Interpreting the TSF

The TSF is commonly used as SL metric in call centers, but it is of interest to pay attention
to its interpretation, especially for the callers. We will consider the regular 80/20 TSF,
where we assume that the number 20 is chosen such that the callers in general find the
service ’good’ when they wait less than 20 seconds and ’bad’ when they wait more than 20
seconds. An 80/20 TSF means that 80% of the customers receive good service, and 20%
bad.
Consider an individual customer that belongs to the 20% that received bad service. To
this customer it is right now irrelevant if the SL was 50/20 or 80/20, in the former case
there are just more unsatisfied customers. To the unsatisfied customer the SL becomes
relevant when he or she tries to call again. If the TSF at that moment is again 80/20, then
the probability of another bad experience is 0.2, or 20%. Almost 1 out of 100 customers
have 3 consecutive bad experiences. And how many customers will try a third time after
two bad experiences if they have alternatives? If the competition is strong then offering
only an 80/20 TSF can lead to churn. Thus whether a 80/20 TSF or any choice is the
right SL for a CC depends on the behavior of the callers. Will they call back after a bad
experience, and is 20 seconds indeed the correct borderline between good and bas service?
Things become even more complicated when we take abandonments into account, see the
next section on this subject.
Choices related to SL become even more difficult when we consider call centers with
multiple types of calls (see chapter 9 for more on this). Consider for example two types of
calls for which we like to obtain both an 80/20 SL. Now what if we obtain 70/20 on one
and 90/20 on the other? and if we have the choice, with the same means, between 70/20
and 90/20 or 75/20 and 80/20? The former has a better average SL (assuming an equal
load), the latter shows less variance. The answer depends again on the behavior of callers
and the nature of the service: will they mostly generate the same type of call, or do they
change type? In the former case we should consider the types independently, in the latter
case we should perhaps focus on the average SL.
The situations becomes even more complicated when we have different SL constraints
for different call types, for example because we want the sales line to have a better SL than
the after-sales line. Here we might have 90/20 and 70/20 constraints, and still be more
satisfied when we realize 95/20 and 65/20, simply because we value individual sales calls
higher than after-sales calls. A SL definition that corresponds with our intuitive notion of
SL might be to have a constraint on the high-value calls of 90/20 and an average constraint
of 80/20.