AI projects stall when data foundations are weak

Four years after the AI boom sparked by ChatGPT, many companies believe they are ready for the “year of AI ROI.” But successful early deployments of chatbots and copilots do not mean an organization is prepared to embed AI across complex business operations.

According to the Chief Product Officer at Nasuni, the biggest obstacle is not model quality, GPU supply, or raw compute. It is data infrastructure. Businesses have long managed to operate with disconnected file environments, inconsistent governance, and information scattered across multiple repositories because human employees could work around those flaws. AI systems cannot.

Traditional file environments were built for a world where data sat in separate locations, teams controlled access independently, and gaps in governance were tolerable. That changes when AI needs reliable, always-on access to information. If data is hard to find, lacks context, or cannot be accessed consistently, its value drops quickly. In practice, enterprises may have plenty of data but still find it unusable for AI.

The article argues that early AI wins are creating overconfidence. Because chatbots and copilots have a relatively low barrier to entry, they can produce visible results quickly. That can lead companies to assume their infrastructure is ready for larger deployments, even when deeper issues around accessibility, governance, and resilience remain unresolved.

That matters most as businesses move toward agentic AI projects. Without stronger foundations, those efforts can run into delays, weak ROI, and failed implementations. The same organizations, the piece notes, are often also struggling with data recovery after cyber incidents.

Recommended reading

Microsoft delays Exchange Online credential cutoff

The prescription is straightforward: treat data as a strategic asset, not just a storage and capacity problem. That means creating more centralized environments with fewer systems, reducing fragmentation, simplifying management, and giving AI consistent access to trusted, well-governed data.

By this measure, real AI readiness is not whether a company has launched a copilot. It is whether its underlying data is accessible, secure, and fit for purpose at scale.