## Sales

Suppose we have a sales call centre with a conversion level which varies on a weekly basis between 20% and 25%, with varying AHTs.

The Sales manager is pretty confident that conversion will increase to 35% in 2 months, and is asking what the AHT will be.

The table below shows the conversion levels and AHTs over the past 8 weeks for two call centres. Notice that Centre 1 has conversion levels which vary quite a bit from one week to the next while Centre 2 is much more predictable.

Let’s consider two different ways of trying to estimate the AHT when conversion gets to, say, 35%. This jump in conversion might relate to a group of customers reaching the end of a contract and looking to buy more product, or perhaps the implementation of IVR technology filtering out non-sales calls before they are presented to advisors.

The two different methods use high-school maths: firstly, we’ll take a look at simply plotting the values on a graph and using a straight line to extrapolate out; next we’ll use simultaneous equations to try to estimate the AHT of successful and unsuccessful calls.

The starting point for both systems, however, is to plot a scatter graph of the data. It is instantly obvious that one of the weeks has rogue data. Looking at the table it seems that we need to remove week 5 from our consideration: On discussion with the operational team we learn that there was a trial going on that week which was not adopted – just the kind of information needed to back up our hunch that we should delete that week’s data from our table.

[table]

[graph]

This leaves us with the following data:

[table]

[graph]

[tip: clean your data]

Now we can make progress.

#### Straight-line estimating

#### Simultaneous equations

#### More advanced methods?

There are other, more advanced, methods available to use if neither of these simple systems provide you with the accuracy you require.

Interpolation methods abound, though are useless for this exercise where we are interested in estimating values beyond the bounds of data achieved to date. It is necessary to choose an appropriate *extrapolation* system to build on the data you already have in order to estimate out.

In most cases, however, the two systems above will enable you to reach your number with accuracy well within tolerance and further analysis may well result in diminishing returns.