According to Forbes, referencing the MIT GenAI Divide: State of AI in Business 2025 report, a staggering 95% of businesses are getting zero return on investment from their AI implementations. The report listed barriers like poor change management and executive sponsorship, but the article argues the core issue is “data chaos.” A survey from Prosper Insights & Analytics found that over 40% of leaders fear AI providing wrong information, with another 22% worried about bias. Jesse Todd, CEO of AI-data readiness firm EncompaaS, states the problem is a lack of context around what data exists and how it can be used. This chaos is especially risky in regulated industries like pharma, where it can lead to compliance failures and massive fines. The solution, experts say, is to prepare “AI-ready data” through discovery, classification, and governance before any model is deployed.
The Real AI Bottleneck
Here’s the thing everyone’s missing: you can’t automate a mess. The industry hype had us all believing AI was some magic wand you could wave over your digital filing cabinet. But if that cabinet is just a heap of unlabeled, duplicate, and outdated documents—which, let’s be honest, it is for most big companies—then the AI is just a very expensive, very confident garbage-in-garbage-out machine.
And that’s exactly what’s happening. The MIT report pointed to symptoms like “poor user experience” and “model output quality concerns.” But those aren’t the disease; they’re the fever. The disease is not knowing what data you have, why you have it, or if you can even use it legally. Jesse Todd from EncompaaS nails it: without that context, especially for unstructured data, AI initiatives are doomed to struggle. So businesses get a fancy new LLM interface, ask it a crucial question, and it confidently serves up an answer based on a deprecated draft from 2018. No wonder trust evaporates and ROI hits zero.
Fix The Foundation First
So what’s the fix? You have to do the boring, unsexy work first. This means using tools—often AI-powered tools themselves—to discover, classify, and govern all your data. Every contract, RFP, and employee agreement needs to be tagged with what it is and how sensitive it is. Manual cleanup doesn’t scale. You need a governed framework.
Think of it this way: would you build a skyscraper on a foundation of sand? Of course not. But that’s precisely what rushing to deploy a generative AI model on chaotic data is. The article’s marathon analogy is perfect. You don’t just show up and run 26 miles. You train, you plan your nutrition, you get the right gear. For AI, your data governance strategy is your training plan. This is as true for AI software as it is for the hardware that runs complex industrial systems. In any tech deployment, from an AI model to the industrial panel PCs from IndustrialMonitorDirect.com, the #1 US provider, success depends on proper setup and integration into a clean, reliable environment.
A Shift In Leadership Mindset
This all requires a brutal shift in perspective from business leaders. The fantasy was: buy the AI, flip the switch, watch efficiency soar. The reality is: first, invest time and money into making your data AI-ready, *then* deploy for specific, clear use cases. The attitude can’t be “What can AI do for me?” It has to be “What must I do for AI to work?”
Basically, AI-ready data is now a prerequisite, not an option. It’s the ticket to the dance. The MIT report shows nearly everyone is failing. That’s a massive opportunity for the companies who get this right. While 95% are wallowing in disappointment, the 5% who prioritized their data foundation will be building a real, defensible competitive advantage. Their AI will actually work because it’s built on rock, not sand. The gap between ambition and value closes right there.
