Why Most AI Investments Fail — Architecture, Not Tools

Only 14% of CFOs report measurable ROI from their AI investments — yet 66% of business leaders expect significant AI impact within two years. Why is nearly everyone betting on AI while so few are seeing it work?

In this debut episode of What Comes Next, former Microsoft Data & AI executive Arunansu (Arun) Pattanayak draws on 20+ years of building enterprise data and AI systems for organizations including EY, KPMG, Deloitte, Citibank, JPMorgan Chase, and Credit Suisse to answer that question — and the answer isn't "move faster."

You'll learn:

• The three assumptions that quietly kill enterprise AI ROI — including why deploying AI is the easy part and building the data foundation is the hard part
• Why AI is a business architecture project, not a technology project — and what happens when it's handed entirely to IT
• Why AI alone creates no competitive advantage: AI is the engine, data is the fuel
• Intelligence Architecture: the deliberate decisions about how data is collected, governed, connected, and activated before a single model is deployed
• The Data Foundation Test: three questions every leader should ask before making any significant AI investment
• Why agentic AI raises the governance bar — and how scaling AI without governance scales risk, not intelligence
• One action to take this week to assess your organization's real AI readiness

Whether you're a CEO, CIO, CDO, or founder planning your AI strategy, this episode gives you a working edge in the language of strategy, not speculation.

Next episode: how to turn your organization's data from a cost center into a competitive product.

