6, some split tours must be employed to produce
reasonably efficient schedules. With split tours completely prohibited,
a scheduling efficiency of 89.16% is realized and 389 agents are
needed to achieve the target service level of 80%. With split tours
completely unrestricted, efficiency increases to 99.67% and only 348
agents are required. For the unrestricted case, the model schedules 90
weekday split tours (26%) and 63 weekend split tours (38%). However,
as the figure illustrates, nearly all of the efficiency improvement can
be achieved with no more than 20% split tours. In practice, the model
is extremely valuable in providing this type of insight.
We also compared our optimization method with the traditional
approach, which is to employ an integer programming model
and minimize the total number of scheduled agents while enforcing
staff requirements on all intervals. Unless the schedule is
perfectly efficient, some intervals will be overstaffed and the
composite service level will exceed the target. The interval service
level requirement can be iteratively modified to converge on a
composite target, but a new integer program must be solved at each
iteration. For the operationally normal case where the total number
of agents is specified, the agent population could be fixed by a
constraint and a different linear objective function could be
employed. For example, we could minimize the maximum normalized
surplus encountered on any interval. From our integer
programming experiments, we observe that the most effective
approach is to retain the objective of minimizing total staff, and
iteratively adjust the interval service level requirement (resolving
the model) until the optimal objective value is equal to the total