By Pekka Hakkinen | Principal | Chainalytics
Demand planning is the first and perhaps most-important planning phase in any end-to-end sales and operations planning (S&OP) process. It’s also key to achieving high-quality S&OP results and minimizing the bullwhip effect throughout the supply chain.
But the demand planning phase is typically the most challenging, since it deals with a hard-to-manage uncertainty: dynamic market behavior.
Most companies carry out their demand planning phase like any other process, tailoring their approach to their industry type, business model or company-specific planning needs:
- Many businesses begin by using statistical forecasting to generate a base forecast from historical actual demand data using mathematical models and algorithms.
- This base forecast provides the underlying data series to reference during succeeding demand planning steps.
- They then feed additional demand elements and market intelligence into the planning process to produce a confirmed demand plan, which is in turn used as input to supply planning, the next phase of the S&OP process.
Unfortunately, statistical forecasting is often overlooked and rarely receives the attention required for maximum accuracy, despite the fact that its quality is key to achieving maximum confirmed demand plan accuracy.
Consider the following to get the most out of your statistical forecasting:
- Segment your portfolio. Statistical forecasting may work very well in creating a base forecast for some products. But for other products it may be totally unsuitable.
In portfolio segmentation, products are classified (based on their demand patterns) into demand segments. Segmentation is the key to determining which products follow statistically meaningful demand patterns and which do not. So it is important to conduct demand analytics to periodically segment your product portfolio based on demand behavior (you can undertake that analysis every six months, annually or at some other frequency, depending on the nature of your business).
- Improve data quality by removing outliers. To prevent exceptional or unexpected historical demand highs or lows from weakening your statistical forecast’s quality, neutralize any outliers so that the demand values fall within desired range. To identify outliers from demand history, you can apply analytics to determine the limits historical demand values should fall within.
- Educate all your demand planners. Any employee responsible for creating a demand plan and using a statistical forecast as a baseline (whether a sales rep or a dedicated demand planner) will need to understand where the statistical forecast is coming from to provide the sharpest possible projections. They’ll need to start with the base forecast, look at other demand elements (order book, quotations etc.) and add their outside market intelligence into the mix. If they do not understand the logic behind the base forecast numbers, they can hardly make use of that information in providing their best demand plan.
In other instances with clients, we have found that some businesses’ demand patterns require simply creating a statistical forecast and making it a confirmed demand plan, with no manipulation or user intervention.
Again, it is vital to have people with skills and knowledge conduct appropriate analysis at the very beginning of the S&OP process and during iterative sessions, to define the most suitable planning procedures and steer the demand planning process accordingly.
- Simplify your demand planning IT system. The tasks of a demand planning IT application are (1) providing a solution for the demand planning process needs and (2) integrating demand planning into the S&OP processes.
The only downside to using a demand planning IT system? From a statistical forecasting accuracy standpoint, the latter task may turn out to be tricky to manage: Integrated S&OP processes require common planning hierarchies to keep data consistent throughout the organization. However, real-life product demand behavior has no respect for data structures; it is derived from customers’ buying behavior.
Through analyzing their demand data, some companies may find that the best statistical planning accuracy for some of their products is achieved when planning is done at a “product family” level of the product hierarchy. For other products, a statistical forecast should be generated at the SKU level for the best outcomes. For other products, the application might indicate planning at some other level. Adding other planning hierarchies such as customer or geography will produce even more multifaceted picture.
Trying to configure a demand planning application to deal with this level of complexity may lead to overly complex setup of planning parameters. Alternatively, for the sake of IT simplicity, a planner might compromise on planning accuracy and prioritize the ease of maintenance of parameters.
There is a solution to this challenge: encapsulating the statistical forecasting task in a statistical forecasting engine that receives historical demand data from source applications; generates an accurate, uncompromised statistical forecast; maps the results against the common hierarchy; and feeds the results back to the subsequent process step.
This solution ensures the quality of the statistical forecast remains uncompromised for optimal accuracy while allowing simple configuration of planning parameters within the demand planning application.
How an Outside-In Expert Perspective Can Help
Without ongoing monitoring, statistical forecasting often devolves into an organizational routine. It’s run as a demand planning process step with little, or infrequent, attention to the accuracy of results and the need for forecasting parameter updates.
This can happen for many reasons including:
- Demand planning managers assigned to the task are too busy with daily tasks to ensure the demand planning process gets executed as scheduled.
- If a costly demand planning solution was implemented a couple of years ago, the development budget and resources may have subsequently been allocated to other areas.
- The company may not have employees with the skills or experience to analyze demand behavior and understand statistical models.
Statistical forecasting does not require a detailed understanding of the company business; however, it takes data analysis and statistical mathematics capabilities, as well as the right tools. Likewise, communication skills are essential: An analyst must be able to translate his/her findings clearly and concisely into business terms and recommendations.
Some of our clients have chosen to outsource their demand analytics and statistical forecasting function to Chainalytics, to avoid hiring high-cost employees with statistical math skills, but also to ensure they gain the value of our proven methodologies, which enable quick discovery of demand patterns and planning dimensions that give the most accurate results.
For many of our clients, statistical forecasting has become a cost of doing business. They rely on Chainalytics as a close business partner in the demand planning process. With constant access to demand data, we can maintain a company-specific forecasting model and system to revise statistical forecasts and to feed results back to the following step in the demand planning process, mapped against the client’s planning master data. This arrangement enables a quick statistical forecasting execution and allows the client to focus on tasks and decisions that require a thorough understanding of the business.
Pekka Hakkinen has more than 25 years of industry and consulting experience in supply chain planning and execution processes. His areas of responsibility have ranged from business modeling, process design and project management to IT implementation. He has conducted process improvement initiatives and implemented IT solutions in areas including manufacturing, logistics, sales and operations planning (S&OP), demand planning, and supply chain planning.
Read more about how better S&OP supports better businesses:
