Statistical demand forecasting?

Our demand forecasting software, Prophecy, naturally has a comprehensive statistical forecasting engine.  However, statistical forecasting in isolation may not be the optimal solution for real world demand forecasting.

More questions....

• How much of your forecasting process involves reviewing existing forecasts in the light of the latest actuals and known future event changes?  Compared to originating a completely new forecast e.g. to initialise a new year?  Your software should help with the first, as well as the second.
• How much does another month of actuals as input to a statistical forecast improve its accuracy anyway?
• Does greater mathematical complexity equate to more accurate statistical forecasts?  (Our tests suggest not necessarily.)
• How do you ‘clean up your data’ for statistical forecasting (without excessive use of judgement)?
• What is the magnitude of the trend (which statistics may be good at forecasting) versus the magnitude of events, such as promotions, competitor activity etc. on the short and medium term sale?
• Purchasing behaviour is behavioural, not mechanistic.  By this, we mean that sales may respond differently to the same causal factor.  For example, where the buyer was overstocked last time they may order low this time.
• How accurate, and at what time, product and customer granularity, do the forecasts have to be?  If you've tested the statistical algorithms and you find they are accurate enough.... then great!  But our experience suggests they are unlikely to be accurate enough in most real-world markets.

So, is there a solution?

Nobody said demand forecasting for business was easy!

Consider the following alternatives as replacements for complimetary approaches to statistical forecasting:

• Use the 80:20 (Pareto) rule or other types of product / customer classifications to focus time and attention on the relatively small number of skus or customers that account for a relatively high proportion of sales.  If you have not already run an 80:20 analysis on your skus you may be surprised how few account for so much of your business.  (Tip: Prophecy™ can be configured to provide instant Pareto views for any selection of products or customers.)
• Forecast the 80% of skus that typically account for the smallest 20% of sales less frequently or by exception.  For example, after next year has been forecasted only revisit the forecast for known significant events or by exception as the year begins to sell.  Prophecy's built in accuracy analysis, trend graphs and Advisor text will help your forecasters quickly identify when these exceptions occur.
• For the 80% of skus that typically account for the smallest 20% of sales, consider whether it is more efficient to let the statistical algorithm generate a forecast and then spend time correcting it, or use one of Prophecy's easy data-entry features to eyeball the sales line and spread / pro-rate a total annual forecast across the year.  (Hint: maybe it's the latter!)
• The same is true for customers. Forecast the 20% of customers who account for 80% of sales individually and the rest in pre-aggregated groups. Counter-intuitively perhaps, but grouping many small customers into a single forecasting entity will smooth out their individual variations and make their sales easier to forecast.
• Remember that a majority of business forecasting activity involves reviewing your existing forecasts in the light of the latest information, rather than generating completely new forecasts.  Another month of history is unlikely to help a statistical algorithm improve the current forecast, whereas applying human judgement to the latest information about the marketplace, buyers, stocks in the trade, promotions, competitor activity etc. really can improve accuracy.  Use a tool like Prophecy™ to help forecasters flex their forecasts in the light of these ever-changing factors.
• Review the existing forecasts with a "can I defend this number?" mindset.  In other words, look at moving annual totals, percentages versus previous years, the sizes of forecasted peaks and troughs versus previous years.  Keep objectively challenging the forecasts until growth rates and peaks/troughs look 'reasonable'.  Although it takes time, Prophecy™ can greatly speed the process up for you.  The ROI, in terms of the financial benefits of improved forecast accuracy, makes it time very well spent.