The Data Problem Nobody Wants to Talk About: Turning AI & Security Strategy into Reality

 In this episode of The Connected Frontier, we explain that most organizations face a data problem rather than a technology problem when implementing AI. We highlight how inconsistent definitions, fragmented pipelines, and unusable data erode the trust necessary for AI-driven decision-making. To overcome these hurdles, we recommend focusing on making data "fit for purpose" by aligning specific data elements to actual decision points. 

Transcript

Speaker 0:02

Welcome to the Connected Frontier, the podcast where we navigate the technology shaping our world. From securing the industrial Internet of Things to decoding the next wave of cybersecurity to preparing for a post-quantum future. This is where complex ideas become clear. This is the Connected Frontier.

Speaker 0:27

Welcome to the Connected Frontier. There's a lot of conversation right now about AI, security, and the future of the enterprise. But most of it lives at a high level, and that's where things start to break down. In this series, we're focused on what it actually takes to turn strategy into execution, what works, what doesn't, and where organizations tend to get stuck. I'm Katherine Blough, and this is where strategy meets reality.

Speaker 1:03

In the last episode, we talked about architecture and why tools can't compensate for a weak foundation. But even with the right architecture in place, there's another issue that shows up almost immediately, and it's one that most organizations know exist, but don't always address directly, and that's the data problem. Let's just say this up front. Most organizations don't have a technology problem when it comes to AI. They have a data problem, and not in a vague general sense, in a very specific, very practical way that shows up during execution. Because data isn't just something you plug into a model. It's something that has to be defined, structured, aligned, and trusted. And if any one of those breaks down, everything built on top of it starts to struggle.

Speaker 2:02

Part of the challenge is how data is talked about. At a high level, it sounds manageable. We have a lot of data. We need better data governance. We need to clean the data. But those statements don't reflect what's actually happening on the ground. Because data problems aren't centralized. They show up differently in different parts of the organization. And they tend to surface only when you try to use the data, not when you're just storing it. This is where things start to break down because data doesn't fail all at once. It fails at the point of use.

Speaker 2:41

Let's make this concrete. There are a few patterns that show up over and over again. First off, inconsistent definitions. This is one of the most common and most overlooked issues. The same term means different things in different systems. Customer, order, incident, asset. Each team defines it slightly differently based on their needs. And that works until you try to bring those data sets together. Then you realize you're not working with one version of the truth. You're working with several. Second, data that exists but isn't usable. Organizations often have the data they need, but it's incomplete, it's delayed, it's poorly structured, or it's difficult to access. So technically, the data exists, but operationally, it's not usable in the way the strategy assumes. Third, we have fragmented data pipelines. Data moves through multiple systems, each with its own logic, timing, and transformation. So by the time it reaches the point of decision making, it may no longer reflect the current state of the environment. And for AI, timing matters. A delayed insight can be just as problematic as an incorrect one. And finally, the fourth one, lack of trust. This is the one that ultimately determines whether anything works. If teams don't trust the data, they won't trust the outputs but on top of it. And when trust breaks down, people revert to manual processes and personal judgment.

Speaker 4:27

Let's go back to a security example. An organization is using AI to prioritize alerts. The model is trained on historical data, incidents, behaviors, response patterns, but that data has inconsistencies. Some incidents were fully documented. Others were partially recorded. Some are handled differently depending on the analyst. So the model learns from a data set that isn't fully aligned. And when it starts making recommendations, they don't always match what experienced analysts expect. So what happens? They start second-guessing the system, overriding decisions, and eventually relying less on the AI. Not because the idea was wrong, but because the data foundation wasn't strong enough to support it.

Speaker 5:17

Now let's look at operations. An organization is implementing AI-driven demand forecasting. They're pulling data from ERP systems, historical sales, supplier inputs, and market signals. But each of those sources has its own structure and timing. Sales data may be delayed, supplier data may be incomplete, external signals may be inconsistent. So when the model generates a forecast, it reflects a version of reality that's slightly out of sync. And the operations team who has to act on that forecast sees the gaps immediately. So again, what happens? They adjust it manually, they rely on experience, and the system becomes a reference point, not a decision driver. This is where the data problem becomes more than just a technical issue because it directly impacts decision making, speed, and confidence. If data isn't aligned, decisions slow down, automation is limited, and outcomes become inconsistent. And over time, that erodes trust in the broader strategy.

Speaker 6:31

So how do you approach this differently? Not by trying to fix all the data. That's not realistic. Instead, focus on making data fit for purpose. Start with the use case and work backward. Ask, what decisions are we trying to support? What data is required to make those decisions? What level of accuracy and timeliness is needed? Because not all data needs to be perfect, but the data that drives decisions needs to be reliable. A few practical shifts. Define critical data elements clearly. Not everything, just what matters for the use case. Then align data to decision points. Make sure the data supports how decisions are actually made. Then you need to prioritize consistency over completeness. Consistent data that's slightly limited is more useful than inconsistent data that's comprehensive. And finally, build trust intentionally. Transparency matters. People need to understand where data comes from and how it's used.

Speaker 7:49

In the next episode, we're going to shift from data to something that sits alongside it. And that's risk. Because as organizations start to rely more on data and AI-driven decisions, the question becomes: how do you manage risk in a way that doesn't slow everything down? At the end of the day, AI and automation are only as strong as the data behind them. And if the data isn't aligned, execution will always struggle, no matter how strong the strategy is. Thanks for listening to the Connected Frontier. I'm Katherine Blough, and this is where strategy meets reality.