Manufacturers are under growing pressure to deploy AI across operations, yet many are still struggling with a more basic problem: understanding how their factories run. As digital tools proliferate, the real challenge is no longer collecting data, but turning it into decisions that improve productivity, energy use, and operational stability on the shop floor.
For all the noise surrounding artificial intelligence in manufacturing, the reality on most factory floors remains stubbornly grounded. Machines stop unexpectedly, utilisation is lower than assumed, and improvement initiatives stall because nobody can agree on where the real problems sit. According to Chris Iveson, CEO of FourJaw Manufacturing Analytics, this disconnect is not caused by a lack of technology, but by a misunderstanding of what technology is supposed to do.
“AI is a really valuable tool and it is clearly changing the world,” Iveson says. “But it is not a solution in itself. Manufacturers do not want technology for the sake of it, they want solutions to very specific problems. If AI can help solve those problems, then it absolutely has a role. If it cannot, then it just becomes another layer of complexity.”
That distinction matters because much of the industrial software market has raced to position AI as a default capability. In practice, many manufacturers find themselves with platforms that promise intelligence but struggle to deliver insight that can be acted upon. “You see a lot of software with an AI bolt-on,” Iveson says. “You play with it, you can see what they tried to do, but it does not actually solve the problem you have on the shop floor.”
From data collection to operational clarity
Early approaches to manufacturing analytics often focused on connecting directly into machine control systems. While this promised detailed data, it also created friction. Factories are rarely homogeneous environments, and integrating with every machine controller quickly becomes expensive, slow, and incomplete. Worse still, much of the data gathered in this way offers little operational value.
“When we started, we were connecting into control systems and collecting huge volumes of data,” Iveson explains. “But when we spoke to manufacturers, most of that data was not what they cared about. What they really wanted to know was very simple: when are my machines running, when are they not, and why.”
That insight reshaped how data could be gathered and used. Rather than trying to understand everything a machine does internally, the focus moved to outcomes that matter operationally. “If you can classify a machine into a productive state or a non-productive state, and then understand the reasons behind that, you suddenly have a very powerful view of what is happening across the factory,” he says.
This approach also lowers the barrier to adoption. When analytics can be deployed quickly, without deep integration work or specialist skills, it becomes far easier for manufacturers to experiment and scale. “We deliberately wanted something that customers could install themselves,” Iveson says. “That way, we could ship it anywhere in the world and they could be up and running almost immediately.”
Why dashboards alone do not deliver improvement
Despite the availability of real-time data, value extraction remains uneven across industry. Many factories are already surrounded by dashboards, yet productivity does not improve. The difference, Iveson argues, lies not in the data itself but in how it is used.
“You can collect all the data you like, but if you do nothing with it, you get nothing from it,” he says. “The manufacturers who see results are the ones who have a culture of improvement. They use the data to ask where the biggest problems are and what they are going to change.”
That process requires discipline and focus. Rather than attempting to fix everything at once, effective teams use analytics to prioritise. “If you can see that most of your downtime comes from one or two recurring issues, then that is where you start,” Iveson says. “You make a change, and then you measure whether it worked.”
In many cases, data also challenges long-held assumptions. “We have seen situations where people believed machines were running most of the time, and the data showed they were running far less,” he says. “Once that is visible, the conversation changes completely. You are no longer arguing opinions, you are dealing with facts.”
Escaping pilot purgatory
One of the most common failure modes in digital manufacturing is what Iveson describes as pilot purgatory. Companies trial new tools on a handful of machines, see some promise, but never move beyond proof of concept. Often, the underlying issue is risk.
“It is very tempting to treat digital projects as big bang initiatives,” he says. “You spend a lot of money upfront, do a pilot, and if it does not immediately transform everything, the project stalls. We deliberately wanted to remove that risk.”
By enabling low-cost, small-scale trials, manufacturers can prove value before committing to wider deployment. “If you can show that even a small number of machines are delivering measurable improvement, then scaling becomes a business decision, not a gamble,” Iveson says. “That is when things actually move.”
This incremental approach also aligns better with how factories operate day to day. Rather than disrupting everything at once, change happens in manageable steps, with clear feedback loops.
The real role of AI on the shop floor
Against this backdrop, AI takes on a more grounded role. Rather than being the starting point, it becomes an amplifier once reliable operational data exists. “AI really comes into its own when you have multiple datasets telling different parts of the same story,” Iveson says. “That is when it can start to cross-correlate information and surface insights that are hard to see manually.”
However, many valuable decisions do not require advanced modelling. “You do not need AI to see that a machine you thought was running 60% of the time is actually running 30%,” he says. “You can act on that immediately.”
Equally important is how data is perceived by the workforce. Monitoring systems can be viewed with suspicion if they are framed as surveillance tools. “There is always pressure on operators to deliver,” Iveson says. “What data does, when used properly, is give them evidence of the things that get in the way of doing their job.”
That shift can be empowering. “If an operator can show that time is being lost because of missing tools, breakdowns, or staffing issues, then the problem becomes visible to management,” he says. “It moves the conversation away from blame and towards fixing the real issues.”
Ultimately, the path to smarter manufacturing is less about chasing autonomy and more about establishing truth. Before factories can automate decisions, they must first understand them. As Iveson puts it, “Once you have a single source of truth about what is happening on the shop floor, everything else becomes possible. Without that, even the most advanced technology struggles to make a difference.”
