Agent Scheduling of Call Center Using Decomposition Technique

Netnawee Um-in, Wipawee Tharmmaphornphilas

Department of Industrial Engineering, Chulalongkorn University, Bangkok, Thailand
(e-mail address: [email protected], [email protected])

Abstract - This research studies an agent scheduling of a 4) Assign agents into work shift.
call center where the number of agents and skills are Our paper covers the last two steps.
predetermined and varied during each period. Agents must
be assigned into work shifts in order to minimize labor Most of relevant papers on daily shift scheduling of call
related cost and also satisfy working constraints such as centers have been studied in single skill and multi-skill
worker availability, work duration and break requirements.
agents. The problems are categorized into two groups.
We decompose this problem into two parts which are solved
sequentially. First is to determine the minimum number of First group is to determine staffing requirement, second
required agents for each shift across a month that covers the group is to determine the number of agents for each shift
predetermined number of agents in each period. The in order to minimize cost. Integer programming models
outcomes from the first part are used as input for the second [2, 3], integer programming heuristics method [4], and
part which is to assign agents into work shifts. Mixed integer metaheuristics method [5] were proposed to solve the
programming models are proposed to find solutions for both problems. Kabak et al. [6] considered part-time staff.
parts. Part-time staff are needed when there are peak demand
periods. The problem is solved by a mixed integer
Keywords - Workforce Scheduling, Call Center, Work
programming. Cordone et al. [1] studied balance between
Shift, MIP
full-time and part-time contract to avoid concentrating on
a single type. Integer linear programming models were
proposed to solve the problem. However, the proposed
I. INTRODUCTION
method did not work for huge size. Hence, linear
Customer satisfaction is one of the crucial part for programming relaxation method was developed to find
business competition. Simultaneously, call centers have solution. Then, Cordone et al. [7] presented a heuristics
become the important channel to communicate between based on decomposition for the same problem but legal
companies and their customers because it is another way conditions were taken into account.
to increase the efficiency of service that can directly and
conveniently provide information, communication and Furthermore, Taskiran and Zhang [8] considered
assistance in other services for customers. Since the overtime. Overtime schedule is an alternative to increase
personal cost typically amounts to about 70% of capacity for peak demand periods. The shift scheduling
operational costs of call center [1], the trade-off between a problems with part-time, full-time, and overtime were
number of agents and labor related costs is the crucial solved by a mixed integer programming model, but the
issue for workforce scheduling. It is important to model is hard to solve. Hence, a two-phase sequential
determine the number of required agents for each shift approach was developed.
because inadequate number of agents has an effect on
customer satisfaction such as long waiting time. On the The staff scheduling problem is to allocate agents into
other hand, if there are too many agents, labor related cost work shifts over a week or a month that has more than
will increase. Hence, appropriate determination of the one shift per day. The staff scheduling problem in the
number of agents and proper agent assignment keep literature basically try to assign agents into work shifts
minimizing cost. during a week and is divided into two phases: to find an
optimize shift schedule and to construct weekly tours.
In this paper, we allocate agents into shift schedule. The Mixed integer programming models were proposed to
shift scheduling process of the call center can be classified find solution for this problem [9, 10].
into four steps [2].
1) Forecast a customer demand for each period which This paper studies a monthly agent scheduling in call
is usually between 15 minutes to 1 hour. center. We consider in single skill agents and overtime
2) Determine the number of required agents for each schedule, hourly and daily overtime. Agents must be
period under expected service level in terms of customer assigned into work shifts during a month in order to
waiting times or abandonment rates. minimize labor related cost and also satisfy working
3) Determine the minimum number of agents for each constraint such as worker available, work duration and
shift over the planning horizon that covers the break requirements. The rest of this paper is organized as
predetermined number of agents in each period. the following sections. In section II describes the problem

978-1-5386-0948-4/17/$31.00 ©2017 IEEE 80

 Proceedings of the 2017 IEEE IEEM

and presents mathematical models to find solutions. Decision Variables:

