How to go from visibility to problem solving

Manufacturers have spent the last decade building visibility. ERP platforms, MES systems, connected assets and industrial analytics tools have created unprecedented access to operational data. Most organizations can now see more of their operations than at any point in their history.

The question is whether that visibility is solving anything. Josh Zable, President and CFO at Minitab, believes many manufacturers have become trapped in a cycle of collecting information without fully understanding how to convert it into operational improvement. While enterprises continue to invest heavily in dashboards, data platforms and AI initiatives, many of the day-to-day challenges facing production teams remain stubbornly familiar.

“Something like 98 per cent of manufacturers report data problems,” Zable says. “There’s a lot of data but accessing that data and making sure it’s accurate can be challenging. Many manufacturers have expensive systems that are built to collect information, whether it’s an ERP, an MES or a QMS. The promise is that you’re going to collect all this data and get all these insights, but the reality is that the person on the shop floor often isn’t getting the information they need, can’t validate that it’s accurate and can’t use it in real time.”

For Zable, the problem is not a lack of visibility; it’s the right visibility. In many cases, manufacturers already possess more information than they know how to use. The challenge is understanding which operational problems need solving and identifying the data that will help solve them. That distinction, he argues, separates organizations that simply monitor performance from those that actively improve it. Shop floor specific solutions, targeted at solving specific challenges, often go overlooked when larger systems are in place, even if they would provide significant value.

Why more data is not helping

The assumption underpinning many digital transformation initiatives is straightforward. Better visibility should lead to better decisions, which should ultimately lead to better outcomes. While there is some truth in that logic, Zable argues that manufacturers often overestimate what large scale visibility alone can achieve.

“These systems create value because they capture data and provide a high-level view of what’s happening,” he says. “If you’re the head of manufacturing or the CFO, they can help you understand costs, performance and what’s happening across the business. But when you start drilling into a specific operational problem, they don’t always help you solve it.”

This is where many manufacturers become stuck. Enterprise systems are often effective at identifying that a problem exists, but much less effective at explaining why it exists or how it should be addressed. A quality issue, a throughput constraint or a recurring downtime event may be visible, but visibility alone does not provide a solution.

“I fundamentally believe there is value in these larger systems, but they come with promises of solving every problem and they ultimately solve a few,” Zable explains. “The person on the manufacturing floor has an issue today. They have a fire they need to put out today. They need to improve throughput today or improve quality today. That’s very different from having visibility into what’s happening across the entire organization.”

Zable compares the situation to the way sales and marketing teams use CRM systems. The CRM provides a central source of information, but organizations still rely on specialist tools to improve individual aspects of performance. Manufacturing increasingly requires the same approach. High level visibility may provide information, but specific solutions are required to identify where attention is needed that can lead to improvement.

“There is value in capturing the data and creating a top-down view of what’s happening,” he says. “That information may help you identify where a problem exists. But once you drill into that problem, you need tools that actually help you solve it.”

Not all dashboard are created equally

The distinction becomes particularly apparent when enterprises invest heavily in dashboards and business intelligence platforms. Almost every manufacturer now has access to visualization tools capable of displaying performance metrics in real time. Yet the presence of a dashboard does not automatically translate into operational improvement.

“The question is whether you have the right dashboards,” Zable says. “Everybody has dashboards. Business intelligence has become commoditized. The real issue is whether you’re looking at information that helps you make a decision. Ask yourself this: do I have a manufacturing dashboard or am I using the same tool they use across my organization?”

In some cases, visibility can deliver substantial benefits. Zable points to examples where organizations achieved significant gains simply by making performance data more accessible to operational teams. “We’ve seen productivity improvements of around 30 per cent just by having visibility into uptime and downtime. We’ve seen quality improvements of up to 50 per cent because people can see a trend developing before it becomes a serious issue. Those are valuable dashboards because they help people intervene before problems escalate.

“However, many dashboards stop at reporting what has happened. They rarely explain why it happened. You can see production going down, but what is causing it? You can see quality changing, but what’s driving that change? If you’re not answering those questions, you’re getting information rather than insight.”

That distinction sits at the heart of the debate around industrial data. Knowing that a problem exists is useful. Understanding the factors causing that problem is what enables companies to improve performance. Manufacturers increasingly need tools capable of moving beyond visualization and into root-cause analysis, prediction and optimization.

Why AI needs statistics

AI has become the dominant technology discussion across manufacturing, but Zable believes many organizations risk focusing on the wrong aspects of the technology. “I work for a statistics and machine learning company, so I always remind people that machine learning was the original AI,” he says. “When people talk about AI today, they’re usually talking about large language models, but that’s only one part of the picture.”

The distinction matters because different technologies create value in different ways. Statistical analysis and machine learning help manufacturers identify patterns, understand relationships and predict outcomes. Large language models excel at making information easier to access and understand.

“Large language models are very good at explaining what you see,” he adds. “They’re not particularly good at predicting what will happen. Statistical analysis and machine learning help you improve processes, reduce variation and predict outcomes. Large language models help you interact with that information.”

For manufacturers eager to deploy generative AI, the implication is clear. The underlying analytical foundations still matter. “If you’re not doing statistics and machine learning and you’re jumping straight to the large language model, you’re missing a lot of value creation,” he continues. “AI can accelerate your ability to understand what’s happening. It can help explain the results. But if you haven’t built the analytical foundation underneath it, you’re limiting what it can achieve.”

The same principle applies to data quality and governance. While these areas attract less attention than AI, they remain critical to success. “People get excited about AI, but governance isn’t exciting. Cleaning data isn’t exciting. Building compliance frameworks isn’t exciting. Yet those things are absolutely critical. Garbage in, garbage out still applies.”

Start with one problem

For manufacturers feeling overwhelmed by data and underwhelmed by results, Zable advocates a simpler approach. Rather than starting with technology, organizations should start with the operational problem they are trying to solve. “Generally speaking, you have a throughput problem or a quality problem,” he says. “Or you don’t have a problem, but you want to improve throughput or improve quality. That’s the place to start.”

Once the problem is clearly defined, the focus shifts to understanding which variables influence performance and which factors are creating unwanted variation. Predictive analytics can play a particularly valuable role because it helps enterprises identify the drivers of performance rather than simply reporting outcomes. “One of the most powerful things about predictive analytics is that it doesn’t just give you predictions. It gives you insight into your process,” he continues. “It tells you which factors matter and which don’t. It cuts through the noise and helps you focus on the things that are actually influencing outcomes.”

This focus on problem-solving is something Zable believes many digital transformation initiatives have lost. Technology should support improvement, not define it. “I think people sometimes start with the technology and then look for a problem,” he says. “The better approach is to understand the problem first. Use something like the five whys methodology. Keep asking why until you get to the root cause. Once you understand what you’re trying to improve, technology becomes much easier to evaluate.”

Looking ahead, he sees human expertise, statistical analysis and AI becoming increasingly complementary rather than competing disciplines. “I think you need all three,” he concludes. “You need process knowledge and critical thinking from people who understand the operation. You need statistical analysis and machine learning to understand what’s happening in the data. Then AI can help explain those insights and make them more accessible. They all work together.”

Manufacturers will continue to invest in visibility, connectivity and AI. The manufacturers that derive the greatest value from those investments, however, are unlikely to be those collecting the most information. They will be the manufacturers that understand which operational problem they are trying to solve before deciding which technology to deploy.