Having built up our monthly forecast, whether simply or in “layers” using lifecycle propensity calculations, we now need to forecast at a daily level.
We’ll take a simple view on this to begin with. First, gather many weeks of inbound call data showing the volumes by day.
Let’s suppose your data looks something like this:
If we graph this we get the following:
It’s clear that there is some kind of “shape” across the months but it’s masked somewhat by the irregular way our calendar is organised. Suppose now that we “slide” the data around a little, emphasising the day of the week rather than the day of the month:
The graph of this data is:
This immediately looks more useable, but suffers from the fact that we’ve “lost” data at either side of the graph. If we fill in with real data from the previous and next months we get the following, much more promising data:
We can add a column to our table of figures that is simply the average of the 6 months of data. By aligning our forecast month to the appropriate day of the week we can use the distribution of calls in this last column to define the daily breakdown of our month’s total forecast.
As a first pass, this is likely to lead to a daily forecast that would provide a moderate level of accuracy.