Results are showed and discussed in section III. Finally, Number of agents assigned into shift j of
conclusion is in section IV. week q, day r for task t
1 if agent i is assigned into cycles c of
II. PROBLEM STATEMENT AND MATHEMATICAL week q, day r, 0 otherwise
MODEL 1 if agent i is assigned into shift j of
week q, day r, 0 otherwise
The problem is involved an agent scheduling in call Maximum number of working day of all
center that belongs to a bank. The call center operates 24 agents in a month
hours a day. There are 115 shifts in a day. All shifts Minimum number of working day of all
include break requirements. The duration of shifts are agents in a month
varied based upon overtime that they carry: 0.5, 1, 1.5 and Number of overtime working day in a
2 hours. For example, shift 1 starts at 6 am ends at 3 pm month of agent i
where shift 2 starts at 6 am but ends at 3:30 pm since it
has a half-hour overtime. Starting time of shifts vary
Mathematical Model:
every half an hour, the first shift starts at 6:00 am and the
last shift starts at 11.30 pm. There is no shift which ends 1=
between 00:30-05:30 am due to transportation constraints. ∑∈ ∑ ∈ ∑ ∈ ∑ ∈ _ +∑∈ _ (1.1)
Shifts are assigned into work cycles that define starting
time of agents. There are 4 work cycles per day. Cycle 1 ∑ ∈ _ ≥
includes every shift starting between 6 am to 8:30 am,
cycle 2 includes every shift starting between 9 am to ;∀ ∈ , ∈ {1}, ∈ {1}, ∈ (1.2)
11:30 am, cycle 3 includes every shift starting between 12
pm to 3 pm and cycle 4 includes every shift starting ∑ ∈ _ +∑ ∈ ( ) _
between 9 pm to 11:30 pm. Agents must remain in the ≥ ;∀ ∈ , ∈ , ∈ , ∈ \ ∈1∩ ∈1 (1.3)
same work cycle in each week. Every agent must be
assigned a day-off. The number of agents and skills are ∑ ∈ ≤ ;∀ ∈ , ∈ , ∈ (1.4)
predetermined and varied during each period (half an
hour). Agents must be assigned into work shifts in order ∑∈ ≥ ;∀ ∈ , ∈ , ∈ , ∈ (1.5)
to minimize labor related cost. To solve this problem, a
mixed integer programming model is developed. ≥ ∑ ∈ ∑ ∈ ∑ ∈ ;∀ ∈ (1.6)

Notations: ≤ ∑ ∈ ∑ ∈ ∑ ∈ ;∀ ∈ (1.7)
Set of agents, i = {1, 2, …}
− ≤2 (1.8)
Set of shifts, j = {1, 2, …}
Set of cycles, c = {1, 2, …} ∑ ≤1 ;∀ ∈ , ∈ , ∈ (1.9)
∈
Set of weeks in a month, q = {1, 2, …5}
Set of days in a week, r = {1, 2, …7} ≥ ∑ ∑ ∑ − ;∀ ∈ (1.10)
∈ ∈ ∈
Set of periods, k = {1, 2, …}
Set of tasks, t = {1, 2, …} ∑ ∈ ∑ ∈ ≥5 ;∀ ∈ , ∈ {1, … ,4} (1.11)

Parameters: ∑ ∈ ∑ ∈ ≤6 ;∀ ∈ , ∈ (1.12)
Number of required agents in period k of
week q, day r for task t ∑ ∈ ∑ + ∑ ∈ ∑ ≤6
_ 1 if shift j covers period k of the current
;∀ ∈ , ∈ {2, … , 5}, ∈ {2, … , 7} (1.13)
working day, 0 otherwise
_ 1 if shift j covers period k of the next ≤1− ∑ ( )
working day, 0 otherwise
Number of agents who can perform task ;∀ ∈ , ∈ , ∈ , ∈ {2, … ,7} (1.14)
t
≤1−∑ ;∀ ∈ , ∈ , ∈ , ∈ {2, … ,7} (1.15)
1 if agent i can perform task t,
0 otherwise ∑ ≥ ;∀ ∈ , ∈ , ∈ , ∈ {4, … ,7} (1.16)
1 if shift j is a subset of cycle c,
0 otherwise ( ) +∑ ≤ 1 ; ∀ ∈ , ∈ {2, . . ,4}, ∈ {2, . . ,5} (1.17)
Minimum number of working day in a
month ∑ ∈ =0 ;∀ ∈ , ∈ {5}, ∈ {4, … ,7} (1.18)
_ Overtime cost of shift j
_ Daily overtime cost of agent i ∑ ∈ = ;∀ ∈ , ∈ , ∈ , ∈ (1.19)

