Getting Started With Forecasting

In this instalment of our video series exploring what you need to know about forecasting, and what they won’t tell you, Call Centre Helper’s Jonty Pearce explains forecasting basics.

#3 Forecasting Basics

I’d like to take you through some of the basics of forecasting. The best way to look at a forecast is just to plot out the data over time, something that’s called time series data in a contact centre.

Monthly information is quite good to look at. And what you’ll see here is that you’ve got an interesting pattern of data here. This shows basically both a trend line, you can see stuff going up over time, and although the trend here seems reasonably consistent over time, it does seem to accelerate a little bit away towards the end.

And what you’ll spot is that there’s also some seasonality, so, for instance, every December there’s actually a little bit of drop-off in traffic. This may well be because the contact centre is closed during that period.

So the first thing you need to do in forecasting is to really try and break the whole forecast down into three different components. And that is what’s called level, which is the level that you’re at currently. The trend, which is the difference between one period and the next.

And if you take the level and trend together you can form a trend line, which is the overall approach. And also to look at the seasonality, that is how much do things vary over season.

Once you’ve done that, you can do something like make a weekly or an annual forecast based on that. So you can see, for instance, here’s a monthly plot of data, and you can see that this does seem to have some form of trend over time.

This is looking at about three and a half years’ worth of data. And you can see that there is some degree of seasonality here, that, for instance, on the 12th month it seems to be down. There seems to be a bit of an anomaly here.

So now you’ve got the data, the first thing you can do to that is try and work out some form of trend line. This is a linear trend line that we’ve applied here. Once you’ve got the linear trend line you need to then try and determine some form of seasonal pattern.

So you try and get some form of regular seasonality being added to that data. And then once you’ve built your seasonality model, and once you’ve built your trend line, you can use that data then to predict off into the future.

One of the common mistakes that a lot of people tend to make when working out seasonality, is to basically say look at 12 months’ worth of data.

And one of the things they do is, for instance, and here’s a whole load of contact data, this is made up data. So what you see here is that you’ve got a lower number of contacts in January, higher in December, and you basically add all of that together, that gives you a number for the annual contact volume.

You then work that out as being 100%, and then you can work at a percentage, and then if you take the average of one 12th of that, you could then work out a seasonality figure, starting at 62%, 132%.

So that’s one thing that a lot of people do, but there is one problem with this and that’s that it’s assuming that this data goes down.

What you see here is that generally at the beginning of year the numbers are lower, and the end of the year it’s higher.

So what you’ve got to work out is what the trend component is of that. So what you have to do is you first of all need to de-trend and then de-seasonalize the data, so you then effectively work out what your trend is, take your trend and remove the trend from your data, and then work out what your percentage seasonality is over the year.

For more in this series watch these videos next: