English Articles

Applying Advanced Analytics in Forecasting

Efficient supply chain planning enables organizations to optimize the balance between supply and demand. As a critical component of demand planning, forecasting aims to capture business intelligence at various planning horizons.

By leveraging advanced analytics to improve forecast accuracy, businesses have the opportunity to realize economic benefits and operating efficiencies, while providing better visibility into financial and operational performance.

Though forecasting requirements vary by business and industry, there are several key elements that enable organizations to apply advanced analytics as an integral part of their forecasting process. Below is an advanced forecasting analytics initiatives checklist:

Outlier Detection & Management

Detect, identify, estimate, and adjust unusual and influential observations (i.e. outliers). Multiple types of outliers should be considered, including level-shifts (seasonal and non-seasonal), transitory change, and innovative outliers. Avoid common pitfalls , such as assuming outliers represent a single-period event, or treating all outliers as additive.

Demand-Shaping Factors

Identify and incorporate suitable internal and external factors based on domain knowledge and discussions with business managers. These demand-shaping factors aim to incorporate additional layers of intelligence and provide greater intuitive understanding of the forecast data.

Select Best-Fit Algorithm

Run the adjusted demand history and demand-shaping factors into a suite of advanced algorithms, such as multiple variants of exponential smoothing, seasonal-trend, local regression, and Box-Jenkins ARIMA models. Compute forecast error (either relative or absolute), depending on the nature of demand, at the appropriate lag and hierarchy by leveraging a robust backcasting mechanism.

forecasting-methods-white-background

Forecastability Segmentation

Estimate the degree of forecastability at different hierarchical levels. Segment along different dimensions such as revenue and forecastability to support decisions about processes that drive efficient demand management.

New Product Introduction Forecasts

Implement a forecasting methodology for new items with limited demand history using techniques such as identifying like items or using a forward looking forecast model.

There is no magic click that will make a mathematical forecast model work flawlessly; the forecasting process and algorithms should include the nuances of each particular business. Combining all the intricacies described above is challenging, yet necessary, to implement advanced analytics in an organization’s forecasting process.

Manuel Peralta
Consultor Supply Chain. Consultor de estrategia y operaciones con más de 12 años de experencia liderando proyectos a nivel mundial, particularmente en las industrias de consumo masivo, retail y manufactura.

5 comentarios

    1. Hi Egor – thanks for the question. One measure of forecastability is the coefficient of variation (COV), defined as the ratio of standard deviation to the mean. Lower the ratio, higher the variability in demand and higher the forecastability. We typically put data on a scatter plot diagram and compare the COV against the MAPE (at whichever LAG the business requires). Hope it helps.

      Cheers,
      Manuel

    2. In addition, this came from a Gartner whitepaper:

      “The higher the volatility, the more difficult it is to have an algorithm-based demand forecasting process that can produce accurate demand forecasts. Demand planners can make a difference in the demand forecasting outcome. However, their time is at the premium and should be spent on developing an accurate and reliable forecast for products that are most important to the business.

      In the process of forecasting accuracy improvement, it is important to identify the critical few products from the many trivial ones. The historical demand volatility associated with specific SKUs plays a big part on the forecastability of a product.

      Demand analysis based on coefficient of variance (CoV) is an excellent method to normalize the volume differences. It produces a relative, unit-of-measure independent view of demand variances across the products in the company product portfolio.”

      1. Hi Manuel! Thanks for your answer. Unfortunately COV shows next to notning in the question of forecastability. If COV is low it only means that “average” is a good forecast. If there is trend and/or seasonality in the data (which is quite common) you can’t use COV. Take simplest line y=x and its COV is about 50%. So what: we can’t forecast it?

Deja un comentario

Tu dirección de correo electrónico no será publicada. Los campos obligatorios están marcados con *