Because the word ‘forecast’ is in everyday use, it’s easy to
think that we understand what it means when the context moves from weather to business.
But forecasting in business can be easily derailed by one or more myths that can
frustrate our efforts to improve forecast quality - or even make matters worse.
In practice, up to 50% of forecasts fail to beat the simplest benchmark - a naïve
forecast, where last period’s actual is used as the forecast for this period. This
can be made even harder to bear by the expense of forecasting software and processes.
Many businesses focus exclusively on forecast accuracy rather than measuring and
managing the value added by forecasting.
While levels of error may be lower, in practice, it is often these forecasts that
fail to add value, because the level of error is higher than the simplistic benchmark.
And if your stable products also have the highest demand, even minor failures can
have a disproportionately large impact on the performance of forecasting as a whole.
In this case, focussing exclusively on simple forecast accuracy measures actually
makes matters worse.
The academic evidence is that there is little correlation between increasing sophistication
and improved forecast quality. It also confirms that so-called optimising functions
in software are no ‘silver bullet’. The ability to fit a mathematical model to
history is a poor predictor of forecasting performance. Most forecasting packages
use a similarly small set of forecasting algorithms – each of which will work well
in some circumstances, but fail in others. In practice, the best way to improve
performance is to identify where the chosen approaches destroy value and stop this,
rather than pursuing the unattainable ideal of ‘optimisation’.
All mathematical approaches work from the assumption that the future is more or
less like the past, which is often not the case. So it is easy to see why manual
intervention, often in the form of consensus forecasting, where collective judgement
is used to hone the statistical forecasts is a common feature of forecast processes.
Again, research shows that, while many interventions do add value, the majority
of judgemental adjustments make forecasts worse not better. The best way to improve
forecasts is often to do less; the trick is to work out what type of intervention
should be stopped and which encouraged.
Forecasting can be made complicated, often unnecessarily so. To avoid this trap,
decisions should be driven by evidence – in this case, if and where forecasting
has added value to the business, delivered in a language understood by all, not
an obscure statistic.
It is unlikely that one person or function holds the monopoly of knowledge and
expertise needed to forecast well; it is inevitably a collective effort. And, until
now, there has been no objective way to set comparative targets for forecast performance
since “good and bad” are dependent on the forecastability of demand, which varies
between products, geographies and over time. But when measured correctly, forecast
performance can be compared and this, allied to a philosophy of continuous improvement,
provides the foundation of a process for managing forecast performance.
Forecasts don’t need to be perfect predictions for them to add value to the business;
any reduction in error will provide better information for decision making and
both statistical methods and management judgement have roles to play. If we accept
that it is always possible to improve forecasts and it can often involve both less
work and expense, the trick is to work out which is the best method to use, where
and to continuously monitor and manage the performance of the process.