81
 Proceedings of the 2017 IEEE IEEM

∈ {0,1} ;∀ ∈ , ∈ , ∈ , ∈ (1.20) A. Part I: Work shift scheduling

∈ {0,1} ;∀ ∈ , ∈ , ∈ , ∈ (1.21) The purpose of this model is to find the minimum

number of required agents for each shift across a month
≥0 ;∀ ∈ , ∈ , ∈ , ∈ (1.22) that covers the predetermined number of agents in each
period.
≥0 (1.23)

≥0 (1.24) Notations:
Set of shifts, j = {1, 2, …}
≥0 ;∀ ∈ (1.25) Set of periods, k = {1, 2, …}
Set of days, d = {1, 2, …}
The purpose of this model is to assigned agents into Set of tasks, t = {1, 2, …}
cycles and work shifts. The objective function (1.1) is to
minimize the total overtime cost. Constraints (1.2) find Parameters:
number of agents for each task that covers the Number of agents who can perform task
predetermined number of agents in each period of the first t
day of the month, the other days of the month that have Number of required agents in period k of
labor supply from the previous day is determined by day d for task t
constraints (1.3). Constraints (1.4) limit the number of _ 1 if shift j covers period k of the current
agents that are assigned to tasks not to be more than the working day, 0 otherwise
number of available agents. For each work shift, number _ 1 if shift j covers period k of the next
of agents who are assigned into work shifts must be more
working day, 0 otherwise
than the number of required agents with constraints (1.5).
_ Overtime cost of shift j
The maximum and minimum number of working day of
all agents in a month are determined by constraints (1.6)
and (1.7), respectively. Constraint (1.8) limits range of the Decision Variables:
number of working days not to be more than 2. Each Number of agents assigned into shift j of
agent cannot be assigned more than one shift in a day with day d for task t
constraints (1.9). The number of overtime working days is
determined by constraints (1.10). The minimum number Mathematical Model:
of working day in a week is 5 shown in constraints (1.11).
2=∑ ∑ ∑ _ (2.1)
The number of consecutive working days in a week and ∈ ∈ ∈

every 7 days are limited by constraints (1.12) and (1.13). ∑ ∈ _ ≥ ;∀ ∈ , ∈ {1}, ∈ (2.2)
In a week, agents must be assigned to same cycle that is
defined by constraints (1.14) - (1.16), constraints (1.14) ∑ _ +∑ _ ≥
∈ ∈
force agents into the same cycle as the previous day,
constraints (1.15) force agents into the same cycle as the ;∀ ∈ , ∈ {2, … ,31}, ∈ (2.3)
first day of the week. Constraints (1.16) force agents into
the same cycle in a week when agents are not assigned ∑ ∈ ≤ ;∀ ∈ , ∈ (2.4)
into cycle of the first day of the week. For days in the next
≥0 ;∀ ∈ , ∈ , ∈ (2.5)
week, agents can be assigned to the same or next cycle as
in constraints (1.17). Constraints (1.18) force to assign
workers only within a month. Constrains (1.19) define The purpose of this model is to find the minimum number
cycles of agents who are assigned into shifts. Constraints of agents for each shift across a month. Therefore, the
(1.20) and (1.21) assign decision variables to be binary. objective function (2.1) is to minimize the total overtime
Finally, decision variables are imposed to be positive cost due to shift assignment. Constraints (2.2) and (2.3)
values by constraints (1.22) - (1.25). find the number of agents for each task that covers the
predetermined number of agents in each period.
The proposed model took up to 39719 seconds to solve an Constraints (2.4) limit the number of agents that are
instance with only 10 agents. A real size problem with assigned to tasks not to be more than the number of
179 agents could not be solved. Hence, we decompose the available agents. Finally, decision variables are imposed
problem into two parts which are solved sequentially. to be positive values by constraints (2.5).
First is to determine the minimum number of required
agents for each shift across a month that covers the From this model, we obtained the minimum number of
predetermined number of agents in each period. The required agents for each tasks, each shift across a month
outcomes from the first part are used as input for the that covers the predetermined number of agents in each
second part which is to assign agents into work shifts. period. The outcomes from this part are used as input for
Mixed integer programming models are proposed to find the second part which is to assign each individual agent
solutions for both parts. into cycles and work shifts.

