Which capabilities need to be shared? Which systems need to evolve — and which can remain as they are?

The goal is not to eliminate IT legacy. It is to determine where it has become a constraint, and where it can continue to serve the business effectively.

Conclusion

AI does not create the problems of an IT landscape. It exposes them.

Established companies cannot erase decades of technological history. Nor do they need to.

What they can do is progressively evolve their existing landscape so that data and AI become genuine enterprise capabilities.

The question is not how to replace legacy with a perfect architecture. It is how to evolve the IT landscape at the pace of the business.

That ability to connect strategy, architecture, data, and execution is what will allow organizations to move beyond isolated AI experiments and build sustainable, scalable business value.