Transcript

SPEAKER_00 0:00

Most conversations about AI focus on the tools, what model to use, which vendor to choose, what feature to launch next. But the organizations I work with that are genuinely winning with AI, they barely talk about the tools at all. They talk about architecture. They talk about governance. They talk about the foundation underneath everything else. And that difference between changing tools and building architecture is what differentiates between AI that delivers and AI that disappoints. Welcome to What Comes Next with Arun. I am Arnand Sipatanaik. Most people call me a rune. I am a former data and AI executive from Microsoft, and I have spent more than 20 years building and deploying enterprise data and AI systems for some of the most complex organizations in the world, including KPMG, EY, Deloitte, Booz Allen Hamilton, Maryland, Citibank, Jeffrey Morgan Chase, Credit Suisse, Macmillan, and so on. This so exists because I believe the most important conversations about AI aren't happening in the right rooms. They are either too technical to be useful to leaders or too generic to be actionable for builders. We are going to fix that. Every episode is built to give you a real working edge in the language of strategy, not speculation. Let's get into it. Let me start with a number. According to a recent study, only 14% of CFOs report measurable ROI from their AI investments. 14%. And yet, in that same research, 66% of the business leaders said they expect significant AI impact within two years. So here is the situation. Nearly everyone is betting on AI, and yet the vast majority are not seeing it work. The question I want to answer today is why? Because I don't think the answer is what most people assume. Conventional wisdom says the problem is adoption speed. Move faster, implement more, get more people using the tools. I have heard that advice given in boardrooms and at conferences across the country, and respectfully, I think it's wrong. In fact, I think moving faster with a wrong foundation is precisely what created the ROI crisis we are looking at right now. There are three assumptions I see organizations make about AI that almost always lead to disappointing results. First, they assume deploying AI is the hard part. It's not. Deploying AI is easy. There are more AI tools available today than any organization could ever implement. Most vendors offer easy point and click or template-based solutions for deploying AI. The hard part, the genuinely difficult, unglamorous, unsexy hard part is building the data foundation that gives those tools something real to work with. I'll say it plainly, an AI model is only as good as the data it sits on. And most enterprise data estates are not ready. I talk to many companies who tell me they need an AI consultant or a vendor to help them deploy AI in their organization. What ends up happening is that the vendor sells them their AI subscription as an add-on instead of integrating AI to their business processes. Today, every developer wants to call themselves AI developer. Every executive wants to have AI deployment on their resume because that is what has been perceived as high value in the market. But these quick fixes are the reason why organizations have challenges seeing meaningful business advantage out of using AI. Second assumption, AI is a technology project. It's not, it's a business architecture project that happens to use technology. When organizations hand AI initiatives entirely to IT with no seat at the table for strategy, governance or business operations, they almost always end up with technically impressive solutions that don't connect to any outcome anyone actually cares about. I talk to companies where IT builds an app, AI app, using the latest and greatest available today, only to find out that the app does not meet compliance or cybersecurity requirement of the organization. This can be avoided if business is involved in defining the requirement for AI implementation. The third and this one might be the costliest. AI creates competitive advantage on its own. It doesn't. Data creates competitive advantage. AI is the engine, data is the fuel. If you don't own proprietary, well governed, strategically structured data, you don't have an AI advantage. What you have is an expensive subscription to someone else's advantage. I talk to founders who think because they have AI, they have an unfair advantage. Well, guess what? Everyone else also has AI. It's not having AI, but having the right strategy for implementing AI into your business workflow that will give you competitive advantage. So, what does it actually look like to build AI the right way? I want to introduce a concept I work with constantly. What I call intelligence architecture. And before you tune out, because that sounds like a buzzword, let me tell you what it isn't. It's not a framework someone invented in a consulting white paper. It's not a technology stack. It's not a vendor solution. Intelligence architecture is the set of deliberate decisions an organization makes about how data is collected, governed, connected, and activated before a single AI model is deployed. Think about building a house. You don't start by picking out the furniture. You don't even start by designing the rooms. You start by understanding the land, laying the foundation, setting the load-bearing walls, get those right, and the house can be almost anything you want. Get them wrong, and it doesn't matter how beautiful the furniture is, the structure fails. That is exactly what's happening right now with enterprise AI. Organizations are buying beautiful furniture and putting it in houses with no foundation. So here is something practical. Three questions I call the data foundation test. Before your organization makes any significant AI investment, ask these three questions and answer them honestly. Okay? Question one, do you know what data you own, where it lives, and who is responsible for it? This sounds basic, but in my experience, fewer than half the enterprise organizations can answer it with confidence. And if you don't know what you have, no AI model can help you use it well. Question two, is your data governed? Meaning, are there clear policies for data quality, access, privacy, and compliance? AI governance is becoming one of the most searched topics in the enterprise technology right now. And for good reason. Agentic AI systems, systems that take autonomous action based on data, require a level of governance most organizations simply don't have in place yet. Without it, scaling AI doesn't scale intelligence, it scales risk. Question three, can your data talk to itself? I mean this literally. Can your customer data, your operational data, and your financial data connect in a way that gives you a unified intelligent view? Or are you sitting on desperate data islands that don't talk to each other? Most organizations are. And until you solve that connectivity problem, AI will always give you partial answers because it only has access to partial truth. I have had the privilege of working with some of the world's most sophisticated data and AI organizations through my work at Microsoft. And I can tell you what the leaders have in common. It's not that they have the best AI tools, it's not that they have the biggest budget. What they have in common is this. They treated data foundation as a strategic asset years before anyone else thought it mattered. They invested in governance, in architecture, in the unglamorous infrastructure that makes everything else possible. And now, when AI has become a top-level priority for every organization on earth, they are the ones able to move fast because they build the foundation first. The organizations scrambling right now are the ones that assume they could skip data architecture and go straight to the intelligence layer. They can't. No one can. The foundation always comes first. And the good news is it's not too late to build it. But it requires a leadership decision to stop optimizing for speed and start optimizing for durability. That's exactly what this show is going to help you do. Before the next episode, there is one action I want you to take. Pull together whoever owns data strategy, technology, and operations in your organization and ask them the three data foundation test questions I talked about. And ask them together, not in an email, not but in a room, because the conversation that emerges from that exercise will tell you more about your real AI readiness than any vendor assessment ever will. I promise you that. Next episode, we go deeper into the second layer of intelligence, architecture framework. How to turn your organization's data from a cost center into a competitive product. That conversation will change how you think about every data asset in your organization. I am around, and this is what comes next. And what comes next is always built on what you build right now.