IT Modernisation: Can Your Information System Support Your AI Projects?

Why does the information system hold AI projects back?

An AI project needs data that is accessible, reliable and up to date. Yet that data often lives in older applications, sometimes built 20 years ago, that are hard to query and poorly documented.

Cognizant’s study “The path to meeting the legacy modernization mandate” (2025, 1,000 executives from Global 2000 companies) highlights this gap. 85% of the executives surveyed fear that their current technology estate will undermine AI integration. Only 17% feel confident that their infrastructure can support AI agents.

The proof of concept shows it is possible. The system shows it holds up. And it is often the existing information system that makes the difference between the two.

What does technical debt mean for AI?

Technical debt is the sum of past technical choices that now make a system harder to evolve: ageing code, unmaintained components, poorly documented dependencies.

It has a direct cost. According to the same study, maintaining existing systems currently absorbs 61% of the IT budgets of the companies surveyed. That money does not go into new projects. The same executives plan to reverse the trend by 2030: they aim for 27% on maintenance and 43% on modernisation and migration.

For AI, it also has an indirect cost: data locked in silos, applications without an application programming interface (API) and response times that cannot support real-time use. An AI is only as good as the data and applications it can access.

How can you tell whether your information system is ready?

Three questions give a first reading. Can your business data be accessed other than through manual exports? Do your key applications expose documented APIs? Do you know who is responsible for the quality of each major data domain?

An application map completes this reading. It lists your applications, their dependencies and their health. Above all, it links each planned AI project to the systems it relies on.

Take an AI agent designed to help a customer service team answer requests. It needs to check customer history, order status and the knowledge base. If one of those sources is only available through a weekly export, the agent will answer with outdated information. The map reveals these blocking points before they cost you a project.

This work follows the approach described in our article “Enterprise AI: How to Go from Business Need to Industrialization?”: start from business needs before choosing tools.

IT modernisation: do you have to rebuild everything?

No. That is often good news: modernisation happens step by step, application by application, following three main options.

Some applications are kept as they are because they do their job. Others are encapsulated: they are wrapped in APIs that make their data accessible without touching the core system. The rest are rebuilt, when their maintenance cost or limits block new uses.

AI projects help set priorities: you first modernise what serves the highest-value use cases. AI itself can also speed up the work, for example by documenting legacy code or making its migration easier.

This step-by-step approach limits risk. Each phase delivers a visible improvement, teams adopt new tools as they go and the budget is spread over time. It also avoids the tunnel effect of large redesign programmes, which tie up teams for years before delivering results.

Skills matter as much as technology. In the Cognizant study, half of the executives cite a lack of talent able to modernise code as an obstacle. Older applications often rely on a few people who know their history: documenting that knowledge is part of the project.

Modernising for AI does not mean redoing everything: it means making what matters accessible.

What role does data play?

Modernising applications is not enough without good-quality data. You need to know which data has value, make it reliable and organise its governance.

In practice, each major data domain needs a named owner, shared quality rules and a common reference repository. Without that foundation, two modernised applications can still show two different versions of the truth for the same customer.

That is the question we asked in “Does All Your Data Really Have Value?”. It becomes critical as soon as an AI agent has to rely on that data to decide or recommend.

Lùkla supports you…

With our Apps & Data offering, we assess with you whether your information system can support your AI projects. We prioritise the applications to modernise and build access to the data your use cases need.

Our Managed Services teams then handle the maintenance and evolution of your applications over time. From strategy to implementation, we co-build an information system that can keep pace with your ambitions.