other queuing and staffing models
As well as the Erlang C model there are several other call
center staffing and queuing models in use.
simulation
This is the most accurate way of modeling as it simulates a
'real' call center with thousands of calls arriving and being handled, using a
'Monte Carlo' simulation. It can also handle a mix of call types, call
priorities and agent skills, along with accurately calculating expected
abandonment rates.
cc-Modeler Professional is the
only low-cost call center tool that includes a simulation module.
'merlang'
Merlang (or modified erlang) is an attempt by one vendor to
overcome two of the restrictive assumptions of the Erlang C model. It adds
another two parameters to the calculations - 'Mean time to abandon' (or MTA)
and 'Expected Retry Percent'.
However there are several obvious issues with this approach -
| Parameter |
Notes |
| Mean Time to
Abandon |
This is not available from any ACD
statistics and must simply be guessed.
There is actually no such
thing as a 'mean' time to abandon - some callers hang up immediately on
reaching the queue; others will hang on virtually forever. The
relationship between abandonment rate and time on hold is quite complex
and varies very much by the type of caller, the nature of their problem
and their expectations. It is not simply a straight line relationship.
The result is that any
estimated MTA probably has an accuracy of no better than 20%. (Actually
improving the 'Wrapup Time' parameter has a much greater impact than
attempting to include a MTA).
(In contrast, the cc-Modeler 'Simulation'
module allows you to use actual ACD data as its inputs to calculate
abandonment rates). |
| Expected
Retry Percent |
This is the percentage of callers who, after
abandoning, will call again some time later.
It can only be guessed at.
In addition, retries will generally arrive in a
later 'interval' - this is not handled by Merlang. |
This model also does not address another Erlang
assumption of a 'steady state' situation where all calls are handled in the
same interval in which they arrive (i.e. there is never any backlog of calls
impacting subsequent time periods).
Our conclusion is that in trying to solve one problem,
Merlang actually introduces several others.
