Optimizing Material Requirements Planning in an Increasingly Configurable World
Evidence of product customization is everywhere, from the laptop ordering station at the local office supply store to the kitchen cabinet aisle at the home improvement warehouse. Customers expect an infinite array of choices and, more and more, easily configurable products. As a result, both the number of products available and the degree of customization possible are growing rapidly.
As an example, one German car manufacturer has a model that offers more than 70 configuration options, theoretically allowing for more than 1024 different production combinations. While the manufacturer may wonder whether it will ever again build two identical cars, customers have come to expect and enjoy rapid, reliable, and accurate delivery of even the most exotic product variants.
So configuration complexity increases to dazzling levels, even as cost pressures continue to dictate minimum levels of inventory. Balancing these conflicting goals requires combining smart forecasting of customer demand with careful supply chain planning and execution.
To compound the challenge, it’s not always easy to make a direct translation from customers’ needs (primary demand) into parts requirements for production (secondary demand). For standard, non-configurable products, ordering parts requires only a bill of materials and a sales forecast.
Yet for configurable products, each individual variant can consist of multiple components – and, to add to the complexity, there’s seldom a one-to-one relationship between a given characteristic and required materials. A specific part may, for instance, be needed only if the desired product has both characteristics A and B, but neither C nor D.
In the example of the automotive manufacturer, such Boolean expressions can be mind-bogglingly complex, with some consisting of sixty or more terms. Recognizing this complexity, how can manufacturers of configurable products accurately predict parts requirements well enough to keep customers happy and inventories low?
Several approaches to this forecasting problem have been advanced by business leaders and academics, including:
- Directly estimating the probability of each Boolean expression in the bill of material explosion
- Forecasting and planning on only a limited number of prototypic variants
- Considering parts requirements as a time series with seasonality and trends, completely ignoring primary demand forecasts
In practice, the choice of method depends on the number of configurable products, average product complexity, and volatility of demand.
When SAP consultants from Performance and Insight Optimization Services discussed demand planning with a German manufacturer of industrial equipment, it became clear that none of the approaches above would yield satisfactory results. A typical product could be described by as many as 150 characteristics. And due to the large-scale project nature of the business, primary demand fluctuates significantly from quarter to quarter. Such situations appeared to be the norm over time.
In a subsequent feasibility study, Performance and Insight Optimization Services research scientists developed a statistical model that relates historical primary demand directly to the corresponding secondary demand – thus creating an approximated bill of material explosion. The model can handle different degrees of input granularity; the more detailed the primary demand forecast, the more accurate the parts requirements prediction. Figure 1 shows how dramatically forecast error could be reduced using a subset of 15 primary characteristics.
Using this model, a company with €1 billion revenues in configurable products could save several hundred thousand euros per year in inventory holding costs. The company would also experience improved delivery reliability and higher customer satisfaction – all by improving forecasting accuracy.
To find out more about how predictive analytics and mathematical models can help to optimize your business, visit www.sap.com/services/pio.
