Don't rely on due dates
In this example, we found that relying on the customer invoice due dates would mean that the forecast would have been more than 20% in error in 47 weeks out of a 53 week year. We tested various scenarios and found that the best algorithm was to adjust the due dates based on the average delay at a customer level adjusted for variability. In this case the variability was being caused by the erratic payment behaviour of a significant proportion of customers.
|
Sometimes a forecast error correction model can be helpful. In this example, that was not the case.
Garbage In, Garbage Out
The best algorithm may not prove to be enough to get an accurate forecast.
|