Demand Sensing Improves Short-Term Demand Forecasting at Arla Foods

Among leading SAP supply chain customers – companies achieving significantly better supply chain business results than the competition – 41% currently use advanced demand planning, and another 18% plan to implement it in the next year.

One aspect of advanced demand planning, known as demand sensing, integrates real-time data such as open sales orders, promotions, social media sentiment, or even weather forecasts into existing time-series statistical modeling, adjusts the mid-to-long term forecast based on patterns, and breaks the forecast down to the daily level to support immediate decisions.

At the SAPinsider SCM event in Orlando, Florida, November 6-8, 2019,  Implement Consulting Group (ICG) and Arla Foods co-presented a session on how the two companies worked together to successfully deploy demand sensing with SAP Integrated Business Planning to improve forecast accuracy.

This research brief shares how SAP Integrated Business Planning demand sensing can lead to improved forecast accuracy and therefore higher order fulfillment rates and reduced inventory costs with more inventory turns.

Products with Short Shelf Life Need More Granular, Flexible Forecasts

Fast-Moving Consumer Goods (FMCG) companies often need to make shipping and distribution decisions in a short-term horizon due to spikes in demand and the short-shelf lives of some products. Various factors including weather, promotional campaigns, and competitor activity can and should influence these decisions. More traditional time-series statistical models for demand planning can struggle to incorporate these types of real-time information.

Demand planning from SAP Advanced Planning and Optimization and SAP Integrated Business Planning can provide useful forecasts for strategic and tactical planning, master production planning, and sales and operations planning more than six weeks out, but without demand sensing, it does not easily split into daily forecasts or react to customer orders or promotions in stores. This makes it difficult to accurately schedule the size of daily deliveries.

To use demand sensing, organizations still need their mid-to-long term demand planning forecasts for tactical and strategic planning and as the primary input for demand sensing. Demand sensing refines and adjusts the demand plan for the short-term future. The traditional demand plan provides the aggregated volumes of products, and demand sensing improves delivery accuracy by supporting decisions involving getting the right mix of products to the right place at the right time.

In the case of Arla Foods, the world’s fourth largest dairy company, it originally attempted to implement SAP Integrated Business Planning demand sensing back in 2017 and integrated the solution with its existing SAP Advanced Planning and Optimization demand planning application.

As part of the initial rollout, the IT department led the initiative and failed to consult the demand managers or educate them on how it worked.

Paul Gibson, Demand Manager at Arla Foods, shared, “We knew we could benefit from more focused short-term planning, but incorrect threshold settings and a rushed rollout led to a failed initial launch.”

When Done Right, Demand Sensing Improves Short Term Forecast Accuracy and Delivery Accuracy

SAP Integrated Business Planning demand sensing uses machine learning and pattern recognition to match and balance weekly and daily forecasts to patterns in demand with inputs from historical and future sales orders, historical shipments, promotions, weather forecasts, and social media sentiment.

When Arla Foods relaunched demand sensing in January of 2019 as part of a deeper engagement with ICG, it took a month to analyze the performance of current operations. Then, it took another month to understand and educate employees on how the demand sensing engine works, identifying the tuning parameters and determining how it could improve. After that, Arla ran tests to tune parameters across product categories and markets in the UK and Denmark. It found that it could effectively apply demand sensing to approximately 80% of all products and improve their forecast accuracy.

Only after all that did it roll out the technical improvements first in Denmark, and then in the UK, onboarding its super users of demand planning and exchanging learnings. Arla Foods and ICG gamified the educational experience to help demand planners better use demand sensing with an interactive gameboard using flip cards with what-if scenarios.

Arla Foods performs weekly sensing runs every Monday morning and can update weekly volumes based on that, and it also runs daily updates when it receives orders, which generally do not change weekly volume but can change product mixes in deliveries.

As a result, the company has seen improved forecast accuracy of 5.5% in Denmark and 2.4% in the UK. Arla Foods has also seen an 8% decrease in manual planning exceptions in the UK and a 1% decrease in Denmark. It also makes the demand planners much more efficient.

“SAP Integrated Business Planning demand sensing saves us three to four hours per day, which frees me up to focus on more long-term planning and promotions,” said Gibson.

What Does This Mean for SAPinsiders?

Based on our research and the presentation from ICG and Arla Foods, the following considerations can help SAP customers improve short-term forecast and delivery accuracy:

• Identify internal and external demand drivers and bring them together in a structured and coherent manner. Like any digital tool, demand sensing only produces value if it has the correct data inputs. If not, garbage in will produce garbage out. Evaluate seasonality, effectiveness of promotions, competitor actions, social media trends, and any other information that could affect demand. Then, organize the data sources in a usable way.
• Educate demand planners before rolling out demand sensing. Educating means more than training. Demand planners should understand not only how to use the tool but why it works the way it does. Only then can they effectively set thresholds and react appropriately.
• Test demand sensing for each individual characteristic value combination (CVC) before applying it live. Demand sensing will likely not prove relevant for every single product. Arla Foods discovered that demand sensing performed worse than existing unconstrained forecasts for 20% of its products, for which it had less stable demand.
• Adjust sales history as an input based on historical promotions and exclude holiday weeks from machine learning algorithms. In its failed 2017 initial launch, Arla Foods did not consider historical promotions and public holidays when setting up the machine learning algorithms for demand sensing. This contributed to its struggles.

Following this strategic guidance should help SAP customers get the most out of SAP Integrated Business Planning demand sensing in a way that complements demand planning on either SAP Integrated Business Planning or SAP Advanced Planning and Optimization.

Make sure to download and read the our full report, “Transforming the Intelligent Supply chain,” available for free for SAP customers and the general population. Also, please contribute to the upcoming report, “When to Move Supply Chain Planning to the Cloud” by filling out our latest survey.