Supply chain management has become an increasingly essential part of companies’ business. Demand forecasting is at the heart of everything in supply chain optimization and there are more and more opportunities today to create better and more accurate demand forecasts through machine learning.
Change in retail trade
The retail sector is becoming more data-driven, and data has become a key factor in the trade sector. As a result of the ongoing digitalization of retail, manufacturers, wholesalers and retailers are collaborating and sharing supply and demand-side data within the framework of existing laws. Data sharing has become an increasingly important part of the retail ecosystem in the new data-driven model.
Competition in the retail sector is accelerating and has forced many companies to streamline their operations further. The values of inventories and the supply chain have long been significant factors in the profitability of companies. As a result of meeting customers’ diverse needs, supply chain management has become an increasingly essential part of companies’ business. A high level of inventory binds capital, which reduces the company’s ability to invest capital in other activities that would contribute to the company’s competitiveness. One of the goals of companies is to reduce the amount of working capital tied up in inventories.
Consumers are increasingly demanding product customizations, and due to that, a company’s inventory can consist of a wide range of different items. The capital tied up in inventory can be reduced as the company pays attention to the storage of items and inventory turnover. With the help of warehousing and inventory turnover reviews, the company can reduce item lead times, optimize order batch sizes, and optimize the entire supply chain so that production volumes match the actual need for products. A company should strive to improve its supply chain to reduce inventory levels. At the same time, it should be ensured that there is a good range in stock and products requested often will not run out.
Working capital management
The average inventory value is a key influence on the inventory turnover rate. The most accurate forecasting of demand is directly related to inventory value, and thus forecasting is another key factor influencing the inventory turnover rate. Demand forecasting aims to assess future demand for products, which is affected by, among other things, future trends and seasonal fluctuations in products. Furthermore, in many of our categories, weather and other external factors bring more uncertainty to the forecast. A company should pay attention to how it intends to forecast future sales because, ideally, it can then anticipate customer requirements correctly. In this case, a company must only store the products ordered by customers, which significantly reduces the stock levels and at the same time significantly improves the inventory turnover rate.
From a retail perspective, it is essential that the supplier delivers the goods on time and has the goods when the store needs them. Thus, underestimated inventories could lead to delivery problems and, at worst, disrupt other production if the production plan is changed on the fly in an emergency. Suppliers constantly assess demand and strive to produce a quantity of goods that meets expected demand. In fresh products especially, overproduction significantly increases loss and, with limited resources, making the wrong products reduces the capacity to make another product. Purpose of Berner “Protecting tomorrow” is based on values and indicates our strong emphasis on sustainability. We work to ensure well-being and good living for all of us – also 100 years from now and the reduction of wastage is one concrete measure.
Machine learning
Machine learning is a part of artificial intelligence that allows computers to learn from data and improve their performance in performing a particular task without being specifically programmed to do so. Machine learning is suitable for situations where the phenomenon is not understood, or modeling is too tedious for manual work, but enough data is available to teach the model.
Machine learning is usually referred to as an algorithm or model. An algorithm is, in practice, an instruction on how to do something so that the problem can be solved. The concept of an algorithm could even be compared to a food recipe. An algorithm is not a specific calculation but a method that is followed when calculating. It is a mathematical model that describes the interdependence between things, and training the model means fitting the function to the data, i.e., enabling the mathematical model to best describe the data and pattern or patterns found. The model chosen may be too simplistic and underfit the data, allowing the model only to explain the problem at a rough level. Or the model may be too complex and overfit, where it explains a situation very well, but with new data the results are inaccurate.
For model training, the data set is divided into parts, where the model is trained with one part and tested with another part of the dataset. If the chosen model is good, then it should also explain the new test data well. If the model seems to work well in the training phase, but the results are inaccurate with the new test data, it is worth considering changing the model. In addition to choosing the model, it is important to understand that utilizing artificial intelligence requires industry expertise, IT expertise and analytical skills. I often act as an interpreter between business and IT. A prerequisite for success is seamless collaboration between functions. Artificial intelligence is not a black box that is the answer to everything, but it answers some well-defined problems.
Behaviour change
Demand forecasting is at the heart of everything in supply chain optimization. Demand can be predicted manually, but it takes a considerable amount of time, and the results may still be only moderate. As a result, there are more and more opportunities today to create better and more accurate demand forecasts through machine learning when enough data on demand is available. Machine learning does the same things as humans, but much faster, on a larger scale, and potentially more accurately when subjective things are left out of the assessment.
Finally, it is good to note that utilizing machine learning in a company not only requires introducing a new system but is a matter of greater change management and changing old patterns of behaviour to support the new way of working.
The future of retail revealed: Read our thought-provoking speculations on how people will buy groceries in the future. Download the Future of Retail 2030 Vision - with insights on opportunities for retailers, trend materials and future scenarios.
About the author
Joona Honka, Data Analyst Manager, Berner Oy
I’m responsible for setting the vision and culture for the use of data in making key business decisions at the company level. As analytics lead, I’m in charge of driving the day-to-day analytical approaches and exploring solutions to problems. I am also tasked with the management and long-term prioritization of the business’ overall analytical needs and opportunities.
