In the first article of this series, I explained why vendor-provided AI features often fail in production. There are three main reasons: they do not understand your business operations deeply, they struggle to integrate with your current systems, and they give confident but wrong answers when unsure. However, the root of the problem extends beyond the limitations of individual features. It lies in how the whole AI estate is put together: piece by piece, department by department, without a clear enterprise strategy.
The Fragmented Construction of the AI Estate
Today, many organisations adopt AI in a decentralised manner. Each function procures its own AI solution in isolation. Marketing teams deploy AI within content platforms to refine campaigns and personalise messaging. Commercial teams integrate AI into CRM systems to enhance customer insights and determine the next best action. Medical departments adopt AI tools within platforms like Veeva to streamline clinical data management and ensure compliance. Meanwhile, analytics teams leverage AI capabilities in their business intelligence tools to uncover patterns and inform decision-making.
Each department evaluates AI features based on its specific needs, negotiates contracts independently, and runs pilot programmes in silos. While this approach may seem logical at the departmental level, it is rarely optimal for the enterprise as a whole. Without a unifying strategy, organisations risk wasting resources, creating conflicting systems, and missing opportunities for collaboration.
No organisation sets out to build a disjointed AI estate. Instead, it evolves incrementally. Every SaaS vendor in your technology stack now includes an AI feature, and each department, operating within its own renewal cycle and budget constraints, adopts these features in isolation. The vendor selling SharePoint also offers Copilot, and the vendor providing your CRM system promotes its proprietary AI assistant. The path of least resistance, both technically and politically, inevitably leads to the proliferation of AI tools: one AI per platform and one platform per function.
The result is not a unified, intelligent enterprise but a patchwork of isolated AI systems. Each tool may excel within its narrow domain, but it remains blind to the broader organisational context.
The True Cost of Eleven AI Projects
Organisations allocate substantial budgets across multiple AI projects. Each initiative is justified by its own business case, complete with compelling projections and clear benefits. However, when these projects are viewed collectively, the return on investment rarely materialises as expected.
In practice, these tools often operate in isolation, unable to collaborate or share insights across departmental boundaries. Even a seemingly straightforward cross-functional question, such as “What is the impact of a delay in our supply chain on our Q3 sales forecast?”, still requires manual intervention. Teams must gather data from disparate systems, reconcile inconsistencies, and piece together an answer.
Here is where the maths do not add up: Eleven AI projects do not yield eleven times the value of a single, integrated project. Instead, they produce overlapping insights, each confined to its own silo. The greatest expense of each AI deployment is not the licensing fee or the implementation cost but the opportunity cost: the insights, efficiencies, and innovations that remain trapped within isolated tools, inaccessible to the broader organisation.
While executives may perceive this proliferation of AI tools as progress, operational teams experience it as fragmentation with added complexity. Rather than simplifying cross-functional work, it adds new challenges, reduces transparency and increases inefficiencies.
The Accelerating Problem of Silos
Each AI deployment not only fails to break down existing silos but often reinforces them. Consider how this plays out in practice. The marketing AI generates content and optimises campaigns using marketing data and historical trends. It becomes increasingly adept at understanding marketing insights but remains disconnected from other areas. Simultaneously, the commercial AI develops account strategies based on commercial data and sales history, deepening the divide between marketing and commercial knowledge. The medical AI focuses on clinical data and regulatory requirements, while the analytics AI identifies trends in performance metrics, and the HR AI concentrates on workforce planning. Each tool improves within its own domain, but the gaps between them widen.
With each passing quarter, the silos grow. Each function’s AI accelerates its separation from the rest, reinforcing the barriers that hinder collaboration and knowledge sharing. Silos were once a slow, almost passive challenge within organisations. With AI embedded within each silo, the problem now compounds at an accelerating rate. Tools designed to simplify operations are, in fact, amplifying fragmentation, making it increasingly difficult to achieve a clear, unified view across the company.
The Solution: A Unified Knowledge Layer
The solution is not another AI vendor. No single AI, no matter how advanced, can resolve the fragmentation caused by the previous eleven. The solution lies in addressing what all eleven lack: a unified knowledge layer, a cross-function, enterprise-wide source of truth that all AI tools can access and contribute to.
Imagine this knowledge layer as the foundation of your AI estate. It serves as a shared repository of critical business information, acting as a map and dictionary of meaning for your business data and the material used by business functions. This can include customer interactions, product data, compliance frameworks, and operational metrics. More than just a database, it acts as a central nervous system for your organisation’s existing systems. This layer enables your AI tools to maintain a coherent and consistent understanding of the business and its collective knowledge.
The implementation can be gradual, with compounding value as each function adopts the knowledge layer. The more functions that integrate, the greater the effective value. Your marketing AI can draw on insights from commercial data, your commercial AI can incorporate medical compliance requirements, and your analytics AI can factor in HR workforce trends. The result is a truly intelligent enterprise, where AI tools work in concert rather than in isolation, multiplying the value of every insight while the process and tool usage remain owned and managed by each department.
This is not about replacing tools, scrapping investments, or buying yet another SaaS promise. It is about building the enterprise AI memory that turns eleven isolated tools into one intelligent organisation.
In Article 3, Forget More Data, You Need a Memory, we will explore this concept in depth. Using plain, accessible language we will explore what a knowledge layer is, how it functions, and why it is the missing piece in most organisations’ AI strategies.
In the meantime, consider this question for your next budget review or strategy session: Of all the AI projects currently running across your organisation, how many can answer a question that requires data or insights from another department?
If the answer is none, or even just a few, then the reality is stark. You do not have eleven AI projects. You have eleven silos equipped with chat interfaces. And an operational drag that compounds with every quarter.
Let’s Discuss
If the challenges described in this article resonate with your organisation, I would welcome the opportunity to discuss them further. At Forge DC, we help enterprises connect what they already have, rather than buy more of what they do not need. Our work focuses on the connections and frameworks that turn existing investments into a coherent AI estate.
Contact us to explore how we can address your specific needs and help you transform your AI estate from a collection of isolated projects into a unified, intelligent system.
Fabio Barboza is CTO and Strategic Partner at Forge DC, where he leads R&D on DataBridgeAI and writes Forge Field Notes. He has spent more than a decade at the intersection of data, technology, and commercial pharma. Based in London.