The Value of AI in Manufacturing is Accelerating
There is a lot of talk about the value of artificial intelligence in manufacturing, and rightfully so. Although AI isn’t new, it appears to be reaching a tipping point where companies are more open to exploring its potential and AI techniques are more accessible than ever. Manufacturers are acting on the opportunity. Our recent Making Manufacturing Analytics and AI Matter research, for example, shows that 99% of manufacturers plan to invest in analytics and AI in the next year.
Further, AI has moved from the experimentation phase to the value phase. Our latest executive survey, Executive Strategies for Sustainable Business Success 2024, found that AI / ML (Machine Learning) was the most common application type providing value, with 88% of responding companies reporting they achieved business value from AI / ML. Further, these results can come quickly. The manufacturing analytics and AI study found that companies achieve fast ROI more frequently from advanced analytics / AI than from their other software investments. For example, the data shows that 73% of respondents received benefits from GenAI in less than a year.
Talking about AI Adoption with the Experts
Our research shows that AI / ML can drive value, and companies can achieve that value rapidly. But that value doesn’t come from buying software. It comes from applying the right solution to solve a real problem. However, many manufacturers don’t know where to start or how to implement these capabilities. To better understand how manufacturers can target and adopt AI, we sat down with two experienced consultants with real-world experience in helping manufacturers improve business value by adopting AI. We sat down with Kalypso: A Rockwell Automation Business‘s Senior Manager of Data Science and Digital Transformation, Chelsea Barnes and Master Data Consultant William Rosengarten to get their perspective. Let’s see what they have to say!
What are Manufacturers Asking For Today?
Jim Brown
Kalypso works with a lot of manufacturing companies. What are manufacturers asking for related to their AI strategy and implementations?
Chelsea Barnes
At one end of the spectrum, some companies are enthusiastic but lack direction. They know they need to do something with AI because it’s a buzzword or their CEO says it’s important, but they don’t know what that looks like. They need help setting a direction. On the other end of the spectrum, some clients have a very specific problem and sometimes possibly even a solution in mind, like needing a machine vision solution to detect malformities or burns on a chip because it’s costing them $10 million in scrap every year. They need help validating and implementing a solution. But there’s also a middle ground where companies have operational targets in mind and maybe some initial hypotheses about how to improve them but aren’t sure where to start.
Regardless of what the starting point looks like, the core of the ask is the same. They want value delivered quickly, at scale, using the best advanced technologies available.
Targeting Business Value versus Technology
Jim Brown
I’ve known Kalypso for some time and I appreciate that you don’t believe in technology for technology’s sake but focus on adding business value. How do you get companies started or help them frame their problem?
Chelsea Barnes
We help them discover where their business problems really are and what technology solutions are best suited to those problems. Then, we bring those things together. When we talk to a company, they know their issues far better than we do. For example, the people operating a line will be able to specifically articulate what problems are happening and have a very good hypothesis as to why they’re happening. Then, we bring our business, operational, and technology expertise to those conversations so that they meet in the middle with solutions.
Jim Brown
One of the things I appreciate is that you’re not just technologists. You are domain experts who understand operations and the manufacturing industry. For example, in the Consumer Packaged Goods (CPG) industry, when you mentioned “burns on a chip,” I knew right away you meant potato chips and not microprocessors. Can you tell me a little bit about why it’s important that advisors don’t just approach their clients with AI knowledge, but also bring relevant business expertise to the table?
William Rosengarten
We’re not coming in cold because we have a depth of expertise in the industry. We already have a point of view on the end goal. If a client comes to us with a problem, we know what best-in-class in the CPG industry looks like, so we can help them create a plan to achieve it.
Chelsea Barnes
Exactly. We bring together a variety of expertise to make that happen. We’re coming in with a really solid set of hypotheses around what the problems typically are. We’re familiar with approaches to improve quality yields and deal with issues like variable material inputs that cause problems for food and beverage clients. The specifics come from the client and their own knowledge, but our experience helps us get to a diagnosis more quickly.
Prioritizing the Right Opportunities
Jim Brown
In a recent cross-industry survey, we asked companies about their AI goals. The most common goals identified across industries were product and service innovation, product and service performance, and workforce efficiency. Those are essential in any industry. A survey specific to the manufacturing industry, however, clearly identified cost reduction as the most common investment driver. What are the CPG companies you’re working with looking for?
Chelsea Barnes
We’re seeing the same thing on the ground. There are two macro trends that are really squeezing manufacturers right now. The first is inflation, which increases cost pressures. The other is workforce turnover, including a wave of seasoned specialists leaving the workforce, which puts a new sense of urgency on workforce efficiency. To meet those cost reduction and efficiency goals, the top AI use cases we hear are around quality control, process optimization and predictive and prescriptive maintenance.
William Rosengarten
We also see a common pain point in accessing the right data, especially when working with time series data. Five years ago or so, manufacturers felt they needed to capture everything from the plant floor, store it in the cloud, and historize it. So many manufacturers have created a giant haystack of all of their data, and they’re struggling to find that needle that will drive specific use cases like the ones Chelsea is describing.
Justifying Projects
Jim Brown
With all of the potential projects you may identify with a client, how do you help them decide on what to focus on? Do you counsel them to focus on the most significant problems, or maybe try to have them find repeatable problems? Or is it purely the project with the largest ROI?
