Demand forecasting (or demand planning) is quite simply the process of estimating how much stuff you expect to sell. There are two basic techniques that you can employ when preparing forecasts:
- Follow your gut (judgemental forecasting)
- Analyze historical data (quantitative forecasting)
One simple forecasting process is simply to ask one or more business experts how many units they think you’ll sell over a given period. As crude as this technique may seem, this kind of judgemental forecasting is actually quite effective in certain situations. If you sell products that have a complex sales process then probably the best way to predict how many units you’re going to sell is simply to ask each of your sales representatives which deals they expect to close in the next X months.
Judgemental forecasting doesn’t work so well when you’ve got a lot of products and/or the sales process is really simple. The sales staff of most retailers, for example, typically don’t known the individual customers, much less which of them would be likely to come into the store or what products they might buy in any given period.
In situations where judgemental forecasting is impractical or ineffective then, another solution is to estimate future sales based on historical sales data.
An alternative to judgemental forecasting is to extrapolate future sales from historical data, using statistical techniques. One common form of this is simply to take the last X days sales and assume that the next X days will look pretty much the same.
For example: if you sold 90 widgets over the last 30 days, your “average” over the last 30 days was 3 widgets per day. Your “forecast” for the next X days then is
simply 3 widgets per day.
This kind of forecasting is sometimes called velocity reporting or, if you’re a statistician, a moving average. A moving average will be OK when forecasting certain products. However it doesn’t work so well for products that have seasonal sales patterns (e.g. if you sell more of something in December than you do in January) and it also doesn’t account for broader trends (e.g. we’re genearlly growing at X% or $Z per month/quarter/year).
Forecasting with SkuBrain
SkuBrain lets you use a mix of judgemental and quantitative techniques. Initial forecasts in SkuBrain are generated automatically using statistical techniques. However these are merely baseline forecast (a starting point) and once prepared, SkuBrain lets you adjust these using your own judgement, before they are used in any inventory planning decisions.
Additionally, SkuBrain makes use of much more sophisticated forecasting algorithms than a simple moving average. In certain cases these algorithms will create forecasts that are similar or even the same as what a moving average would yield, but the moving average is just one of many statistical models that SkuBrain will try.
Indeed, rather than relying on any single algorithm, SkuBrain runs a forecast tournament for every forecast that it creates. During the forecast tournament SkuBrain runs multiple algorithms against your sales history and automatically selects the one that produces the best results.
Finally, SkuBrain lets you create hierarchical forecasts, which allow you to organize your forecasts into a hierarchical tree. This makes forecasts easier to manage, however it also allows SkuBrain to create far more accurate forecasts than would otherwise be possible, since it can infer trend and seasonal patterns for products where this would otherwise be impossible to detect.
The following sections cover each of these topics in detail:
The first 2 of these topics might perhaps be considered just FYI. You could use skuBrain effectively without knowing anything about these. However the topic (on hierarchical forecasting) should probably be considered “recommended reading” as it deals with stuff that won’t be familiar to most readers and which may have a significant impact on the quality of your forecasts.