Data-driven Demand Planning: Managing Disruption in CPG with ML and Demand Sensing

Even in normal times, demand volatility is a constant challenge for consumer packaged goods (CPG) supply chain management. But these are not normal times. The COVID-19 pandemic has upped the ante, completely disrupting supply chain planning.

As unforeseen external factors trigger extreme consumer demand changes, multiple supply chain components in multiple locations are affected, and online channel growth has skyrocketed. Traditional planning and forecasting methods simply are not up to the task of handling this complexity.

Supply chain teams that are struggling to keep up can attest to the fact that sophisticated tools are needed to get supply chains back on track in the new normal. Advanced machine learning (ML) tools can work alongside existing systems to help ease the burden of creating operational forecasts in a volatile business environment.

At AWS, we are at the forefront of using artificial intelligence (AI) and ML, offering the broadest and deepest set of ML services and supporting cloud infrastructure. AWS provides the technology underpinning Amazon.com, driving the automation that allows Amazon to better understand customer needs and respond by exceeding their expectations.

Machine Learning: The Foundation of Forecasting in CPG

Forecasting helps CPG supply chain managers anticipate demand and take timely steps to react appropriately. Traditionally, forecasting has been based on historical information (time series data), with the idea that the past can predict the future. But now, we are in a new normal that past patterns cannot accurately predict.

ML is a vital tool that can help CPG supply chain managers better plan for, and respond to, demand disruption—in the current moment and beyond. ML is an application of AI that allows systems to automatically learn and improve. It uses algorithms to discover patterns in data, then uses these patterns to construct mathematical models that create predictions for future data. The system continues to learn and adjust actions automatically without human intervention, improving its accuracy over time.

In supply chains, ML can help organizations generate more complete and accurate predictions. ML technology helps to eliminate the guesses and errors that people can introduce at each point in the supply chain, allows more variables to be integrated, and creates a more accurate forecast that is better trusted by the people using it. This dramatically improves overall planning, leading to better decisions and greater support for customers.

ML is key to Amazon’s ability to create daily forecasts for over a half-billion products at a post code level around the globe. ML also allows Amazon to correlate consumer buying behaviour for a group of products and form buying relationships between SKUs. Using ML, Amazon can predict that when a customer buys a particular product, they’re also likely to buy another specific product in the next week, even if not at the same time. A spike in demand for mouthwash, for example, might be related to a spike in demand for toothbrushes.

ML can make forecasts more complete and accurate by providing the capability to analyze hundreds of thousands of combinations—for instance, evaluating a large number of different models to see which forecast has the best results.

And since ML is a feedback loop and not static, it empowers you to continually tune your supply chain. That’s a sharp contrast to the prevailing manual method of setting business rules and keeping them in place until there’s a problem, which means you’re failing to optimize along the way.

Demand Sensing is Critical for Advanced Forecasting

ML on its own provides a foundation for long-term forecasting, but implementing demand sensing technology is now a critical capability to enable advanced demand planning capabilities. Demand sensing adds a new dimension of sophistication to identify trends and changes in demand patterns sooner from a diverse set of internal and external influences.

Demand sensing uses ML to account for the impact of various internal and external factors. These factors can include anything from weather patterns to socioeconomics, competitor activity, geographic data, internal sales data, social media, point-of-sale data, and more. Demand sensing gathers the data from these sources, known as demand signals, and integrates it into planning to help predict and manage demand volatility. This is especially valuable now, as extreme peaks and valleys have disrupted many businesses’ planning process.

As the technology that powers demand sensing becomes more accessible, economical, and easy to use, visionary companies are exploring how to incorporate it into their supply chain planning. This allows them to pick up on changes in demand sooner, from sources closer to the end consumer, and react in time.

For example, a CPG company that uses customer promotions and/or competitor activity in its demand sensing application has more awareness of the price elasticity of demand and can forecast short-term changes in demand for specific products within a certain market. This allows the company to not only improve forecast accuracy but also respond better to volatile demand by prioritizing inbound and outbound flows of these products to improve service levels.

We believe that capturing the right demand signals is one of the most effective steps CPG companies can take to make supply chain planning more accurate and efficient. When companies miss the signals that create demand, they are missing crucial information. Capturing those signals with demand sensing allows organizations to analyze short-term fluctuations, adapt longer-term forecasts as needed, and take appropriate action.

Demand Sensing in Action

Promotions and price elasticity of demand—including marketing promotions, price discounts, and price matching due to competitor activity—have become important elements in short-term forecast changes in the CPG and retail world. However, these promotions are often disconnected from the supply chain planning process. Companies can use demand sensing and ML to improve inputs to the planning process and ensure that supply chain managers are factoring in these demand-affecting elements.

A CPG distributor in Europe used ML to analyze historical promotion performance data to generate a model to improve promotion planning and predicting short-term changes to forecast. For future promotions, this model was used to generate a more accurate promotion plan by analyzing various inputs, such as promotion price, pricing dynamics, marketing campaign/medium, competitor activity, and demographics/geography/stores. However, with demand sensing, the distributor was also able to predict changes in factors impacting the model as promotions were started to further enhance forecast accuracy, minimize lost sales, and improve promotion performance.

Consumer behavior is another area that can provide valuable demand signals. Online companies with ecommerce channels can track customer journeys and product searches on their websites. If a particular product reaches a certain threshold in terms of numbers of searches or new clicks on the product page—even if the customer isn’t always converting the search to a purchase—a new order could be triggered without waiting for the normal ordering cycle. This allows companies to predict demand accurately and minimize lost sales without creating significant inventory risk.

Since spring 2019, online channels have contributed to nearly 70% of overall CPG growth. As online channels for CPG companies grow, this will become an even more important element for CPG companies in responding to demand.

Weather is another example of a short-term variable that impacts demand for many CPG businesses and can be included in demand sensing to mine valuable insights. For example, a CPG company that sells ice cream products may already know that demand for their products rises with summer heatwaves. However, using demand sensing, one such company identified that if winter temperatures warm by two degrees Celsius, people buy 30% more ice cream than they did the day before—a new insight that would affect forecasts and supply.

In another example, a CPG PPE company looking to get ahead of demand fluctuations may incorporate information on current and predicted pandemic outbreaks to make decisions on where products will be most needed.

Using demand sensing to pick up short-term demand signals doesn’t mean that companies need to change their usual planning cadence, but rather that the short-term data can be fed into the regular process for improved response through inventory, supply, and distribution planning.

Time for a Change

The events of the past year have caused major disruption in the CPG supply chain world. But this disruption also presents an opportunity to reinvent supply chain processes. Combining ML with demand sensing in the cloud is fundamentally transforming the forecasting landscape. CPG firms can cost-effectively forecast unique product groups using unique statistical models, resulting in a 5-10% improvement in forecast accuracy.

Using AI/ML to generate better forecasts and adding in demand sensing can help CPGs create a better planning system that is more efficient and accurate, allowing organizations to ride out this storm and those to come.

Learn more about AWS for CPG, or get in touch here.