By Tony Dear, principal of Inventory Management Associates.
The driver of the Sales and Operations Planning process (S&OP) is the sales forecast. It is also in many cases the weakest link. In this article we outline some basic principles of Sales or Demand Forecasting for the S&OP process.
Basic approach: Forecast like the weather man
There is a fundamental difference between weather forecasting and sales forecasting. The weather man is only interested in obtaining the best possible estimate of future conditions. If his boss wants a sunny weekend because he is going sailing it tends to have little influence on the weather forecaster.
Its not quite like this in Sales Forecasting. I do not know of any sales forecasting person who firms up his forecast without at least some reference to the budget. In fact it is not at all uncommon where there is a trend of decreasing sales for the forecaster to forecast up in direct contradiction of this trend. We have seen formal systems mostly spreadsheets where there is a simple forecast calculation: subtract the sales so far this year from the budget and split the remainder over the rest of the year. We have always thought that a more appropriate starting point for a forecast is to use an algorithm to pick up the trend in the sales and extrapolate it into the future.
This raises the fundamental issue in Sales Forecasting. Are the forecasts a tool for motivating the sales force to greater endeavours or a method of obtaining the best possible estimate of future sales? In many situations a REALISTIC forecast (ie one that is the best estimate of the future) is not necessarily an ACCEPTABLE one (ie one that meets the budget).
If our interest is S&OP then the realistic forecast is the one we want. If we drive our S&OP process from acceptable forecasts then we can easily end up with overstock in addition to missing the budget!
This is a balancing act because we cannot deny the importance of the motivational force of working to the budget. We have found that there are two things that help.
Be aware of the issue of the two priorities in forecasting
Generate the initial forecasts based on market trends. Then the forecaster must explicitly change it to meet budget and such changes can be highlighted and discussed.
It reminds me of a comment from a managing director at a seminar I was giving a while back. All I want, Tony, he said, Are good forecasts that meet the budget. If only this was not so often a contradiction of terms.
The forecasting process
The most common method of Sales Forecasting for S&OP is the spreadsheet. These vary from the inane to the sophisticated with the majority clustering around the inane. They also grow as time passes and many have considerable complexity but they all retain one feature. They rely on the forecaster putting in some numbers which is his estimate of future sales for that month. They may or may not have past sales history on the spreadsheet.
We have found that a two stage approach to Sales Forecasting for S&OP is more appropriate.
Step 1: Generate a Background Forecast from past history.
Step 2: Present it to the forecaster to change in the light of his Market Knowledge.
This has two prime advantages.
Source of Inaccuracy: It enables the two inputs to be kept separately. When the actual sales become available it is possible to make two judgments:
Was the Background Forecast any good?
Did the forecaster improve the forecast where he changed it?
We have found that when forecasters change the Background Forecast they do not always improve it. Identifying the source of inaccuracy enables a more directed approach to improving it.
Activity Focus: It can target the forecasters activity to the process we want him to do ie inputting Market Knowledge. His task becomes relatively simple. If he knows something that is happening in the market then he can and should change the forecast; if he doesnt then he shouldnt.
Since the first process the generation of the Background Forecast - is totally automatic it takes no time. There are a multitude of algorithms out there to use and provided they take due account of Trend and Seasonality there is not much to choose between them.
The second process inputting the Market Knowledge is the one we need to focus on. The key thing here is the user interface. Remember the prime activity we want from the forecaster is to change the Background Forecast if and only if he has Market Knowledge.
We need to gear him into the forecasts he should be looking at in addition to those where he has some Market Knowledge of his own using simple exception reports. Some examples:
Products with unusual demand patterns
Products where accuracy has been poor
The format of this presentation to facilitate Market Knowledge input is most important. This forecaster input is the make or break input to the Sales Forecasting and the S&OP process. It is a simple truth in this particular activity that the quality of the output the forecast is highly dependent on the way the forecaster is presented with the data.
A spreadsheet showing all products with a row per product with a line of figures showing sales and forecast is probably hard to beat for inappropriateness in terms of matching the user interface to the task to be done. The forecaster has little visibility and it makes the task of converting the qualitative some Market Knowledge often vague and imprecise into a quantitative figure that is the forecast quite difficult.
Graphical representations are enormously beneficial in forecasting. We are much better at recognising a pattern from a graph than from a line of figures. With a graph we can see what is happening.
There are few topics that interest managers more about forecasting than the measurement of accuracy. The basic idea seems to be that if we can feedback to the forecaster a measure of his accuracy than he will improve over time. It sounds a good idea but we find very few situations where there seems to be a constant improvement in accuracy arising from such feedback.
The prime reason for this is the nature of the feedback. Feedback can be used effectively to limit bias ie if a forecaster tends to consistently forecast too high then constantly feeding back a measure of this bias to him will often move him nearer to reality. But constant improvement of the quality of forecasts is much more difficult. Just feeding back loads of forecast accuracy data month by month doesnt seem to have much impact.
There is one approach to accuracy measurement that is very simple that stems from the two step approach we mentioned above. In that approach the forecasters input is to change the forecast when he has specific Market Knowledge. We have found that a feeding back a simple percentage measure of the number of forecasts that he improved by changing them is a very effective approach to both focussing the forecaster on his Market Knowledge Input Only task and on improving accuracy. A forecaster who makes 100 changes in a month and ultimately is found to have improved 42 of the forecasts and made 58 worse is not doing a good job. A forecaster who improves 70% of his forecasts is doing his task well. We have found that feeding back this information enhances the percentage of improvement from such interventions.
Heres another point. When people raise a forecast they are more likely to be in error than when they lower it. There is for example a tendency to raise a forecast after two high demands in a row especially if it results in an out of stock situation. There is far less tendency to lower a forecast after two low demands in a row.
This article has set out three points that can serve as foci for improving the forecasting driver of the S&OP process.
Forecast Like the Weatherman: In the ideal world we should be aiming for the best estimate of the future. In a world of budgets that is not always the case. Recognition of the pressure of budgets to bend forecasts and in particular ensuring reasonable reality is maintained in forecasting when performance is below budget is a key aspect of forecasting for the S&OP process.
The Two Step Approach to Forecasting: Developing a Background Forecast based on sales or demand history and focussing the forecaster into changing it only in the light of specific market knowledge, improves accuracy and saves considerable time.
Measure Market Knowledge Improvements: The simplest measure of the effectiveness of the input of Market Knowledge the kernel of the forecasting process is to feedback to the forecaster the percent of his interventions that improved the forecast. This must exceed 50%.
About Inventory Management Associates
Inventory Management Associates (IMA) specialises in demand forecasting, supply chain planning & inventory management solutions both through sophisticated software & technical expertise. IMA has been helping companies improve their supply chain in many countries for over 30 years and in a wide variety of industries including retailing, merchanting (paper & steel), automotive spares, fashion, music and food & beverages. IMA's software & support compliments legacy & enterprise software to deliver improvements to clients key supply chain metrics. Unique in its approach to client support through rental of it's software IMA ensures that the improvements to the clients business are continually met.