The enterprise world is currently gripped by an AI-driven paradox. On one hand, there is a global, top-down mandate to adopt Generative AI. On the other hand, a recent MIT study confirms what many leaders are quietly admitting: a majority of companies are struggling with AI implementation and failing to see a meaningful return on investment.
This disconnect between ambition and reality has created a landscape of expensive experiments and frustrated expectations. The critical question, then, is why? Why are so many AI projects failing?
In a recent discussion with Satchit Joglekar, Managing Director of Snowflake ASEAN, I sought to diagnose this widespread problem. His analysis was immediate and clear. The failure isn’t in the AI models themselves or even a lack of strategic desire. The failure lies in a collective distraction with “fancy” new tools, leading businesses to “forget” the “basics”.
Diagnosing ‘Pilot Purgatory’
Satchit aptly describes this state of arrested development as “pilot purgatory”. It is the operational limbo where “everyone tried something”—a chatbot here, a content generator there—but none of these pilots ever mature into full-scale, production-level business use cases.
“A lot of these early initiatives were focused so much on the tools and on the tech itself… because it was fancy, it was new. So, a lot of the basics were forgotten.”
The reason for this stagnation, he argues, is a fundamental misunderstanding of what AI requires. We’ve become enamoured with the consumer-grade experience of tools like ChatGPT, which are “fancy, it was new”. “It’s easy to talk to ChatGPT as a consumer,” Satchit notes, but “in the enterprise world… just the tool itself is not going to do the job”.
The primary reason? An enterprise’s data is a complex, fragmented, and often chaotic entity. The sleek, simple interface of a consumer AI tool belies the immense infrastructural complexity required to make it work reliably within a business.
The Foundational Crisis
The central thesis of Satchit’s argument is that these AI failures are, almost universally, data foundation failures. We are attempting to build a futuristic skyscraper on crumbling, uneven ground. “There is so much fragmentation in the data,” Satchit explains. “There are so many silos that still exist”.
This is the reality for most established companies. “Structured data is residing somewhere. Unstructured data is residing in 10 different places”. Customer records, financial reports, logistics data, maintenance documents, and support logs are all scattered across disparate systems. When an AI model is pointed at this fragmented landscape, it lacks the unified context to deliver accurate or trustworthy results.
“Building [the data] foundation is super critical, and the majority of organisations that started early failed to look at that foundational aspect.”
“Without bringing all of that together and building the strong foundation… the Al is going to fail,” Satchit states. “And that’s what has happened for many organisations”. The projects were doomed from the start, not because the AI wasn’t powerful enough, but because the data it was fed was fundamentally broken.
A Strategic Blueprint for an AI-Ready Foundation
The challenge, therefore, is not an AI problem; it’s a data governance and architecture problem. According to Satchit, the technology to solve this is available. What’s missing is the “will” and “the cohesion across teams” to execute a clear plan.
He outlined a practical, three-step strategic approach for any organisation committed to escaping pilot purgatory.
1. Establish a Holistic Data View
The first mandate is to stop viewing data in departmental silos. Leaders must “look at all of our data” as a single, holistic enterprise asset. This includes everything: “whether it is stuck in databases… in some SharePoint… unstructured data… maintenance documents… invoices or customer onboarding documents”. Only by cataloguing this complete picture can a business begin to unify it.
2. Consolidate to a Single Source of Truth
With a complete inventory, the next step is consolidation. The goal is to “bring it into a single source”, eliminating the “multiple copies and versions of the truth” that plague decision-making.
“Too many times, the same data can be seen by different people and processed by different people and differently. So, we got to get rid of those silos, bring all of that into a single source of truth…”
3. Govern, Secure, and Embed Context
With data unified, the final layer is governance. This involves making the data “trusted, governing it, and understanding where this data came from”. Most critically, this is the stage where “enterprise context” is applied. An AI model doesn’t inherently understand an internal product SKU or a company-specific acronym. The data foundation must be structured to provide this context, ensuring the AI’s outputs are relevant and accurate. “These are all the practical steps,” Satchit affirms, “that need to be done [to] truly have a foundation that’s AI ready”.
Affin-Hwang Asset Management: A Malaysian Case Study
This framework is not merely theoretical. Satchit highlighted Affin-Hwang Asset Management, a Malaysian firm, as a prime example of this strategy in action.
In a data-intensive industry, Affin-Hwang resisted the temptation to chase tools. Instead, “what they’ve really done well is to consolidate and build the data foundations right”. They methodically broke down their internal silos and unified their enterprise data.
The result? “They are very well set up,” Satchit notes. “Now… they are ready to really leverage AI in the right way and successfully with the right business outcomes” . They built the launchpad first, ensuring their AI initiatives will lead to production, not purgatory.
The Barrier Has Shifted
For leaders who see this foundational work as a multi-year obstacle, Satchit argues that perception is outdated. This process used to take “months to years”.
“… with the AI-enabled tools that we are able to do some of these conversions from legacy code into Snowflake in a matter of days”
Modern platforms now offer connectors to easily source data from “SharePoint, Slack, Salesforce, Workday… even just Google Drive”. The technology to build a proper data foundation is more accessible and rapid than ever. The barrier is no longer technical. It is strategic.
The ultimate takeaway from our discussion is that the path to successful AI implementation does not begin with the AI. It begins with the unglamorous but “super critical” work of building a single, governed, and context-aware data foundation28. Businesses that continue to ignore this will remain stuck in pilot purgatory, while those that do the hard work first will be the ones who actually see a return on their AI investment.
