Despite machine learning going mainstream, the technology still mystifies many supply chain practitioners. Here’s how to apply machine learning to solve business problems and create value.
Start with clear objectives
It’s essential to develop a clear charter of what you want to accomplish and why. Without baseline metrics on what you want to improve on and why, there’s no way to judge whether your strategy is working. By gathering the necessary data, you’ll be able to accurately compare previous results with those enhanced by machine learning. And, since machine learning systems get smarter over time, you can potentially measure growing ROI over time and strengthen your business cases.
Walk, then run
Next, plan to build a solid baseline for sustainable success. Combine machine learning technologies with a probabilistic forecasting approach. These work together to allow planners to forecast at a granular level on different time horizons. Start by establishing an adaptive, probabilistic model for demand forecasting using existing historical data (walk). When this is established, layer in more sophisticated machine learning using external data sources (run). Machine learning depends on having a reliable demand forecast and yields significant benefits.
Consider multiple data dimensions
Volume. It’s important to have enough data to draw upon and to derive statistical significance from your model.
Granularity. Unlike old approaches where data was often aggregated to filter out noise from the model, machine learning analyzes that noise to find correlations to train the model and make it more powerful.
Quality. Machine learning projects should include governance programs to clean, filter and maintain information quality through the data lifecycle. That’s because machine learning is unable to assess data reliability.
Variety. The more different types of data sources to factor in (e.g., promotions, advertising, new product introductions, social media, weather, economic indicators and others), the more robust and accurate your planning outcomes will be. It’s important not to also factor in your traditional “small data” sources; every business has small data related to historic demand readily available.
Decide how to operationalize
It can be tempting to want to build a machine learning solution to tackle a one-off business challenge and get a quick win. One-off projects also create “black boxes” that only the developer understands and can support. These can often lead to mistrust among business stakeholders. To gain sustainable business value and get the best returns from all your groundwork, you need to operationalize your results. That’s why it’s important to apply adaptive models that don't require continuous manual tuning; otherwise changing business conditions will render models unreliable.
The top use cases for machine learning in supply chain planning
Demand planning processes like demand forecasting, sensing and shaping are prime candidates for applying machine learning. And, for good reason -- increasing forecasting complexity and rapidly shifting consumer demand are often exacerbated by myriad factors like seasonality, new product introductions, promotions and other causal factors (e.g., weather, social media), making demand planning extremely complex. These are some of the top supply chain planning machine learning use cases.
Seasonality. Clustering and classification of multiple seasonality patterns (day-in-week, week-in-month, month-in-year).
Promotion management. Clustering of past promotions, classification of new promotions based on attributes and uplift calculation.
New product introduction. Clustering of past launch profiles, classification of new items based on their attributes and regression for baseline forecast generation.
POS demand sensing. Advanced techniques to improve sell-in forecast using sell-out demand data.
External demand causals. Weather, social media, Internet of Things (IoT), market trends, indicators and other external data.
Product lifecycle management. Algorithms weigh up attributes and sales of similar items to estimate the shape and duration of the product life cycle.
People play the most important part
Like all things, machine learning has its limitations. That’s why employees’ business knowledge and process expertise play an important role in tuning the machine learning models into results. With artificial intelligence (AI)-augmented planning, systems get smarter over time by factoring in human input. The beauty is that the people get smarter in turn by learning from the success rate of the probability forecasts. This frees up planners to focus on more “human” tasks like improving service to customers and the business, working on strategic projects and adding business insights to the system.
In a primer on developing future supply chain professionals in the digital age, Gartner identified business acumen, adaptability, political savvy and the ability to collaborate as keys to improving “digital dexterity”. This reinforces how important it is for digital supply chain organizations to focus on the human side of supply chain planning as more of demand forecasting is automated through machine learning.
We’re only really at the start of enhancing supply chain planning using machine learning. Data-driven decision-making in supply chain planning is on track to be more automated, visual, transparent and intuitive for a wider range of business users. There are many exciting innovations already in development. That’s why supply chain professionals who develop more human skills like negotiating, communicating business priorities and simplifying complex data will be increasingly valuable to their organizations.
But, don’t wait to get started with machine learning in supply chain because companies have been able to substantially improve service levels while cutting their supply chain costs, waste and also the stress levels of their planners. Just be sure to invest the time in preparation and keep an eye on metrics.