Hi All,

This is my first ever topic and am new to the forum, so please be gentle! If this question has already been asked i'd be grateful if someone could point me in the right direction.

We've recently opened ourselves up to the minefield that is Interval forecast accuracy without thinking about what good forecasting looks like first.

Historically we've always used forecast calls by interval against total calls offered to get an accuacy figure. That's never been a problem at either daily, weekly or monthly levels as we're normally within a 5% variance. However because interval calls can vary anywhere between 30 calls and 400 calls per interval being 2 calls out on a 30 call forecast shows roughly a 7% variance and because we didn't think about the how this should be reported, this is being reported as red on our RAG (Red Amber Green) report. My argument is that this isn't a bad forecast and that the measurements are wrong.

I want to move to a sliding scale of accuracy based on the expected calls per interval and I wondered if anyone had any advice on the subject and what scale to use?

Thanks all,
Dale.

Hi Dale,

Its a much sought after topic in WFM world.
Accuracy can be gauged in many ways and most aggressive being using MAPE (Mean Absolute percentage error). Mape doesn't allow errors on either side of scale to compensate for each other.

For e.g a variance of +10 in one interval and -10 with consolidate to 0 overall on average.
However MAPE takes the absolute values to show error rate at 10% approx

Having said that its quite a tricky metrics to gauge interval level accuracy since LEAN intervals would show higher error rate due to lower volume base even if volume is off by couple of calls.

I would suggest to set accuracy targets for Lean and Peak periods individually and chase the target on standard deviation.

Lean intervals can be defined as those intervals which have less than 2% share of total day volumes in 24 hours operating window. (definition may change as per requirement)

Hope that helps.


Pamposh