82
 Proceedings of the 2017 IEEE IEEM

B. Part II: Staff scheduling ≥ ∑ ∈ ∑ ∈ ∑ ∈ − ;∀ ∈ (3.7)

The purpose of this model is to assigned agents into ∑ ∈ ∑ ∈ ≥5 ;∀ ∈ , ∈ {1, … ,4} (3.8)
cycles and work shifts in order to minimize the daily
overtime cost and also satisfy working demand. The ∑ ∈ ∑ ∈ ≤6 ;∀ ∈ , ∈ (3.9)
outcomes from the previous model (Part I), the number of ∑ ∑ + ∑ ∑ ≤6
∈ ∈
agents that is assigned into shifts ( ), are used as input
( _ ) for this model. However, model in Part I ;∀ ∈ , ∈ {2, … , 5}, ∈ {2, … , 7 (3.10)
uses index d as days in a month but model in Part II
defines them as day r in week q, since there are working ≤1− ∑ ( ) ;∀ ∈ , ∈ , ∈ , ∈ {2, … ,7} (3.11)
constraints for days in a week and days in the next week.
≤1−∑ ;∀ ∈ , ∈ , ∈ , ∈ {2, … ,7} (3.12)
Moreover, this model solves each task independently so
index t is removed from the model. ∑ ≥ ;∀ ∈ , ∈ , ∈ , ∈ {4, … ,7} (3.13)

Notations: ( ) +∑ ≤ 1 ; ∀ ∈ , ∈ {2, . . ,4}, ∈ {2, . . ,5} (3.14)

Set of agents, i = {1, 2, …}
Set of shifts, j = {1, 2, …} ∑ ∈ =0 ;∀ ∈ , ∈ {5}, ∈ {4, … ,7} (3.15)
Set of cycles, c = {1, 2, …}
Set of weeks in a month, q = {1, 2, …5} ∑ ∈ = ;∀ ∈ , ∈ , ∈ , ∈ (3.16)
Set of days in a week, r = {1, 2, …7}
∈ {0,1} ;∀ ∈ , ∈ , ∈ , ∈ (3.17)

Parameters: ≥0 ;∀ ∈ , ∈ , ∈ , ∈ (3.18)
_ Number of required agents in shift j of
week q, day r ≥0 (3.19)
Minimum number of working day in a
month ≥0 (3.20)
_ Overtime cost of shift j ≥0 ;∀ ∈ (3.21)
_ Daily overtime cost of agent i
1 if shift j is a subset of cycle c, The purpose of this model is to assigned agents into
0 otherwise cycles and work shifts. The objective function (3.1) is to
minimize the total overtime cost. Constraints (3.2)
Decision Variables: determine the number of agents that covers the
1 if agent i is assigned into cycles c of predetermined number of agents in each work shift. The
week q, day r, 0 otherwise maximum and minimum number of working day of each
1 if agent i is assigned into shift j of agent in a month are determined by constraints (3.3) and
week q, day r, 0 otherwise (3.4), respectively. Constraint (3.5) limits range of the
Maximum number of working day of all number of working days not to be more than 2. Each
agents in a month agent cannot be assigned more than one shift in a day with
Minimum number of working day of all constraints (3.6). The number of overtime working days is
agents in a month determined by constraints (3.7). The minimum number of
Number of overtime working day in a working day in a week is 5 shown in constraints (3.8).
month of agent i Constraints (3.9) - (3.10) limit the number of consecutive
working days. In a week, agents must be assigned to same
Mathematical Model: cycle that is defined by constraints (3.11) - (3.13) and for
days in the next week, agents can be assigned to the same
3= or next cycle as in constraints (3.14). Constraints (3.15)
∑∈ ∑ ∈ ∑ ∈ ∑ ∈ _ +∑∈ _ (3.1) force to assign workers only within a month. Constrains
(3.16) define cycles of agents who are assigned into shifts.
∑∈ ≥ _ ;∀ ∈ , ∈ , ∈ (3.2)
Constraints (3.17) assign decision variables to be binary.
Finally, decision variables are imposed to be positive
≥ ∑ ∑ ∑ ;∀ ∈ (3.3)
∈ ∈ ∈
values by constraints (3.18) - (3.21).
≤ ∑ ∈ ∑ ∈ ∑ ∈ ;∀ ∈ (3.4)

− ≤2 (3.5)

∑ ∈ ≤1 ;∀ ∈ , ∈ , ∈ (3.6)

