Move Forward Faster

Machine Learning & Supply Chains: What Works, What Doesn’t

By Maria Marchesi, GAINS

Artificial intelligence (AI) and machine learning (ML) hold the potential to improve supply chain performance dramatically. Businesses can use AI and ML to optimize performance across the supply chain. But their effectiveness varies depending on the company’s data and the understanding of the business need.

Optimizing ML Effectiveness

Companies can set themselves up for success by making the right choices for supply chain solutions and aligning around specific business issues. Here are a few tips:

  • Deploy solutions with time to value in mind
    Many companies invest in a collection of point solutions with long implementation times, require a lot of human intervention, and are disappointed by the results at the end of the deployment. This approach limits ML capabilities as it relies on isolated data often stored in Excel spreadsheets.

  • Focus on specific business needs
    To capitalize on the true potential of ML usage in the supply chain, you need to understand the issue or opportunity you wish to address. What are the key supply chain processes that need immediate intervention? What results are needed now, not next year? What is the current maturity of your process? What data do you have available? Then, you can evaluate what’s the best fit for that specific problem and automate areas that can deliver the most impact.

  • Partner with trusted experts
    One of the most important factors to a successful ML strategy is teaming with the right partner. Look for experts with experience in similar industries and environments. They can guide the process using best practices and establish performance goals based on similar engagements.

The Limitations of Machine Learning

Machine learning uses computer algorithms to improve processes by studying performance and data automatically. ML is only as effective as the data it can access. Its value and impact improve as it discovers relationships—the more observations included in the data, the better the results.

An ML solution doesn’t use the same reasoning as humans and is not trying to imitate human behavior. Sometimes ML makes suggestions that a human would consider silly. But, it’s not the fault of ML. Usually, the issue is that it has not been exposed to enough data, or the data had errors. On the other hand, ML can uncover relationships that a human wouldn’t notice. That’s why ML supply chains work best when there is human oversight of the outcomes.

Two Supply Chain Scenarios Where ML May Falter

Machine Learning is not infallible. There are many areas of the supply chain where it dramatically improves performance, but here are a few exceptions:

  • Demand sensing capabilities
    ML is a predictive tool, but it is not always the best forecasting tool. It likely won’t pick up disruptions in the moment. ML algorithms rely on historical data and will identify patterns, learn from the past, and make future predictions. However, if the near future looks vastly different from the past, ML cannot provide accurate forecasts.

  • Multiple trade-offs and business constraints
    Due to the number of permutations and combinations involved in supply chain decisions, it is tempting to rely heavily on ML. But, be cautious. It doesn’t do conventional optimization, so you cannot trust it entirely. ML is good at understating patterns, making predictions, and sorting data. It can also optimize based on fixed rules. But supply chain networks don’t have all the rules defined.

So, when reviewing recommendations such as optimal product flow, order minimums, inventory investment priorities, product portfolio optimization, and more, it’s could be better to use other analytics.

ML Experts Can Deliver Value, Faster

There is value in leveraging ML to solve specific supply chain challenges. With the GAINS P3 methodology, we can identify when to use AI and ML to address a challenge and when an alternative optimization method is best. GAINS has experience applying the best mathematical methods (AI, ML, optimization, heuristics, etc.) and understanding your business needs. This methodology is why our customers achieve results in months, not years.

The bottom line is to know where ML can automate decisions and where it can’t. If you need help identifying a strategy, contact GAINS. We are machine learning supply chain experts that can help you automate and optimize with confidence.