Over the course of my career, I have witnessed numerous supply chains forecasting demand at an irrelevant level of aggregation. This is why I have created a simple approach to find the right model for demand forecasting.
Here is my 4-dimensional forecasting framework that allows you to set up a personalized demand forecasting process for your own supply chain.
I myself use this demand forecasting framework at the start of any forecasting project.
Demand Forecasting — The 4-Dimensions Forecasting Framework
Supply chains are living organisms that make dozens of decisions every day. To make the best possible decisions, you and your company need the right information. Most decisions result from demand forecasting. It is therefore an indispensable piece of information that supports your decision-making and helps your supply chain to optimize its service levels, organizational plans, reduce wastage and overall costs.
Because the forecasting-insights trigger specific operations, it has to be created at the right level of aggregation, tracked with the necessary metrics, and supported by an efficient review process.
In summary, an accurate demand forecasting in itself is not enough. It must be useful.
The forecasting process should be built on four dimensions:
I will call it the 4-dimension forecasting framework.
Demand Forecasting — 1. Dimension: Granularity
Firstly, we need to determine the appropriate geographical and material granularity for your demand forecasting. For this, we need to find the answers to these two questions:
- Geographical: Should the demand forecasting be made for each country, region, market, channel, customer segment, warehouse, store?
- Material: Should the demand forecasting be made for each product, segment, brand, value, weight, type of raw material required?
To best answer these questions, we need to reflect on the decisions your supply chain would make based on the demand forecast.
Remember, a demand forecast is only useful if it helps your supply chain to take the right action.
Let’s shed some light on this with a few examples:
We’ll assume that you need to decide which products to ship from your plant to your regional warehouses. At this point, it might be a good approach to add up demand by warehouse regions and forecast demand directly at that geographical level.
Caution: Forecasting warehouse demand based on historical orders is an insufficient practice, since logistical difficulties may have impacted historical deliveries in the past. Instead, demand forecasting should be done from the geographical region that the warehouse is intended to serve, regardless of the warehouse that actually served those orders. In other words, warehouse demand should be based on what should have been fulfilled from the warehouse if there were no restrictions.
Secondly, an explanation of a bad example:
There are many supply chains that forecast their demand per country. This is despite the fact that they have several warehouses serving different areas of the same country. In this situation, there is a clear discrepancy between the decisions to be made, and the information used to make them. The result of such a discrepancy is often a poor distribution of inventory across warehouses.
If you need to manufacture your products in different specific packaging, you should create a demand forecasting per packaging. During the forecast review, you should then examine what affects the ratio of each type of packaging: commercial events, promotions, etc.
It is recommended that you use different distribution channels if there are different warehouses or processes. These should also be forecasted separately, if necessary.
If, on the other hand, you only have one warehouse, you should ask yourself whether you really need to forecast per region or whether a single demand forecasting at global level would not be more adequate.
Demand Forecasting — 2. Dimension: Temporality
After discussing the first dimension of granularity, you need to determine the right demand forecast horizon and time aggregation.
Many supply chains are oriented towards forecasting demand 18 or 24 months in advance – however, the time of demand planners and others working on the demand forecasting is finite. You need to choose a limited time horizon and focus on it.
- Temporal aggregation: Which temporal aggregation range should be used? (daily, weekly, monthly, quarterly or annually)?
- Time horizon: How many time periods do you consider forecasting (one month, six months, two years)?
As you answer these questions, keep in mind what you want to optimize and achieve on your supply chain, and remember the lead times that are associated with these resolutions.
Let’s discuss two examples in this regard:
Your supplier needs to receive monthly purchase orders three months in advance. You should operate with monthly buckets and a horizon of 3 months (M+1/+2/+3).
You should not even consider forecasts that are above M+3. While forecasting demand — regarding goods to be delivered from your central warehouse to your local warehouses — you should concentrate on a horizon that corresponds to your internal lead time.
Models and forecast horizon. Other than machine learning models, statistical models can easily forecast demand over an infinite horizon. Aiming for long-term predictions, you may stick to the latter.
