Establishing a demand prediction tool that helps you control product inventory at your company

The Analytical Approach – Pairing Prediction and Optimisation

One of the greatest challenges for companies from different economic sectors is controlling the inventory levels of their products, both during the final and intermediate stages of the supply chain. The bigger the added value of the products, the greater the cost of the opportunity to maintain a high inventory level to ensure that there are no shortages along the supply chain. However, in cases with a high product turnover, significantly reducing inventory levels makes the operation extremely vulnerable to shortage. In this scenario, what is highly recommended is to use an analytical approach, one based on a precise estimate of demand and an optimisation model to define the best moments to introduce and remove a specific product along the supply chain. Accordingly, it becomes a more efficient way of minimising operation, logistics, and financial costs, while controlling the trade-off between service level, logistics costs, and financial costs.

It is essential that stakeholders are involved in the discussion, construction, and validation of the model from the very start of the problem modelling process.

The implementation of analytical modelling for the management of high added value inventory should pay close attention to operational details to ensure that the results obtained during simulations are in fact reached. It is essential that the specific characteristics of the operation are considered from the very start of the problem modelling, and that stakeholders are involved in the model discussion, construction, and validation process. Accordingly, the details of the operation will be translated in the mathematical modelling, including the treatment of special cases to guarantee that all stakeholders will have full trust in the model for decision making.

In a supply chain’s inventory management analysis, there are two factors susceptible to a high degree of variability: the demand for products and time necessary to ensure that these products are made available.
Using a methodology with a technical base offers a series of benefits for the operation through the following:

  • Guaranteeing the supply of products with a service level suited to the operation
  • Balancing the total cost of the supply chain, seeking global optimum scenarios
  • Effectively planning the movement of inventory, adding visibility to product exits and entries along all links of the supply chain

Estimated Demand – The First Step Toward Good Inventory Planning

The construction of a model starts with an accurate estimate of demand. Being able to precisely estimate the inventory flow, minimising errors and identifying variables that truly impact variations in the demand for a product along the supply chain is the first step for ensuring that the volumes during each phase of the value chain are optimised.
In order to develop a demand estimate model, it is first necessary to understand the nature of the operation to evaluate the main factors that influence its variation over time. After achieving this understanding, it is possible to assess which methods best suit the specific characteristics of the operation.
The possible options can be separated into two categories: time methods and causal methods.

Learn more: Demand Forecasting Methods

Classical Decomposition: Statistical method that involves decomposing a time series into trend, seasonality and cycle components to project these values in the future.

Exponential Smoothing: Similarly to classical decomposition, this method involves decomposing the series into trend, seasonality, and cycle components to project future values. However, it is possible to differentiate the relevance of more recent data compared to older data – by given more (or less) weight to more recent observations.

Multiple Regression: Causal prediction method that seeks to explain the behaviour of the variable of interest by other explanatory variables that are related to the explained variable. This is one of the simplest methods of the machine learning technique

Neural Networks: Machine learning technique that uses different variables to create layers of relationships between the same, simulating the network of neurons in a brain and their connections

Time methods are based on analysing demand data chronologically and applying this to predict future values. The following methods are classified in this category, for example: Classic Decomposition and Exponential Smoothing.

Casual methods are based on the use of variables that explain demand, with the estimate constructed using the projection of these variables. Multiple regression and neural networks are examples of causal methods. The advantage of this type of method is the possibility of including variables to explain seasonal differences, such as the day of the week, business day of the month, and week of the month, together with external factors like inflation and holidays. The main difficulty using causal methods is the need to predict future values for these variables.

Based on the demand prediction methods cited above, it is possible to create a tool that tests each method, determine which is the most accurate, and then apply this method to estimate demand. This tool is applicable to a range of different businesses, operations, and industries, such as the prediction of money inflow and outflow at bank branches and ATM machines, products and support services offered by back office areas, and the sale of products by a gift retail company.

Managing Stock Levels – The Trade-off Between Overstock and Stockout Risk

After obtaining an estimate of demand, it is necessary to define which strategy will be adopted for inventory planning. Considering inventory of high added value, it is crucial to have a full overview of costs and factors that impact the supply chain, such as, for example:

  • Product transport cost
  • Opportunity cost associated with the inventory volume
  • Product receipt and shipping cost for each transport type
  • Security risks associated with product transport
  • Security risks associated with maintaining a very high inventory level
  • Operation Service Level

Once the stock management model is defined, it should be implemented with a great deal of attention to the operation details, allowing the team involved in the daily work to feel like they are part of the new model. The user needs to recognise the final value of the work and guarantee an alignment between all stakeholders, preventing conflict of interest between the areas impacted by the new model.

Because it is a process that has the potential to involve significant changes in the operation, change management is essential. The transition to a new decision-making model is a critical process, interfering in the routine of areas, with multiple employees, that are impacted by the inventory levels should have indepth training to confidently operate the newly implemented model.

Factors Leading to Successful Implementation

With the change in company mindset caused by adjustments to the inventory management policy based on mathematical modelling that minimises costs in a global sense, certain phases of the supply chain may end up costing more. If the company chooses a model that reduces the inventory volume, there may be an increased cost with transport, but this will be compensated by a reduced opportunity cost, impacting the reduction in the overall operation cost. Three essential points must be followed to ensure that this transition in the operation mentality occurs successfully: alignment of goals, team training, and model accuracy.

  • Alignment of Goals – The guarantee that everyone will be working toward the same objectives: The new inventory management model will not reach its full potential if there is a misalignment between the areas impacted by the operation. To mitigate this risk, the operational goals should be defined in agreement with the inventory management model, to prevent, for example, if one stakeholder’s goal is to minimise the logistics costs while the new model increases the logistic cost to generate a larger return in terms of opportunity cost.
  • Team Training – Guarantee that the Inventory Planning area is aligned with the new policies: The change of mindset related to the inventory policy will raise questions from different areas of the company. For example, updating the policy may work with lower security levels, leading to more occurrences of system disruptions. To ensure that this idea is disseminated, it is necessary for the Inventory Planning key stakeholders, who are responsible for this change, to be fully trained to respond to questions and transmit a positive vision of the model to the other stakeholders impacted by the change.
  • Model Accuracy – Guaranteed effectiveness for those situations commonly faced during the company’s inventory management. When implementing a new inventory planning model, specific characteristics will appear that may not have been mapped when developing the mathematical modelling that resolved the problem. It is essential that these situations are analysed and the model be adapted to fit the specific characteristics of the operation. If not, important focus points may start using exceptions as the rule and discredit mathematical modelling. That is why it is important to pass on knowledge and develop skills among members of the team responsible for the operation and maintenance of the model, guaranteeing that adjustments and progress are implemented with a high level of responsiveness.

Observing all points above, the implementation of a robust model focused on minimising the overall cost of the operation has enormous potential to generate results, primarily in scenarios marked by an increase in interest rates and the consequent growth in capital cost associated with the company’s inventory. The capacity of the operation management, paired with analytical expertise for the development and operation of a prediction and optimisation model, is key for the review of the Inventory Planning process and for ensuring that the evolution occurs organically and sustainably.

About the Authors

Marcus Sousa is a Visagio consultant specialised in projects focused on Management Model, Supply, and Analytics in the retail, financial market, metallurgical industry, and energy sectors. Marcus also serves as leader of the Visagio Research & Intelligence area, with a focus on market research and analyses.

Felipe Pena is a Visagio consultant specialised in projects focused on Budgetary Management, Process Engineering, and Analytics in the acquisition and banking sectors.