Credit: Mimi Phan | Na_Studio/Shutterstock

6 steps to succeeding with industrial AI

Beth Stackpole

Sep 29, 2025

What you’ll learn:

Adapting system dynamics concepts to industrial AI initiatives can accelerate adoption and help organizations avoid mistakes.
Companies need to gather the right data and make sure it is shared with the right people at the right time. Employees need this information to make real-time decisions.
Frame challenges as control problems and remember to simplify, by limiting project scope or using smaller datasets, for example.

Despite widespread access to new technologies, many companies struggle or fail to integrate industrial AI into their manufacturing operations. Adapting system dynamics concepts to these initiatives can help organizations accelerate adoption and circumvent mistakes, leading to better business outcomes.

Companies need to harness data to build feedback loops that will lead to more intelligent insights, increased efficiency, and, ultimately, better decision-making. MIT Sloan senior lecturer explores these concepts in the new MIT Sloan executive education course Strategy, Survival, and Success in the Age of Industrial AI.

“We need to think differently about our systems if we want to be the winners in the industrial AI adoption game,” Carrier said in a recent webinar. “Data collection and data filtering technologies, machine learning, and new types of AI can elevate informal feedback loops, making them much more successful. Anywhere we can put tools in place that get information faster to the people that need it is a huge win.”

An action plan for industrial AI

To refocus on industrial AI and automation efforts through a system dynamics and controls lens, consider the following:

Don’t just aim for a lot of data; gather the right data. Data is central to industrial AI efforts, but low-quality data or an overabundance of ancillary data can do more harm than good. Carrier pointed to the example of an oil refinery that missed an opportunity to avoid a dangerous malfunction because critical indicators were masked in a sea of alerts. “They had too much data, so the information needed got lost in the abundance,” he said. “What you need to do is take data and turn it into information that helps reduce risk.”

Get data in the hands of the right people at the right time. Sharing industrial data through spreadsheets, websites, or well-crafted dashboards only goes so far in building useful feedback loops. The real art is taking these dashboards and signals down to the factory floor so the people doing the work have the data they need to make real-time decisions.

Foster a culture that’s open to data-driven insights. The best leading indicators and AI-driven insights are useless if employees are distrustful of data or not open to information they may not want to hear. For example, indicators could show that maintenance is required or that more training is necessary — activities that might demand additional funding or invoke downtime on the line. “A large part of benefiting from AI is having a culture that is going to accept data that tells us things we don’t want or filtering out data that’s not relevant at the moment,” Carrier said.

Apply control theory for problem-solving. Frame challenges as control problems rather than AI technology problems. Applying quantitative measures to sequence processes is an effective way of fully capitalizing on systems to boost profits or reduce risk. “Realize the systems are trying to speak to us and we help by listening to what they are saying and creating that corrective feedback loop,” Carrier said. When applying control principles, it’s also important to play close attention to getting the time constraints right.

Buy time for your system. The OODA loop is a decision-making process that helps guide where to invest time and money to develop faster responses to disruptions. Industrial AI data can be used to optimize the four-step process (observe, orient, decide, and act), especially in the observe and orient stages, where data can have an impact. Visual recognition and pattern detection, powered by data and AI, are valuable for the final two acts in the OODA loop, especially to help detect defects and solve problems in a high-speed production line.

Remember that simpler is better. The complexity of AI systems, models, and data can quickly subsume projects and slow down results. Simplifying those systems, whether by using a smaller set of models and data for large language model implementations or limiting the scope of the initiative, is a proven way to ensure faster and more meaningful results. “Before you go out and buy these massive technologies that can do everything, focus on the problem at hand,” Carrier said. “Applying the 80/20 rule is really what it takes to be successful.”

John Carrier is a senior lecturer of system dynamics at MIT Sloan. He instructs senior managers on improving manufacturing and business processes and serves as an on-site coach in support of projects. His research focuses on strategic marketing and new business development in high technology, specialty chemicals, and service segments.