Neural network demand forecasting
Harness the power of company and market data
DATURE ENTERPRISE software uses deep fuzzy neural networks in the demand forecasting process
Modern supply chain management requires analysis of large data sets. This is a particularly important element in the process of generating forecasts. The accuracy of estimating future sales depends on how well we are able to “read” and use the data affecting demand. Often we have huge amounts of data, but due to our inability to incorporate it into traditional forecasting models, we fail to make business use of it.
Neural networks allow for the combined analysis of a large number of potential variables affecting demand. e.g. promotions, selected macroeconomic and sectoral data, meteorological, etc. They can also learn from them. Thus, they give us a chance to use them business-wise and get the benefit of much more accurate predictions of future sales.
Advantages of a neural network
Among the advantages of using neural networks in the forecasting process are:
- the ability to learn and generalize the knowledge acquired in the learning process,
- the ability to realize associative (associative memory - similar to how memory works in humans) memory
- the ability to model nonlinear relationships,
- the ability to solve problems without knowing the analytical relationship between inputs and expected outputs,
- adaptability to changing conditions in which the forecast is made,
The use of Dature Enterprise in demand forecasting, moreover, allows you to estimate the importance of the data used in the model from the point of view of the forecasting results. This allows you to know the “validity” of the characteristics of these data and use them in product management processes.
Machine learning
An example of a machine learning process in a neural network starts with loading selected data describing demand. Typically, we have a sales history - the question to be answered is whether this data describes demand well? For example, it is worth making our sales history more realistic by filtering out sales with out-of-stock or incomplete inventory. If we know or suspect a cause-and-effect relationship between our activities (e.g. price changes, promotions), public holidays, events, ordinary seasonality then we add such data. In this way we get variables that will support our neural network demand forecasting model.
The final forecast is made based on the best model, parameters or variables.
Such a process gives us great assurance that we will get the best possible forecast based on our data and knowledge of the business process
Inventory optimization
using a mathematical objective function
DATURE ENTERPRISE software uses mathematical constrained objective function models for inventory optimization.
Simple optimization models are usually concerned with determining the optimal order quantity and are based on the concept of so-called economic order quantity (EOQ). These models can take into account the cost of out-of-stock, but are in this regard based on the assumption of a fixed, arbitrarily adopted level of service (this can be derived, for example, from internal norms based on ABC classification). This “one-dimensional” approach is therefore suitable for rational management of cyclical inventory, especially for single assortment items. In the case of inventory control of multiple assortment items, this approach is implemented in terms of value (so-called economic order value).
In many cases, however, such an approach is insufficient, as the required (optimal) level of service may depend on the cost of possible out-of-stock and its relationship to the cost of maintaining the inventory. This leads to the necessity of building complex models, in which the independent (optimized) variables include both the volume of deliveries (more generally, the frequency or number of deliveries during the assumed period) and the level of service (or the corresponding safety factor). These quantities interact in a complex way with both the cost of replenishment and the cost of maintaining and running out of stock. This is illustrated in the figure below.
Optimization vs. objective function: cost minimization / margin maximization
Optimization of goods flows in supply chains, in addition to the need to define the objective function, requires consideration of a number of constraints. These can vary in nature and concern various issues, such as the permissible availability of warehouse space or the budget that finances inventory,. In cases where one of the factors affecting the value of the objective function is the level of inventory, the level of service (the level of availability of goods dispensed from inventory) will be a constraint.
Optimization of both links of the supply chain (supplier-customer)
Operating in the supply chain, we typically function in a customer-supplier context, collectively generating value for end customers. Optimization of average inventory levels should therefore be considered from the perspective of both the customer and the supplier.
In the Dature Enterprise application, this is possible thanks to the developed algorithms of the model that allows solving two optimization tasks:
- minimize total logistics costs
- maximizing the total margin while taking into account:
- a set level of service - at the level of the recipient
- constraints from storage capacity, transportation means, etc.
- variable replenishment costs (e.g., transportation) as a function of delivery volume
- the cost of stock shortages
- etc.
Demand forecasting is an important issue from the standpoint of inventory management in supply chains. However, an equally important factor affecting the reliability of supply processes, service levels, and consequently the level of safety stock and associated costs, is the timing of the replenishment cycle. The replenishment cycle must be carefully defined by identifying its components and their interrelationships, covering not only the periods that are the responsibility of the Supplier, but also those that lie with the Recipient. The high complexity of the replenishment cycle means that the optimization of inventory levels with respect to replenishment time requires consideration of a number of economic factors. This means identifying these factors in detail.
Optimization of average inventory levels should take into account the total costs of holding and moving inventory throughout the supply chain, the effects of out-of-stock, while maintaining the identified constraints (e.g., storage capacity at each link).
The approach available in Dature allows the implementation of a complex optimization process on the following sequence:
- definition of baseline service level indicators and their value from the Recipient to its customers
- determination of necessary service levels for particular (all, or selected, key) assortment items managed by the Recipient
- definition of service level indicators and their value on the part of the Supplier to the Recipient
- determination of necessary service levels for individual (all or selected) assortment items managed by the Supplier
- determination of optimal inventory locations, taking into account various organizational arrangements and the total cost of maintaining inventory and moving goods.