Machine learning in forecasting

From GEST-S482 Digital Business
Jump to navigation Jump to search


The decision for a customer to buy a product or a service is impacted by various factors. It can be the weather fluctuations, something they heard, how they feel at that particular moment, pretty much anything. The problem is that many of these things happen unexpectedly making it a real pain for business executives to make decisions. But there are technologies that can improve the accuracy of demand forecasting. A saying about forecasting is that the only thing sure with forecasting is that it will always be wrong. The goal is just to be precise enough to help achieve your business goals.

The place and role of forecasting in demand and supply planning

Demand forecasting is the estimation of a probable future demand for a product or service. Demand planning on the other hand, serves as the starting point for many other activities, such as warehousing, shipping, price forecasting, and, supply planning that aims at fulfilling the demand and requires data on the anticipated needs of customers.

Traditional statistical forecasting.

Statistical forecasting is often called historical because of the use off data from to past to predict the future. Statistical forecasting is also popular because the techniques used can be integrated with excel and existing Enterprise Resource Planning systems without requiring additional tech expertise. The most advanced systems consider seasonality, market trends and numerous methods to finetune results. Problems reside in the simplicity of the approach. By using historical data, models expect that situations that have occurred in the past will reoccur which is a wrong assumption.

Machine learning for demand planning

The idea now is to increase the accuracy of the forecasting at the cost of higher tech expertise needs. Machine Learning takes into account internal and external sources of information to make better data-driven predictions. Data can also be structured or unstructured. Internal and unstructured data come from websites, reviews, marketing campaigns and so on. Internal and structured data regroup e-commerce sales data, sales transaction and more. External and unstructured is for example the data found on social media, the likes, the comments or the internet of things. Weather, macroeconomic indicators or customer POS information is part of the external and structured data.

Machine learning uses complex mathematical algorithms to analyze huge volumes of information recognize patterns on its own, adapting them to changing conditions and spot complex relationships in large datasets. Algorithms are procedures or formulas that follow a sequence of finite operations or specified actions.

Here are some of the forecasting algorithm models:

  • Clustering analysis: Creating smaller subgroups of the data to make the analysis easier.
  • Descriptive analysis: Describes past behavior keeping it in mind to predict similar events later on to take profit off the possible opportunities.
  • Factor analysis: Understand relationships and dependencies and comprehend how the different data affect each other. Allows us to predict future development of a variable thanks to the predicted development of the related factors.
  • Time series analysis: Collect historical data to make time based predictions with various patterns that repeat themselves on precise dates.

These are the reasons why machine learning based software offer more accurate and reliable forecasts in complex scenarios. Certain companies have reported an increase of 5 to 15 percent in forecast reliability (up to 95 percent sometimes). Because of the additional complexity, the results can be hard to interpret. The maintenance needed increases the equipment and human resources required. Machine learning is not always the best solutions. Some good situations are: -Short-to-mid-term planning -volatile demand patterns -fast changing environment -new product launches The reason for this remains in the fact that lots of data is required to best forecast demand in theses cases and there is a need to consider numerous variables. Traditional forecasting performs well in mid/long-term planning, established products and stable demand.

Machine learning solutions for demand forecasting

Machine learning can help not only estimate demand but also go further and comprehend the reasons that drive customers to buy more and under what conditions. This is called predictive analytics.

This is how it works:

  • It saves up data from different sources such as Customer relationship management, point of sales, sensors, customer demand studies, social media, surveys…
  • Determine which forecasting algorithm to use
  • Building predictive models to find the different relationships between various factors
  • Monitor models to measure the results and improve prediction accuracy

This enables companies to combine all the buying decisions of their customers. However, this is as we said for machine learning in general, very costly and it needs to find the right patterns meaning the forecasts are set for at least a month. When we need to forecast demand for shorter terms, we use demand sensing to manage real time changes. This is basically just a feature to help adjust existing forecast methods. The system collects daily data from point of sales and other sources to evaluate the significance of each divergence and reevaluate the forecast. Doing this has reported a decrease of 30 to 40% of errors.

Amazon forecast

Amazon forecast is a fully managed service that use machine learning to deliver highly accurate forecasts.

In a world where companies use more and more data for various tasks such as complex financial planning, demand forecast and so on… Amazon Forecast comes right in place.

Based on the same technology used at, Amazon Forecast uses machine learning to combine time series data with additional variables to build forecasts. Amazon Forecast requires no machine learning experience to get started. You only need to provide historical data, plus any additional data that you believe may impact your forecasts. For example, the demand for a particular color of a shirt may change with the seasons and store location. This complex relationship is hard to determine on its own, but machine learning is ideally suited to recognize it. Once you provide your data, Amazon Forecast will automatically examine it, identify what is meaningful, and produce a forecasting model capable of making predictions that are up to 50% more accurate than looking at time series data alone. (Source: