+44 (0)7460 109038
Virtually every company needs to generate forecasts of their short to medium term sales. Being able to forecast demand more accurately has major commercial advantages, whether the forecast is used:
Yet within real world markets, many factors conspire to make accurate demand forecasting difficult to achieve.
In the first place, demand forecasts are frequently used for all the purposes suggested above. This leads to conflicts between optimism and pessimism and potentially introduces 'political' influences into the process. Examples are the different role of the profit forecast (probably conservative) and the sales plan (probably optimistic), or where marketing expenditure is loosely linked to the sales revenue of products (and therefore leads to defensive forecasting to protect planned marketing spends).
There are also conflicts in terms of which units should be forecasted - orders-based for production forecasting or invoice-based for financial forecasting.
Similarly, forecasts by week by sku (stock keeping unit) at the total customers level for the next 12 weeks may be required by production planning. But, this time horizon is far too short and this level of product and time detail is potentially much too detailed for marketing and sales planning purposes. In addition, it omits any reference to customers detail. Frequently, sales to major customers need to be planned and managed.
The important point is to have a clear vision of who the primary customer or customers of the forecasts are. Select the appropriate level of detail and time horizon accordingly and accept that secondary customers of the forecast will probably have to accept sub-optimal forecasts. In many situations it is helpful for both Marketing and Sales to generate demand forecasts. Sales are often more likely to possess the detailed short term knowledge whilst Marketing need to 'own' the forecasts as a result of their role as brand profit 'custodians', and possibly have a clearer knowledge of longer term influences. It is vital here that each area is clear about the role and purpose of the forecasts they produce, and that forecast publication schedules optimise the currency of the data used as inputs, and given as outputs, by each forecaster.
The second major difficulty of forecasting in real world markets is the very nature of these markets. They frequently exhibit some or all of the following characteristics:
In essence, the dominant characteristic of real world markets is "NEVER THE SAME THING TWICE", especially with the major customers, combined with 'lumpy' or irregular demand from the smaller customers.
This makes it hard for commonly used forecasting approaches such as statistical forecasting to provide acceptable results over a short to medium time horizon. 'Acceptable' means able to predict the size and timing of those short-term, but business critical peaks and troughs in sales that result from the factors listed above.
All statistical methods either even out the peaks and troughs in sales history to produce trend-based forecasts, or else they look for repeated patterns in the historical peaks and troughs to make future forecasts.
But, if the peaks and troughs in sales are often caused by factors that are genuinely 'random' events, such as promotions or competitor activity, how can statistical methods help you forecast? On the one hand, a smoothed forecast has little value if the primary purpose for forecasting is to predict the short term sales peaks and troughs. On the other hand, how valid is the second approach (looking for repeated patterns) given the random nature of historical peaks and troughs? (Even 'seasonality' can vary with the weather, public holidays, world events etc..)
If you cannot use statistics, what can you use? In the majority of situations, informed management judgment (or 'finger to the wind' as cynics might describe it) is actually more likely to produce better results within real-world markets.
The essence of judgmental forecasting is the application of the business manager's knowledge and interpretation of past events and activities, and their effects on sales, to planned future events and activities. The result is a 'judgmental' forecast for the future sales periods.
The key factors to consider are well known:
Although there is never the same thing twice, developing and applying an understanding of how sales respond to different types and combinations of events is the most effective way of generating a forecast. It has spin-off benefits too, because it forces marketing and sales people to think long and hard, and hopefully objectively, about which factors really drive their sales.
The method most likely to succeed is forecasting from the 'bottom up', and reviewing from the 'top down'. This means generating the forecasts at the lowest (relevant) level of detail using the process described above : the 'bottom up' method.
One then compares how the resulting forecasted year on year growth rates and Moving Annual Totals compare to expectation, historical or current growth rates and Moving Annual Totals. If the 'bottom up' results are out of line with the 'top down', then the 'bottom up' forecasts need to be revisited to identify the sources of the difference.
This process must continue until the 'top down' and 'bottom up' forecasts are consistent.
The forecasting methodology recommended in this article places a lot of emphasis on the knowledge and judgment of the forecaster. That is unavoidable given the nature of the market, but it follows that developing a good forecast can be a labour-intensive process.
Forecasting software can help here, by providing the forecasters with a productive and flexible environment in which to analyse and manipulate their forecasts. A lot of companies use Excel spreadsheet based systems. Some use an option from their ERP (Enterprise Resource Planning) system.
None of these approaches are ideal.
Spreadsheet based systems are generally difficult to maintain, in terms of adding new products or customers, updating actuals or rolling forward years. They also tend to show the data in fixed views due to the fixed rows and columns structure of spreadsheet programs. Some analytical capability can be introduced by building clever spreadsheet macros, or by users reformatting data in different ways within their spreadsheets, but this approach tends to be clumsy and labour intensive. In addition, aggregation / consolidation of data across products and customers tends to require considerable manual processing.
In addition, spreadsheets are essentially single-user productivity aids, whereas business forecasting is normally a multi-user activity. Delays and inaccuracies get introduced through the need for consolidation of spreadsheets. One change can require the whole, cumbersome round-trip process to be repeated.
Terminal / browser-based systems and ERP options overcome the maintenance problems but tend to be inflexible, lacking in functionality, and do not provide the variety of instant graphical views that a specialised PC based system makes possible. In addition, such systems can sometimes have performance problems - where transaction processing systems and decision support systems operate on the same host, transaction processing systems necessarily get preference in receiving processor time. In addition, it is hard to give these systems the degree of user-friendliness which sales and marketing users generally prefer.
Many purpose-built systems for demand forecasting are available, though they don't all cover the same range of features. Here is a checklist of features to look out for:
Forecasting in the real-world is a difficult process which does not invariably lend itself to automated statistical approaches. The so called 'finger in the wind' / 'judgemental forecasting' method, if carefully implemented and with appropriate systems support, can yield quality improvements in forecasting results.
You need a good system, forecasters who really understand their markets, and above all, the strongly held determination to put it all into practice.