Demand forecasting — 3. Dimension: Metrics
Many professionals fail to see the issue of forecast metrics, even though choosing the right metric for a demand forecasting process is easier than you might think. It also has far-reaching effects on the resulting demand forecasts. Depending on the metric chosen, it could happen that too much importance is given to outliers (RMSE weakness) or even a biased forecast is risked (MAE weakness). For an in-depth discussion of this topic, see my article “Predicting KPIs: RMSE, MAE, MAPE biases” (insert link to article).
Below, I’ll give you some advice on choosing the right forecasting method:
- Avoid MAPE: Many professionals are still using MAPE as a demand forecasting metric. It is a matter of a highly biased indicator that leads to under-forecasting. Therefore, please avoid MAPE.
- Combine KPIs: A good compromise is determined by combining KPIs. This allows you to track accuracy and bias while avoiding the famous traps and pitfalls.
- Observe consistent bias: if you see consistent bias (over/underprediction) for a particular element, it usually means that there is something wrong with the model.
- Weighted KPIs: My recommendation is to weigh each product when calculating overall KPIs based on its profitability, cost, or impact on the entire supply chain. You are supposed to pay more focus on the SKUs that matter most. This is especially important as we want to be able to find a metric that works for your supply chain: A good score on your demand forecasting metric should match the business value.
But beyond the maths, it’s important to match forecasting KPIs to the material and time granularity that is required.
Let’s assume you would like to order goods from a foreign supplier with a 3-month lead time. In this scenario, you should calculate accuracy over a forecast horizon in months +1, +2, and +3 – or, better yet, evaluate the cumulative error over three months – rather than looking only at the accuracy achieved in month +1.
Demand forecasting — 4. Dimension: Process
Since we have discussed material and temporal aggregation, horizon, and metrics, you can install the last dimension: “process.”
This process is defined by three specific aspects:
- Stakeholders: Who is involved in the demand forecasting and is also willing to review it?
Various viewpoints must be included so that an accurate demand forecasting can be made. Different sources of information must be considered and used. All of this can only be accomplished if the review process is thorough and accurate.
- Periodicity: When is the demand forecasting reviewed?
The accuracy of your demand forecasting is optimized by frequent updating. However, don’t overdo updating, as this could also lead to chaos by overreacting to changes in demand and consuming too many resources for limited added value.
- Review process: How will the demand forecasting be reviewed?
The focus of any demand forecasting process should be a measurement of forecasted value. To improve the efficiency of your demand forecasting process, you should monitor the value added by each team member and work on it accordingly.
Demand Forecasting — Summary of the 4 Dimensions
Let’s summarise the three examples again:
- Short term forecast: We assume that you have to decide what you want to deliver to your stores, every week. An update of each forecast could take place weekly, giving a horizon of a few weeks in advance. For each store, the granularity would be one SKU. By updating the forecast weekly, there is little time left for review. Therefore, they should only be validated by a few demand planners. Black-box machine learning models should generally be preferred for this.
- Medium-term forecast: if they want to estimate what will be produced in the coming months, they need to gather information from many stakeholders, such as sales, finance, marketing, etc. Obtain. This is called the typical S&OP forecast. Such a forecast can be collected on a global level per SKU, once a month.
- Long-term forecast: Setting an annual budget is a prerequisite. This is a long-term forecast on a very aggregated level. Create a causal model where the weighting of inputs can be set and discussed, including different scenarios. Avoid machine learning models as they are a black box and difficult to forecast long term due to lack of data.
What is demand forecasting and its methods?
Demand forecasting is a methodical and scientific evaluation of the future demand for a product. There are various methods of demand forecasting that are used in relation to the purpose of the forecast, the data required, the availability of data, and the time frame within which demand is to be forecasted.
What are the types of demand forecasting?
Three basic types exist — qualitative techniques, time series analysis and projection, and causal models.
Why do we forecast demand?
Demand forecasting is so important because it helps a company set the right inventory levels, determine the right prices for its products, and identify how to expand or curtail its future operations.