By Tim Long, Global Head of Manufacturing, Snowflake
Universal truths
Some laws are universal. Nothing travels faster than light. Energy can be neither created nor destroyed. And, every business technology has its limits. That third one may not be a law of physics attributable to Newton, but it’s a reality that businesses are facing as we speak.
The hardest part of adopting a new technology is figuring out where the ceiling actually is. And it’s not the first time a lot of us have had to discover it. For example, the internet's ceiling, for decades, wasn't the technology itself. TCP/IP existed in the 1970s. What held it back was how few things were connected to it. A network is only as powerful as its nodes.
More recently, we learned that there’s a hard limit on social media’s capacity as a marketing platform. That limit is the number of hours a day someone spends scrolling on their phone.
And now, we’ve come to another significant technological breakthrough. One with a limit that, in my experience, most leaders fundamentally misunderstand. It’s easy to assume that the limitation of AI is the model. But it’s not. It’s the data.
How do we define “data”?
When most people hear the word data, they picture tidy rows and columns. Spreadsheets. Database tables. Pixel-perfect reports pulled out of a system once a quarter. That definition isn't wrong, but it's the smallest possible version of the word. I think of data as anything observable: drawings, photographs, video, audio, sensor readings, geospatial information. Any observation a machine can capture is something AI can now learn from. The surface area of your data is far larger than the contents of any database.
At this point you might realize what this means. Your business doesn’t lack the required data. In fact, you probably have…
More data than you know what to do with
A typical construction firm is running a dozen specialized platforms at any given moment. One for project management. One for accounting. One for design. One for field operations. One for fleet. Each of those systems is generating data every day, whether or not anyone is using it. Some of them have AI built in already, which sounds like progress until you realize that whatever AI sits inside a system can only see what that system sees. None of these tools have visibility across the rest of your business.
Companies have so much data in so many places, that no single tool can see across enough of the business to make a meaningful decision. If you're starting to wonder how you're supposed to connect all of this, don't worry, we'll get there. But first, we need to talk about why AI needs the connection in the first place.
Why is data so significant?
AI doesn't reason the way a person does. It doesn't deliberate. What it does, at its core, is recognize patterns in observations it has seen before, and use those patterns to make predictions about new situations. It uses probabilities to guide it. The more it has seen, the more patterns it can recognize, and the more reliable the predictions become.
The patterns that matter most in a business, in my experience, almost never live inside a single system. They live at the intersections. A material delay matters because it touches the schedule. A schedule slip matters because it touches the labor plan. A design change matters because it touches the budget and the field crew. The real value of AI is its ability to pull insight out of the relationships between datasets that no one has ever looked at side by side.
So why is this rarely done? That work is impossible if the data isn't connected. But it’s absolutely within reach if it is.
Fixing data fragmentation
So what do you actually do about this? The most common mistake I see is starting in the wrong place. Companies hear AI is important and go shopping for AI tools. They end up with a stack of impressive technology that still can't see across the business.
The order has to be different.
Start with value. What decisions in your company are slow today? Which are expensive? Which are being made with information that should be more complete than it is? Those are the questions that should drive everything else. From there, work through the risks you need to govern. Security, compliance, the cost of a bad recommendation. Then figure out how to enable your workforce to use these new tools safely. Only at the end does the data conversation come into focus. Which datasets need to be connected. In what formats. At what speed.
Done in that order, the work has a clear destination. Done in reverse, you spend years cleaning and securing data without knowing what it's for.
And underneath all of it, the practical move I see in every organization that's pulling ahead right now is investing in integrating data across the entire landscape, so the next tool you adopt has a connected foundation to work from.
Where will the future of AI go?
The companies most fluent with AI today are still using it the way they used the first generation of chatbots. You ask it something, it gives you an answer, you move on. What I think changes over the next few years is that workers stop being users of AI tools and start being builders of them.
The interesting move right now is that I can describe a task in plain language and have AI package it into something I can drop into my own toolbox. The next person on my team doesn't need to know how I built it. They just use it. Multiply that across an organization and you have something fundamentally different from a chatbot in the corner. You have a workforce that is quietly inventing new capabilities every day. But every one of those custom tools is going to draw on whatever data the worker can reach.
Custom tools are only as powerful as the data powering them. If the data is fenced off in disconnected systems, the tools built upon it are small. If data is poor-quality or remains trapped in spreadsheets, outcomes will be unreliable. Only when data is connected and accurate can the tools we build be unbounded. The ceiling on individual productivity, it turns out, is the same ceiling we've been talking about all along.
Hear more from Tim Long and leaders from Microsoft, NVIDIA and Trimble in the “Beyond the Hype” AI webinar recording.
About the author
Tim Long is Global Head of Manufacturing at Snowflake, where he leads industry strategy focused on helping manufacturers and their partners unify data and apply AI to drive operational performance. Tim brings deep experience in advanced analytics, machine learning, and applied AI across manufacturing, semiconductors, and consumer goods. Prior to Snowflake, he led the data and analytics function for North America at Adidas, and the enterprise analytics and data science function at Micron Technology.
