Quick Start Guide to Using Machine Learning for Demand Planning
Anyone who has done demand planning knows it is extremely complex, with forecasting challenges and rapidly shifting consumer demand, often exacerbated by seasonality, new product introductions, promotions, and myriad causal factors (e.g. weather, social media). The good news is that these characteristics make demand planning a perfect fit for machine learning solutions. In fact, supply chain executives put demand forecasting with machine learning as a priority use case.
While machine learning and demand planning may go together like peanut butter and jelly, successfully harnessing this technology requires careful consideration and preparation. Gartner research shows that “three of the top five reasons why organizations haven’t adopted AI are related to an inability to articulate a roadmap.”1 We’ve compiled a few considerations to quickly get you on your way to successfully incorporating machine learning into demand planning processes.
Tips for Using Machine Learning for Demand Planning
Set Specific Business Objectives at the Start
With all the buzz about machine learning for supply chain planning, it’s tempting to want to go from zero-to-60 with machine learning. Unfortunately, the transition requires a more moderated approach to achieve success. Without baseline metrics on what you want to improve on and why, how can you be confident your strategy is working? Having a solid charter of what you want to accomplish and why is essential before charging down the machine learning path.
Gathering the necessary data to formulate an accurate comparison between previous results and those provided by machine learning is an excellent way to establish confidence in your results. And because machine learning systems get smarter over time, having a consistent method of measurement is even more important to ensure you can accurately track how outcomes and ROI are improving against established metrics.
Take a Phased Approach: Establish a Foundation and Layer on Complexity
After you’ve established your project objectives, it’s time to build a solid baseline/foundation for a successful and sustainable initiative. We’ve found that the best approach is to leverage both probabilistic forecasting and machine learning technologies, which work together seamlessly and automatically, giving users the ability to forecast at the most granular level, on different time horizons. This walk, then run approach begins with establishing an adaptive, probability-based model for demand forecasting using existing historical data, then layering in more sophisticated machine learning using external data sources. A reliable demand forecast is critical to success with advanced machine learning and yields significant benefits on its own.
Don’t Forget the Four Dimensions of Data: Volume, Granularity, Quality, Variety
It’s important to have the right amount of data to draw upon, but it’s equally important to have enough data to derive “statistical significance” from the model.
Unlike many approaches from the past where data was often aggregated to weed out noise from the model, it is the examination of that noise and its use to find correlation between it and other seemingly innocuous data elements that help to train the model and give it its power.
It’s a good idea to implement regimented data governance programs to clean, filter and maintain information quality through that data’s lifecycle.
The more different types of data sources you factor in (e.g. promotions, advertising, new product introductions, social media, weather, economic indicators, and others), the more robust and accurate the planning outcomes can be.
Operationalizing Your Machine Learning Solution is Key
Often businesses will build a machine learning solution to tackle a one-off business challenge, not considering long-term sustainability, or how they will adapt the solution as business challenges change. For sustainable business value you need to operationalize your results for continued success. Here are a few tips to help you prepare:
Choose self-adapting models: To achieve the stability and adaptability required for operational use, it’s important to use models that are self-adaptive and do not require continuous tuning by experts, otherwise changing business environments will make them unreliable. This is common with traditional demand planning processes that use multiple forecasting algorithms that are assigned to each item/location according to its demand behavior. The forecast generated by these algorithms degrades as the demand patterns evolve over time. This discrete selection and tuning of the algorithms require human skills that most businesses cannot afford.
A connected solution is key: One-off science projects create “black boxes” that only the developer understands and can support. Business users remain skeptical, and if the developer leaves the company, these models are shelved or discarded altogether. These isolated machine learning projects also require continual manual work to refresh the model when business needs change. The better method uses a self-adaptive model as part of a fully integrated business solution, with models updated automatically on a frequent basis to react to changes in the business.
Get the Right People on the Team
The automation machine learning brings enables planners to do less manipulating models and spreadsheets and more value-add, strategic work. As your business changes over time, you’ll have new questions to answer, and will need to adjust your existing models so that they remain accurate and useful. It’s critical to understand the skills and resources you’ll need for success before kicking off your project.
Part of the strategic work of planners is applying domain knowledge to the process. Machine learning can only do so much; business knowledge and process expertise is required to properly tune machine learning models and evaluate results. This AI-augmented planner role is the ideal symbiosis between human and machines: The system gets smarter over time by factoring in human input and the humans get smarter by learning from the success rate of the probability forecasts. This frees up planners to focus on service, work on strategic projects and add their business insights to the system.
Machine Learning Case Example: How Lennox Tackles Company Expansion and Demand Complexity
Lennox Residential Heating and Cooling faced the challenge of managing an ambitious North American distribution network enlargement while simultaneously transitioning to a hub-and-spoke model with 55 shipping and 161 selling locations. The company wanted to improve service levels and optimize inventories to reallocate working capital and balance inventory in the changing network. But the supply chain environment was daunting, with a multi-echelon distribution network about to grow by 250%,
450,000 SKU-Locations, many slow movers, and new product introductions.
Lennox implemented a transformational supply chain planning solution to dynamically rationalize the inventory mix and create an operational plan that sets inventory stocking targets and balances service levels with inventory cost. Lennox uses machine learning to reliably model highly variable seasonal demand patterns. It sifts through hundreds of thousands of SKU-Locations to identify “clusters” of those with similar seasonality profiles. These enhanced seasonality clusters substantially increase peak period forecast accuracy.
- Improved service levels by 16%
- Increased inventory turns by 25%
- Supported significant increases in sales and market share growth