Hi,

Can some one please tell me the correct average method to use when working out averages such as AHT, Avg ACW, Avg Talk Time.

If I use the 'mean' average to work out lets say the AHT, the average will be skewed if the data consists of some very high ACW or Talk Times. Shouldn't we be using the 'median' average instead to work out the above metric if we want to set an accurate baseline average?


Would appreciate if others can share their thought and advice.

Box and Whiskers

Regards

DaveA

Hi,

If you are reporting an average AHT you need to use a weighted average.

eg:
250 AHT * 100 Calls = 25000 Seconds
250 AHT * 150 Calls = 37500 Seconds
249 AHT * 100 Calls = 24900 Seconds
600 AHT * 2 Calls = 1200 Seconds
-------------------------------------
= 88600 Seconds / 352 Calls

The above would give you a AHT of 252 seconds. (In excel you can use the sumproduct function)
This will assume that the majority of your AHT's are in the 250 regegion and add almost no weighting to the few calls with high AHT's



When forecasting AHT's I use a "poisson" distribution. This is a discrete probability distribution that expresses the probability of a number of events occurring in a fixed period of time.

Hope this helps.

I think this is a very basic, but very important question which we don't think about enough. I think the answer depends on why you want to know your AHT? Couple of observations:
1: there are three common measures of "central tendency", the mean, median and mode (modal). We tend to use the mean but as you point out, if the distribution is skewed this is misleading, for instance if you have a few very long calls the mean will be higher than most of the calls. The median is better, in that it tells you the AHT in terms of the length of call where 50% are longer than that, and 50% are shorter. But you can argue that the best measure is the modal AHT, which is the most common call length (you'll need to categorise call lengths, into say 15 sec or 30 sec units to do this)
2: but if you are intersted in variations in how agents are handling calls, and how this relates to some measure of output (eg overall call quality) then perhaps the most informative measure is one of variation eg the standard deviation of call durations for each agent. Agents who are "going through the motions are likely nto have low sd's (they do the same thing irrespective of the task/customer) whereas agents who are good communicators/responsive to the customer are going to have higher sd's: their call duration will vary as a function of the customer and their task eg FCR is likely to be higher for agents with higher sd of AHT.
Hope this helps
Guy (horizon2)