# Methods to Calculate Forecast Accuracy

871

In this video Chris Dealy, WFM Evangelist at injixo, outlines the methods that contact centres can use to calculate forecast accuracy

## Methods to Calculate Forecast Accuracy

Forecast accuracy, I mean when it comes to forecasting, the rubber really meets the road when the time comes to build shift schedules for your agents.

You want schedules that consistently match the supply of agents with demand from customers, and you need to take into account the peaks and troughs across the days of the week, and the weeks of the month, and so on, and the M curve, or whatever, across each working day.

That at the end of the day is why we do forecasting. So you should start by taking a snapshot of your forecast at the moment you use it to create schedules.

Now a good workforce management application will have the means to do that. In injixo it is called ‘Forecast Versions’.

So the second step is after the event, so the days arrived, the weeks arrived, and you want to compare the actuals with what you saved in the forecast snapshot. Now a tip here is to be careful in choosing the evaluation timeframe.

Averages are dangerous – if I put my head in the oven and my feet in the freezer, on average I’m comfortable.

Averages are dangerous – if I put my head in the oven and my feet in the freezer, on average I’m comfortable.

Of course, it doesn’t work out like that. On a typical day there’ll be some intervals that have got more volume than your forecast, and other intervals with less volume than your forecast, and these two can cancel each other out.

That’s called compensating errors. So if you’re just looking at daily or weekly totals your forecast will probably look more accurate than it is.

So what I recommend is to use a technique for measuring forecast accuracy, and what you’re fundamentally doing is you’re comparing the numbers in two columns, the forecast and the actual column.

If you’re into statistics you can find all sorts of techniques for doing that comparison to the two columns.

There’s something called standard deviation which is measuring the amount of noise or spread in a sample of data, and the bigger the number, the more the spread.

You could use correlation coefficients where you know a perfect match, a perfect accurate forecast, gives a score of one, plus one, and complete inaccuracy is minus one.

The thing is, it’s quite difficult to interpret the results of those comparison, so I recommend using something called MAPE, Mean Absolute Percentage Error.

There’s a little image there showing the scary mathematical formula, but actually what it boils down to is something you can do easily with a spreadsheet.

And there’s a little example there showing at 15-minute intervals forecast compared with actual, looking at the absolute percentage error in each interval, and then averaging that across the day to give you a figure.

So that’s my first tip: use Mean Absolute Percentage Error, it’s powerful and easily understood. Another tip is don’t forget average handling time, it’s just as important when working out your workload, just as important as as volume.

Another tip is what does good look like? Well, if you’ve got a centre with at least 100 agents in, you should be aiming for within 5%, so a Mean Absolute Percentage Error of 5% or less. If you’re a bit smaller than that, because you haven’t got the safety in numbers, then 10% is a reasonable target for a smaller centre.

And the final tip for measuring is you’ve got to find the cause of deviations so you can avoid them next time.

With thanks to Chris Dealy at injixo for contributing to this video.

If you are looking for more great insights from the experts, check out these videos next:

Author: Chris Dealy
Reviewed by: Robyn Coppell

Published On: 28th Nov 2023 - Last modified: 30th Nov 2023
Read more about - Video, , , ,