AI and data integration: an indispensable symbiosis

AI continues to be a defining factor this year, with forecasts pointing to a $250 billion market, with generative AI as the growth engine. In their quest for increased efficiency, better decision-making and higher customer satisfaction, organizations are exploring what AI can do for their processes in full with more than half of companies expecting to increase their AI spending in 2025.

Pim Simons, Integration Domain Lead at Codit

“In large companies with diverse IT systems, integration is the heart of the business,” Francis Defauw, Chief Portfolio & Marketing Officer at Codit, says. “Without an integration layer, the business does not function. Use cases rely on these data flows.”

A concrete example: a supplier of fresh fruits and vegetables optimizes its ordering and delivery processes by accurately predicting demand which helps with efficient personnel planning and less waste. Data integration is crucial here.

Accurate demand forecasting requires data from CRM, inventory and transportation systems. This data sits in separate applications, known as data silos. Integration is necessary when extracting value from this data, to make better forecasts. With data integration, companies operate more efficiently and respond better to market movements and customer needs, thus adding strategic value.

What makes data integration complex?

Data integration goes beyond mere data collection; it is about bringing together data from different systems into a coherent whole, according to a certain reference architecture. “Instead of linking systems directly, we bring all the data together through an integration layer,” Pim explains. “This makes the data flows more manageable and provides the flexibility to add new functionalities quickly.”

Francis Defauw, Chief Portfolio & Marketing Officer at Codit

At many companies, data resides in data silos spread across different applications such as ERP, CRM and WMS. Those silos make it difficult to combine data. “Just that makes it all very complex and again underscores the importance of data integration,” Pim continues. “The integration layer breaks through all those silos and brings the data together in a central environment. That provides a single source of truth, without the company having to pull data from different applications each time.”

Scalability and cost

Scalability is essential for data integration, especially as the data volumes used increase. Those looking to efficiently store massive amounts of data quickly turn to cloud-based solutions, such as Microsoft Azure, which offer almost unlimited scalability. The challenge here is to pay close attention to costs.

“Those costs are usually not so much in the storage of data, but rather in the tools you use to develop and implement AI applications,” Pim explains. In addition, companies often do not know what data they will need in the future. “It is only later that they discover that they are missing certain datasets. Storing raw data right from the beginning lets you build a history that can be valuable later.”

But there are practical challenges to that too. “As more applications enter use, the number of data formats also increases,” Pim says. “But working with common, readable data formats does make it easier to perform more complex tasks with them later on. Data integration is therefore preferably done in consistent formats, such as JSON or XML.”

Don't forget data governance

A strong data integration strategy requires more than technology alone, it needs a combination of governance, advanced tools and automation to ensure scalability and reliability.

“Data governance is essential to ensure data quality and privacy, especially now that AI models are using business-sensitive data more often,” Pim explains. “A tool like Microsoft Azure Purview then aligns nicely with DevOps processes, providing support for data governance within large-scale data integration.”

To realize that vision, the importance of data integration increases even further. “The quality and consistency of data determine whether Agentic AI works well,” Pim concludes. “Errors in the data stream can lead to wrong suggestions or decisions, possibly with unwanted consequences. That kind of autonomous system can't work without a solid integration layer.”