Chelsea Barnes
Manufacturers are absolutely looking at ROI. They need to understand how it will affect the process, the tangible value they will get from the initiative, and how they will measure achievement. It’s critical that they know what their quantifiable goal is.
However, when it comes to investments in digital, sometimes the value isn’t as clear-cut as a 12-month payback. In some cases, companies are looking to stay ahead of the competition by operating on the bleeding edge of innovation. This might justify a more long-term investment approach to allow the transformation they’re looking for to take root.
Getting back to determining ROI, we’re big proponents of rapid use case identification and prioritization, where you quickly narrow down your short list of high-potential opportunities before investing too much time rigorously evaluating all options. To do this, you need a good value calculation framework, which we bring to all of our assessment projects. But,you also need the right technologists in the mix to help you quickly vet the solutions and estimate implementation complexity to understand the cost of an initiative.
Choosing the Right Technology
Jim Brown
Generative AI is on most peoples’ minds and has become popular in conversations because of OpenAI and ChatGPT. However, many other AI and machine learning (ML) techniques are available. AI can be applied at different levels, ranging from companies wanting to retrieve data more effectively to the other end of the spectrum where they are pursuing AI-driven autonomous, real-time decisions to drive equipment behavior on the floor. How do you help your clients decide what technologies to apply for a specific problem?
Chelsea Barnes
We always start by confirming the business needs and what’s the problem to solve now. Even if they come to us with a very specific request, like “I need a machine vision solution,” we will diagnose the issue together and then confirm that’s the right solution. We don’t tell our clients to “go GenAI” their business. That wouldn’t be good business for them, and it’s not good business for us. We have a collection of tools in our toolbox to bring to this equation depending on the problem to solve and the data they have to work with.
William Rosengarten
We make sure to map technologies to business needs. For AI technologies, we consider:
- What data types are we working with?
- Is it structured data?
- Is it time series data?
- Is it natural language data?
- What kind of action or decision are you trying to take?
- What is the risk of error in that decision?
- Is there a human interaction component that would be an essential decision-making factor?
The choice will be different if they just need to organize and retrieve data quickly or if they’re looking for insights from the data they have. For example, you shouldn’t use copilots for autonomous control, but it’s valuable when a human is in the loop for decision-making. These questions help drive considerations about the modeling and architecture that should come into play.
Chelsea Barnes
We always look at what kind of algorithmic approach will be best suited for the scenario. While generative AI is the topic of the day, there are plenty of cases where you should not be using it. For example, a GenAI model will not help make a prediction to autonomously control a production process – think predicting fill by monitoring time series data so you can adjust your filler dosage so it comes exactly at target. It’s just not suited to do that. But if you are trying to process something like a year’s worth of shift logs to find anomalous patterns in those free text shift logs, that’s a situation very well-suited for a large language model.
Two other important decision criteria in regulated scenarios are the risk of error and whether the results are explainable. GenAI models, which are neural networks, are by nature black boxes where you don’t know how it arrived at a decision, so having a human operator in the loop is critical to confirm the results.
A Closer Look at Copilots
Jim Brown
AI copilots are gaining a lot of traction to streamline and improve human workflow. When do you find those applicable for your clients?
Chelsea Barnes
A copilot makes sense when they are trying to augment what a human can do, to make them more efficient in a process, or help them with the decisions that they are making. A good example for an operator would be a troubleshooting copilot. For example, a line is down and a fault code comes up. Instead of looking that up manually, the copilot could take the operator through a decision-making process and walk them through the troubleshooting steps.
Copilots are attractive because the manufacturing industry has not fully rebounded following COVID, and many companies still have jobs left unfilled. Retaining institutional knowledge in manufacturing is even more of an imminent and challenging concern as a substantial percentage of the workforce nears retirement. Many companies would love to get to lights-out manufacturing, but that can be decades away. So the goal is to find the best way to augment and assist the workforce they have. Copilots can help make them more productive and efficient, and equip them with decades of institutional knowledge, even if they haven’t worked on the line for 15 years.
William Rosengarten
Agree. Copilot assistants are an excellent solution for capturing and retaining institutional knowledge. For example, they are very good at taking notes. A technician running a troubleshooting process is trying to get the line back up and running and typically doesn’t have time to document what they’re doing. They are making decisions on what steps to take based on their experience. A copilot could take notes about the decisions they make and the impacts they have on the troubleshooting process. Doing this creates a feedback loop that typically only exists in free text or just in a technician’s head and tribal knowledge today. In that way, copilots can help guide troubleshooting and feed information into a knowledge repository to assist in future troubleshooting efforts.
Key Takeaways
You’ve shared a lot of insights into how manufacturers can identify the right business opportunities and apply AI to solve them. Two of my key takeaways are that it’s essential to have industry expertise to help diagnose the problem and that it’s critical to have diverse technical knowledge to be able to apply the right AI capabilities to get the job done. This is an exciting topic, and we’ll stay in touch about it.
Additional Resources
To learn more from Kalypso, explore the Kalypso website, a Kalypso interview about operational CoPilots, or Kalypso Insights on GenAI.
Thank You
Thank you to Kalypso, a Rockwell Automation Business, and Hadley Bauer for arranging the interview. We learned a lot from the discussion and know manufacturers will, too.