83
 Proceedings of the 2017 IEEE IEEM

III. RESULTS AND DISCUSSION IV. CONCLUSION

This section shows the results from agent scheduling In this paper, agent scheduling of call center is
of call center in a real setting. First, the minimum number studied. As a personal cost is expensive, the minimum
of required agents for each shift is determined. Then, the number of agents is determined. Then, the agents are
agents are assigned into work shifts. assigned into work shifts. Working constraints including
work duration, worker availability, break requirements,
In this problem, there are 179 agents. The minimum and overtime are considered. The purpose of this research
number of required agents for each shift is determined by is to find the optimal staff schedule in order to minimize
model in Part I. Then, each agent is assigned into shift by labor related cost. The main cost consists of salary, hourly
model in Part II. These proposed models are solved by overtime cost, and daily overtime cost. We decompose the
CPLEX, using a computer with core i5, 1.8 GHz with
problem into two parts. Mixed integer programming
turbo boost up to 2.7 GHz and 8 GB memory. Cost
models are proposed for both parts. Proper assignment
comparison between our method and the company method
can reduce the labor related cost. The complexity of the
that a bank used are shown in Table I. The company
method is assigning workers based upon experience problem has an effect on the computational time. This
planner. For the proposed method with proper assignment, method works well for this problem size. However, once
there is no overtime required resulting in only salary. The problem size increases, other methodologies should be
company method uses the same number of agents, but explore, which becomes our future research.
without good assignment, it results in overtime cost.

However, using the current number of agents may be REFERENCES

more expensive than reducing them replacing by
overtime. Therefore, finding the optimal number of agents [1] Cordone, R., et al., “Optimization of Multi-skill Call
is needed. As in Table II, overtime cost increases when Centers Contracts and Work-shifts”, Service Science, vol. 3,
no. 1, pp. 67-81, 2011.
the number of agents is reduced. Meanwhile, the total cost [2] Ertogral, K. and B. Bamuqabel, “Developing staff
also decreases until 162 agents. The number of agents schedules for a bilingual telecommunication call center
lower than 162 gives more total labor related cost. with flexible workers”, Computers & Industrial
Therefore, the optimal number of agents for this problem Engineering, vol. 54, no. 1, pp. 118-127, 2008.
is 162. Noted that, using lower than 160 agents cannot [3] Bhulai, S., G. Koole, and A. Pot, “Simple Methods for Shift
satisfy demand. Computational time depends on the Scheduling in Multiskill Call Centers”, Manufacturing &
complexity of the problem. The number of agents has an Service Operations Management, vol. 10, no. 3, pp. 411-
effect on the computational time of model in Part II. 160 420, 2008.
agents require more computation time than 179 agents [4] Ingolfsson, A., et al., “Combining integer programming and
the randomization method to schedule employees”,
because of tightening between the number of agents and European Journal of Operational Research, vol. 202, no. 1,
demand. pp. 153-163, 2010.
[5] Avramidis, A.N., et al., “Optimizing daily agent scheduling
TABLE I in a multiskill call center”, European Journal of
Cost comparison between the proposed and current methods Operational Research, vol. 200, no. 3, pp. 822-832, 2010.
[6] Kabak, Ö., et al., “Efficient shift scheduling in the retail
Number of Cost sector through two-stage optimization”, European Journal
agents Overtime Salary Total of Operational Research, vol. 184, no. 1, pp. 76-90, 2008.
Proposed method 179 0 2685000 2685000 [7] Cordone, R., et al., “Optimal selection of contracts and
Company method 179 16900 2685000 2701900
work shifts in multi-skill call centers”, EURO Journal on
Computational Optimization, vol. 2, no. 4, pp. 247-277,
TABLE II 2014.
The optimal number of agents and its total cost. [8] Taskiran, G.K. and X. Zhang, “Mathematical models and
solution approach for cross-training staff scheduling at call
Number Computational time centers”, Computers & Operations Research, 2016.
Cost
of (sec)
[9] Cuevas, R., et al., “A mixed integer programming approach
agents Overtime Salary Total Model 1 Model 2
to multi-skilled workforce scheduling”, Journal of
179 0 2685000 2685000 21 580
Scheduling, vol. 19, no. 1, pp. 91-106, 2015.
178 0 2670000 2670000 21 701
176 0 2640000 2640000 21 1152 [10] Ağralı, S., Z.C. Taşkın, and A.T. Ünal, “Employee
174 0 2610000 2610000 21 1237 scheduling in service industries with flexible employee
172 0 2580000 2580000 21 1230 availability and demand”, Omega, vol. 66, pp. 159-169,
170 0 2550000 2550000 21 1191 2017.
168 0 2520000 2520000 21 1242
166 0 2490000 2490000 21 1328
164 2750 2460000 2462750 21 1285
162 10350 2430000 2440350 21 1369
160 45772 2400000 2445772 21 2381

84