Demand planning and forecasting are about people, processes and technology. The three broader aspects of demand planning today are finding the right model, understanding the performance of the forecast and managing the demand planning and forecasting processes.
There are many developments on the technology front like the emergence of data sciences and predictive analytics that are playing an important role in shaping the practice of forecasting today. In this article, I intend to briefly discuss the art and science of demand planning.
The first pressing question is how does one choose a forecasting method?
Most often firms often choose a single forecasting technique as one fit all solutions for all products. They are in the lookout for the “forecasting technique that will predict the future without much error”. The most commonly used method is the time series method.
When the time series method is not able to predict a landscape which is dynamic today, we move to the judgmental method. The most dynamically used judgmental model is “sales force composite”. This is closer to the concept of predictive analytics based on last mile demand assessment. The use of the causal method is limited to a few organizations.
In a world where demand has become dynamic, using a single method of forecasting can be dangerous. The uncertainty emerging due to industry completion and its impact on organizations to become more agile and improve the service levels has made the task of forecasting even more difficult.
There is a need to understand segmentation based forecasting as multiple supply chain resides in the same business. The practitioner needs to realize that a single forecasting method may not work in such a fast-changing landscape.
I have been often repeating this in my speaking assignments at various forums. Sunil Chopra’s strategic fit analysis and framework is needed to solve much of the problems that confront demand-planning analyst today.
Here, data science can provide a good clue to demand planners about multiple supply chains and their characteristics. Using data science, demand planners can understand the segment, which shows static demand and hence can comfortably use peak to average time series techniques like Holt and Winter Model.
Use of big data and causal methods may help identify many macro and micro factors important in explaining the causal effect. Using the coefficient of determination and variable coefficient one can easily predict a lot of implied uncertainty and decide on supply chain strategy.
Yes, where predictability is not possible, firms need to build their information channel and their responsive capabilities to react in the shortest possible time. Most firms have failed to incorporate this into their supply chain.
The second issue that is important is to measure forecast performance. It is important mainly because the buying of the forecasting process in an organization becomes easier and so does supply chain integration. The goal of any forecasting system is to be as accurate as possible. One measure of accuracy is to minimize error, which includes two components bias and magnitude.
Demand planner should use at least one measure for bias and another for magnitude on an ongoing basis to assess accuracy. Bias is defined as a systematic difference between forecast and actual result over a period of time.
The bias could be measured with the help of mean error, cumulative forecast error or mean per cent error. A large and consistent negative or positive error signifies a forecast that is consistently high or low. Once the bias has been identified, management can take active steps to correct the forecast by adding and subtracting the amount of bias from the predicted forecast.
The magnitude indicates the variance between the actual result and the forecast. This can be measured through mean absolute deviation and mean absolute percentage error. When a magnitude reflects high forecast inaccuracies, it cannot be corrected as easily as the problem arising from bias inaccuracies. The problem may be with the forecasting systems and processes. In this case, the forecasting system needs to be revised and streamlined.
This takes us to the third pillar of a good forecasting organization. Firms, in order to make forecasting as a lifeline of their organization, need to build in dynamic alignment. There is a need to align the organization to its operating environment. Forecasting requires a larger buying from the organization and hence the chief executive should lead it.
The sales and planning exercise has a meaning if it starts by mobilizing change and this is possible only through executive leadership. The supply chain strategy should be translated into operational terms. All the functional area should be aligned with this strategy as we know these areas have their own forecast.
The big question is how do we arrive at a single forecast? The process is to iron out various issues facing marketing, supply chain and finance function and arrive at a single forecast which aligns sourcing, production distribution, marketing and sales. The organization should motivate to make forecasting strategy everyone’s job. There is a need to make sales and operations strategy a continual process.
These are the broader pillars of a good forecasting organization. These pillars will make forecasting the biggest aligning tool and supply chain of such organization will help create a competitive advantage for the firms.
This article is authored by Rakesh Paras Singh, Chairman, Institute of Supply Chain and Management