Local modeling

Using uncertainty modeling to better forecast demand

The Covid-19 pandemic has caused widespread supply chain disruptions across the world: chip shortages are forcing auto and medical equipment makers to cut production, while the Suez Canal blockage and the lack of shipping containers have inflated delivery times and shipping prices. . Their effects have been exacerbated by management practices such as just-in-time manufacturing that aim to reduce redundancy in operations: with the layoffs gone the security buffers that companies’ supply chains once had.

Of course, companies understood the risks of eliminating buffers in the supply chain as they increasingly invested in sophisticated data analytics. If they could better understand the bottlenecks in their supply chains, companies would in theory be able to operate with less redundancy without incurring additional risk. But the disturbances persist.

Our research across multiple industries, including pharmaceuticals and fast-moving consumer goods, shows that the reason for this persistence is less because of software shortcomings than because of its implementation. To begin with, managers tend to base their analysis on departmental units. While sales and marketing teams can contribute important insights and data, their input is often unsolicited by operational decision makers.

Additionally, analytical solutions focus closely on the company’s own supply chain. Best practices remain case-specific, and analytics models too often remain disconnected from broader ecosystem trends. As the examples cited above illustrate, a seemingly local disruption can snowball around the world.

How can companies best avoid these pitfalls? Let’s start by looking at what data analysis entails in more detail.

What is Data Analytics?

Data-driven analytical methods can be categorized into three types:

Descriptive analysis.

These address ‘what happened’ and ‘what is happening’ questions and are rich in visual tools such as pie charts, scatter plots, histograms, statistical summary tables and correlation tables . Sporting goods chain The Gamma Store, for example, uses statistical process control charts to identify in-store customer engagement issues.

Predictive analytics.

These are advanced statistical algorithms to predict the future values ​​of the variables on which the decision-makers depend. They address the question of “what will happen in the future”. The predictions generated are typically based on observed historical data on the decision’s response to various external changes (eg, changes in interest rates or weather conditions). Retailers like Amazon rely on predictive data on customer demand to place orders with suppliers, while fast-moving consumer goods producers like Procter & Gamble and Unilever have invested in predictive analytics to to better anticipate retailer demand for their products.

Prescriptive analysis.

These support decision makers by informing them of the potential consequences of their decisions and prescribing actionable strategies to improve business performance. They are based on mathematical models that stipulate an objective function and a set of constraints to put real-world problems into an algorithmic framework. Airlines are leveraging prescriptive analytics to dynamically optimize ticket prices over time. Logistics companies, like UPS, also apply prescriptive analytics to find the most efficient delivery routes.

Businesses typically use all of these methods, and they reflect the stages of decision-making: from analyzing a situation, to predicting key performance drivers, and then to the optimization analysis that results in a decision. . The weak link in this sequence is prediction. It’s the inability of its famous predictive data analytics to accurately forecast demand and supply that has forced Amazon to destroy around 130,000 unsold or returned items every week in just one of its UK warehouses.

The reason why predictive analytics fail is in most cases related to assumptions and choices about the generation of the analyzed data. Abraham Wald’s study of post-mission aircraft in World War II provides the classic example. The research group he belonged to was trying to predict which areas of the aircraft would be targeted by enemies, and they suggested reinforcing frequently hit areas. But Wald disputed that recommendation and advised reinforcing undisturbed areas, as damaged aircraft there were more likely lost and missing from the observed data. It is by looking at how the the data was generated that military officers were able to correct the decision on which aircraft areas to reinforce.

The solution lies in an approach to analysis known as uncertainty modeling, which explicitly addresses the issue of data generation.

What does uncertainty modeling do?

Uncertainty modeling is a sophisticated statistical approach to data analysis that allows managers to identify key parameters associated with data generation in order to reduce uncertainty surrounding the predictive value of that data. In a business context, you create more information about the data in a predictive model.

To understand what is happening, imagine that you are a business-to-business business that receives an order every three weeks from a customer for one of your products. Each order must be delivered immediately, which makes the request time negligible. Now suppose the customer’s first order is for 500 units and she plans to increase that quantity by an additional 500 units for each new order, but she does not inform the company that this is her plan.

What does the company see? The customer will order 500 units in the third week, 1,000 units in the sixth week, 1,500 units in the ninth week, etc., which generates monthly demand values ​​of 500, 1,000, 1,500, 2,500 and 3,000 units for the five average 2,100 units per month. But since the actual demand data shows substantial deviations from the average, the latter is a highly uncertain forecast. This uncertainty disappears completely, however, once the company obtains the information that the customer systematically increases its purchases by 500 units with each order.

For production managers to spot this type of information, they need to look beyond purchase numbers. In most businesses, customer order information is stored in an order management system, which tracks data such as when orders are placed, delivery dates requested, and products requested in what quantities. This system is generally owned, managed and maintained by the sales department. After customer orders are fulfilled, aggregate information about completed orders is transferred to the demand processing system, usually owned by production and operations, which those responsible for these functions then analyze to forecast future demand.

The problem is that the aggregation process often results in loss of information. With uncertainty modeling, however, managers can apply key parameters identified from the order management system to feed back information to their prescriptive analyses.

Information about the rescue in Kordsa

Kordsa, the Turkish tire reinforcement supplier, gives a concrete example. The company receives large orders from its customers (tire manufacturers) but the number of orders, as well as the quantity and delivery date of each, is uncertain in each period. Previously, the company simply aggregated customer order information to calculate historical monthly demand values ​​which were then analyzed. As a result, the number of uncertain parameters went from three to one, resulting in a significant loss of information.

Using uncertainty modeling, we showed Kordsa how to avoid information loss and achieve significant performance improvements based on key performance indicators (such as inventory turnover and fill rate ). By applying advanced algorithms such as Fast Fourier Transformation, we were able to integrate into the company’s demand prediction model the key customer order parameters that we identified by studying the company’s CRM data.

To better leverage the power of uncertainty modeling, Kordsa has since created an advanced analytics team drawn from R&D, sales, production, planning and IT. Team members regularly interact with different departments to better understand and identify data and sources used in decision-making processes outside of their own functions, which can then be factored into their predictive analytics.

This type of going beyond borders should not stop at the company’s gates. It is not just the decisions of its customers and suppliers that can affect demand uncertainties — the decisions of players in adjacent industries producing complementary or substitute products can also affect demand. Getting closer to the data generated by these players can only help reduce uncertainty around the performance drivers you need to be able to predict.

. . .

Although manufacturers and retailers are investing in data analytics to improve operational efficiency and demand fulfillment, many benefits of these investments are not being realized. Information is lost as data is aggregated before transformation across silos, which amplifies the level of uncertainty surrounding predictions. By applying the mathematics of uncertainty modeling to incorporate key insights into how data is generated, data scientists can capture the effects of previously overlooked parameters that can significantly reduce the uncertainty surrounding demand forecasts. and supply.