Cleverly dealing with material shortages with AI
"Our material is in short supply." At the beginning of 2021, four out of five German companies this statement. Around a year later, the Ukraine crisis is turning this picture around. In spring 2022, four out of five German companies complained of a significant shortage of materials. But there are clever ways to improve the cash-to-cash cycle with the help of data-supported automated forecasts.
In addition to the European war, the shortage of materials in recent years is also the result of drastic factors such as the coronavirus crisis, temporary economic upturns or the increasing international interdependence of supply chains and the resulting dependencies. The economic index, which is currently measured on a monthly basis, fluctuates to a greater or lesser extent. Unfortunately, the overall picture is not a positive one. German companies are registering a growing backlog of orders that they are unable to fulfill. A key trigger for this is the lack of materials.
Modern AI-supported solutions improve sales forecasts
Diese Entwicklung zeigt eindeutig, dass herkömmliche Methoden zur Vorhersage des Materialbedarfs nicht mehr zeitgemäß sind. Einige Unternehmen setzen auf moderne <a href="https://www.taod.de/services/artificial-intelligence-consulting" data-webtrackingID="blog_content_link" >KI-gestützte Lösungen</a> und Intelligent Automation zur zuverlässigen und kostengünstigen Vorhersage ihres Materialbedarfs und der Optimierung ihrer Bestellmengen und Bestellzeitpunkte. Andere drohen abgehängt zu werden. Manche Betriebe begegnen dem Mangel mit einem Ausbau der Lagermengen. Die erhöhte Kapitalbindung in Kombination mit steigenden Zinsen vervielfachen allerdings die Kosten und schränken somit die Flexibilität dieser Unternehmen langfristig ein.
Betriebe mit einer <a href="https://www.taod.de/demand-forecasting-mit-ai" data-webtrackingID="blog_content_link" >automatisierten Nachfragevorhersage</a> können eine Übereinstimmung von 95 % zwischen der tatsächlichen Nachfrage und der Vorhersage erreichen. Auf dieser Basis können sie den tatsächlichen Bedarf an Rohstoffen und Vormaterialien präzise bestimmen und ihre Lagerstrategie auch in unsicheren Zeiten optimieren. Ein Beispiel hierfür ist das Unternehmen HoneyWell HSP, ein Hersteller für verschiedene Sicherheitsprodukte. HSP stand vor umfangreichen Herausforderungen in der Bedarfsprognose, wie schwankende Nachfrage, Erweiterung und Anpassung des Produktportfolios, Verwendung mehrerer Systeme zur Nachfragevorhersage, hoher manueller Prognoseaufwand und fehleranfällige, langsame Prozesse.
The company then decided to use an AI-supported forecasting system. Initially, a test phase with parallel operation was set up to continuously develop the model and check whether the predictions were actually coming true. Within a few months, a continuous improvement could be seen. The deviation of forecasts from actual demand fell from 25% in the first quarter to 13% in the third quarter. Adherence to delivery dates was increased by 20 % and stock turnover by 16 %.
{{whitepaper}}
Increase in fulfilled orders from 80 % to 95
Brooks, a manufacturer of sports shoes, was also facing immense challenges. The sales growth of new product lines fluctuated greatly and was between 0% and 50%. The Dutch company usually makes the decision to adjust production capacity 18 months before the start of production of new product lines. Brooks recorded a significant increase in orders with immediate delivery. Furthermore, the sell-through cannot be calculated as the specialist stores have little or no means of transmitting their sales data to Brooks.
The management decided to set up a data infrastructure and develop a forecasting model. The model's forecasting accuracy was tested by comparing it with real demand volumes. Reports were automatically generated for various positions in the company with different levels of detail. Experienced employees were given input options to incorporate their years of experience into the AI solution and thus obtain the best possible forecast.
The accuracy of the forecasts was 40% better than that of manual planning. Whereas previously 20% of orders could not be processed, with the system it was only 5%, despite greater fluctuations in the market. Overproduction was reduced by 60%. Overall, Brooks has achieved higher margins and lower costs by switching to automated demand forecasting.
Components of an AI-supported prediction of demand
In order for demand forecasting to offer companies added value, it requires three components.
1. reporting
Current demand forecasts and key figures on the reliability of the forecast, presented in tools such as Power BI or Tableau, provide access to the most important information. All tools can be used by all relevant employees after a brief introduction. The complicated computational work is performed by the algorithm in the background.
2. algorithm
There are countless possibilities for the computational part, which are selected by AI experts to suit the individual case. The algorithm for a demand forecast should be individually tailored and trained on the basis of many years of business experience. Normally, the more extensive and complex the data, the longer it takes to train and optimize an algorithm. However, with the right data basis, high speed can be achieved without compromising on quality.
3. data basis
An algorithm is based on necessary and additional data. Necessary data is demand data from the past and known central influencing factors. Additional data are influencing factors that are important but not central.
With a clear focus on the necessary data, an investment in an AI-supported system can be profitable particularly quickly. In a proof-of-concept project, experienced demand forecasting experts very quickly begin to analyze what data is actually needed to make reliable predictions. This lowers costs and reduces the time frame until the first forecasting solution can be used productively.
Introducing automatic forecasting - but how?
The probability of success of planned automation projects can be determined very well using a proof of concept. As a rule, a test project for demand forecasting takes around two months. During this time, the AI model is trained and tested in parallel with regular operations. If the project is rolled out over a longer period of time, the costs, which are in the mid to upper five-digit range, are usually amortized within nine months. A first step towards professionalized automation is usually the selection of an experienced service provider who can incorporate their experience in intelligent automation into consultative discussions and workshops in order to design an overall package tailored to the company. Feel free to ask us!
.webp)
