In the global race to implement artificial intelligence, many organizations are discovering that the success of their AI projects hinges critically on the quality of their underlying data. This fundamental dependency often leads to ambitious initiatives faltering, rarely advancing beyond the experimental proof-of-concept stage. To understand how to transform these experiments into tangible revenue generators, AI News spoke with Martin Frederik, regional leader for the Netherlands, Belgium, and Luxembourg at data cloud giant Snowflake.

Frederik unequivocally states, “There’s no AI strategy without a data strategy.” He elaborates that the effectiveness of AI applications, agents, and models is directly proportional to the data they are built upon. Without a unified, well-governed data infrastructure, even the most advanced AI models are prone to underperforming. The common narrative for many organizations involves a promising proof-of-concept failing to evolve into a profitable business tool. According to Frederik, this often stems from leaders perceiving technology as the ultimate objective. He advises, “AI is not the destination – it’s the vehicle to achieving your business goals.” Projects frequently encounter roadblocks due to misaligned business objectives, departmental silos, or, most commonly, chaotic data management.

While statistics might suggest that 80% of AI projects never reach production, Frederik offers a more optimistic interpretation, viewing this as an integral part of the “maturation process.” For companies that successfully establish the correct data foundations, the rewards are substantial. A recent Snowflake study reveals that 92% of companies are already realizing a return on their AI investments, with every £1 spent generating £1.41 in cost savings and new revenue. The paramount factor, Frederik reiterates, is establishing a “secure, governed and centralised platform” for data from the outset.

However, even with cutting-edge technology, an AI strategy can fail if the organizational culture is not prepared. A significant challenge lies in ensuring that data is accessible to all who require it, not just a select group of data scientists. To scale AI effectively, robust foundations must be built across “people, processes, and technology.” This necessitates dismantling interdepartmental barriers and democratizing access to high-quality data and AI tools. As Frederik explains, “With the right governance, AI becomes a shared resource rather than a siloed tool.” When teams operate from a single source of truth, they can eliminate disputes over data accuracy and collectively make quicker, more intelligent decisions.

The current breakthrough in AI lies in the emergence of AI agents capable of understanding and reasoning across diverse data types, irrespective of their structure or quality. This includes everything from organized spreadsheet rows and columns to the vast amounts of unstructured information found in documents, videos, and emails. Considering that unstructured data typically constitutes 80-90% of a company’s data, this represents a monumental leap forward. New tools now empower staff, regardless of their technical proficiency, to pose complex questions in natural language and receive direct answers from the data. Frederik describes this evolution as a move towards “goal-directed autonomy.” Historically, AI has functioned as a helpful assistant requiring constant direction; one would ask a question and receive an answer, or request code and receive a snippet. The next generation of AI agents, however, are fundamentally different. Users can assign an agent a complex goal, and it will independently devise the necessary steps, which could involve writing code or gathering information from various applications, to deliver a comprehensive solution. This will automate the most time-consuming aspects of a data scientist’s role, such as “tedious data cleaning” and “repetitive model tuning,” thereby freeing up highly skilled individuals to concentrate on strategic initiatives and drive significant business value.