The growing momentum for advanced technologies such as machine learning, deep learning, and Industrial Internet of Things (IIoT) in manufacturing is undeniable. But we all know that, just because technologies hold promise for improving processes or productivity, it doesn’t make them a good investment for your business. And figuring out if these types of solutions make business sense is challenging.

First you need to track down reliable information about how much value you could realistically expect a solution to deliver and weigh that against the potential costs.

Based on our research, you are likely to find that many machine learning and IIoT solutions aren’t driving investment-worthy value … yet.

It’s not that the solutions don’t deliver improvements. Rather, it’s that the costs or challenges associated with them are currently prohibitive and/or the improvements are often statistically insignificant compared to more traditional methods and solutions. For example, our review of several machine learning research papers shows that you might only get something like a 2% improvement for demand forecasting using machine learning compared to multiple linear regression analysis. Clearly, that improvement could be of value to a large and complex enterprise. But for most organizations, there are usually far less expensive ways to make similar gains.

Of course, there are many different potential applications for machine learning and IIoT solutions, and some hold more promise than others. Let’s take a look at what some research has shown as well as a few key things to consider as you look for ways to drive continuous improvements in your operations:

Breakthrough or incremental improvements? Research insights at a glance.

This paper explores the potential of machine learning for forecasting “distorted demand at the end of a supply chain (bullwhip effect),” comparing machine learning against several other methods. The researchers used a data set from a simulated supply chain as well as a data set from Canadian Foundries orders as the basis for their research. They discovered that, although the advanced techniques did lead to more accurate forecasts, the improvements were not significant enough to justify the additional complexity and cost compared to a simpler linear regression model.

This paper provides an overview of deep learning techniques and a brief history of machine learning. It includes a summary of some of the different ways advanced analytics methods have been used for smart manufacturing as well as a comparison of traditional machine learning and deep learning techniques. The paper, which is a good resource for understanding what it takes to make a machine learning algorithm succeed, discusses many uses of advanced techniques throughout the manufacturing process, from machine failure to product failure. Interestingly, it highlights the importance of good data (not just lots of data) and well-defined use cases to success with these technologies.

This paper looks at trends related to machine learning in manufacturing and includes a case study about analyzing tool wear in a small automotive shaft manufacturer using regression algorithms. The research highlighted in the paper makes a good case for the potential of machine learning to predict tool wear quickly and accurately, which could lead to quality improvements. The costs per machine, however, would be very high, making the business case hard to justify in many cases.

Analysts are certainly more bullish on the potential of combining sensors and cognitive models to improve supply chain performance. For example, IDC predicts that analytics-driven cognitive capabilities could help one-third of manufacturing supply chains improve cost efficiencies and service performance by 10% and 5% respectively by the end of 2020.

It’s also worth considering that, although sensors may be cheap and getting more data may be easy, these solutions will still add significant costs. Just consider that data scientists are in high demand, and new graduates are earning more than $100,000 just out of school. In other words, in most cases the incremental value a solution could create versus the cost you will add to your business to analyze all of the data and do something useful with it means net business value is debatable.

These simple examples are not meant as an indictment of or an argument for the technologies. But they are a good reminder for why it’s worth proceeding with caution and a well-thought-out plan.

Fortunately, you don’t need a huge budget or vast resources to start your journey toward IIoT and machine learning. Given the uncertainty of the benefits with machine learning as well as security risks and other challenges with IIoT, most companies will benefit from a phased approach. And before you start seriously thinking about either technology, it’s important to consider how well your company is currently able to use its data from across key systems to keep operational performance humming. For example:

  • Do you have capabilities for collecting and integrating data from across the organization?
  • What capabilities and methods are you currently using for data analysis, and how do your related processes help drive continuous improvements?
  • How well do your teams track and manage improvement projects?

Depending on your answer to these questions, it might make sense to pursue a more advanced technology solution. But even then, finding a way to run a contained pilot that will allow you to accurately gauge the value of a given solution should be a top priority. And to that point, it’s important to ask:

  • What exactly do you want to do or achieve?
  • What capabilities are needed to do that and why?
  • Do you have the process rigor and necessary resources to handle a potentially complex project and all of the additional data it will produce?

If you struggled with the first set of questions above, then it’s crucial to consider how you start improving performance across the organization and building a foundation for advanced technologies. The right approach could help drive operational performance improvements of 15% or more while setting you up for greater success with future initiatives.

Integration is a key first step because knowing what is happening at all critical points of your business in real time has become essential to staying competitive. Once you integrate your data and can really see what’s happening in the business and why, you can continually fine tune operational performance and processes. And in the meantime, you’ll gain a much better idea for what other types of technologies you really need to continue improving.


Ken Koenemann leads TBM’s Supply Chain and Technology practices and is currently leading new product development for TBM’s proprietary software business, Dploy Solutions.