Case Study: A DSS DESIGN FOR WORKFORCE MANAGEMENT IN CALL CENTERS

1
Case StudyA DSS DESIGN FOR WORKFORCE
MANAGEMENT IN CALL CENTERS
MIS 463 Decision Support Systems for Business
An MA Thesis Study by Burak Gedikoglu Department
of MIS April 06
2
Outline

• WFM in Call Centers
• Background and Recent Literature Review
• The DSS Design for WFM
• LP Model
• Simulation Model
• Illustration with the Company Data
• Conclusion

3
Workforce Management Systems

• Computer programs, software and analysis tools
  used for
• Demand forecasting
• Labor staffing
• Shift scheduling
• Staff allocation to the shifts
• Most of the existing WFM tools use Erlang C
  calculations to find out the minimum number of
  staff required.

4
Shortcomings of Erlang C

• Erlang C uses the basic M/M/s queuing model and
  this
• model is based on the assumptions listed below
• The call arrivals have a Poisson distribution
• Agent service times have an exponential
  distribution
• A fixed number of agents are available
• All servers are identical, having the same skill
  and service time distributions
• Queuing rule is First Come First Served (FCFS)
• The service process consists of a single phase
• Queue length is unlimited and no balking or
  reneging occurs

5
Recent Literature

• Thompson, G. M. (1997)
• Two new models of the labor staffing and
  scheduling problems that overcome the limitations
  of existing models
• Green, L. V., Kolesar, P. J., Soares, J. (2001)
  
• SIPP (Stationary Independent Period by Period)
  approach
• Ingolfsson, A., Cabral, E., Wu, X. (2002)
• Integer programming and the randomization method
  to schedule employees
• Atlason, J., Epelman, M. Henderson, S.G. (2004)
• Cutting plane method to optimize the scheduling
  of agents
• Cezik and LEcuyer (2005), Koole and Pot (2005)
• Multi-skill call centers

6
The DSS Design For WFM

• The proposed design integrates a mathematical
  optimization model with a computer simulation
  model.
• Use of computer system simulation as a supportive
  model is not a common method in the literature.
• Simulation is also more reliable tool than Erlang
  C for what if analysis.

7
Company Overview

• Company A operates in mainly three locations.
• Company A makes call forecasts for 15 minute
  intervals in a week.
• Minimum number of required agents for each
  15-minute periods is calculated by using Erlang
  formula.
• Duration of a shift is 8hrs/day and an agent can
  work in a single shift in a day.
• Customers are segmented into four groups I1, I2,
  I3 and I4
• Over 100 different services under three main
  headings Routine calls, Regular campaign calls
  and Temporary campaign calls.

8
Problem Definition

• Allocating the calls between the locations that
  have agent and seat capacities, while attaining
  the target service and utilization levels is a
  major problem.
• Try shifts with flexible durations, say 4 to 8
  hrs?
• - Duration, start and finish time of each shift?
  (Shift scheduling problem)
• - Number of agents in each shift?
• (Staff allocation problem)

9
Linear Programming Model

• Decision Variables
• Gi The number of employees that start to work at
  the beginning of slot i, i 800,, 2200 in a
  day.
• Qj The number of employees that end work at the
  end of slot j, j1100,, 2400, 0100 in a day.
• Xij The number of agents allocated to the shift
  which starts at the beginning of slot i and ends
  at the end of slot j, i 800,, 2200 and t
  1100, ..2400, 0100.

10
LP Objective Function

• Labor costs constitute the major part of the
  operational costs in Company A. For this reason
  the
• objective is to minimize the number of agents
  working in a day.

• Minimize Total Number of Allocated Agents

11
LP Constraints

• 1. Minimum workforce requirements should be met
  in each slot.

• Min. staff requirement in slot k, k800,
  ..,0100

• 2. An agent should work for at least four hours.

• k800, 900,..., 2100

12
LP Constraints

• 3. An agent should work no more than eight hours.
  

• k 800, 900,..., 1700

• 4. At the end of the day, all agents should quit
  work.

• j 1100,, 2400, 100

13
LP Constraints

• 5. Total number of agents working in a period
  shouldn't exceed total seat
• capacity.

• Total seat capacity, k800,..., 2400, 100

• 6. Total number of agents allocated in a day
  shouldn't exceed agent
• capacity.

• Agent capacity

14
LP Constraints

• 7. Total number of agents who start working at
  the beginning of
• slot k should be equal to the sum of the agents
  allocated to the
• shifts that start at the beginning of slot k.

• k800,..., 2200

• 8. Total number of agents that quit at the end of
  slot j must be equal to
• the sum of the agents allocated to the shifts
  that end at the end of slot j.

• j 1100,..., 0100.

15
LP - Results

• Allocation of the agents to the
  shifts

16
LP - Results

• Hourly allocation of the agents in the proposed
  system

17
Simulation Model

• Constructed in Arena in order to monitor
• Service levels per period The percentage of
  calls handled within 20 seconds in a period.
• Abandonment rates per period The percentage of
  callers who hang up before their call is answered
  by an agent.
• Average speed of answer (ASA)The average delay
  in the queue experienced by a customer waiting
  for an agent.
• Utilization The percent of the available time of
  agents that is spent actually for handling
  incoming calls.

18
Simulation of the LP Solution

• Service levels on day 1

19
Simulation of the LP Solution

• Abandonment rates on day 1

20
Simulation of the LP Solution

• Utilizations

21
Simulation of the LP Solution

• Average speed of answer

22
Improvement by Simulation

• 1. An arriving call can be allocated to any of
  the three locations in accordance to the number
  of available agent capacities.
• 2. The first shift of the day in Location 3
  starts in 07.00 instead of 800 to avoid the
  significant decrease in the service levels
  between 0700 and 0800.
• 3. We allocate additional agents in some shifts
  in order to increase service levels.

23
Improvement by Simulation

• Allocation of additional agents to the
  shifts

24
Improvement by Simulation

• Service levels on day 1

25
Improvement by Simulation

• Abandonment rates on day 1

26
Improvement by Simulation

• Utilizations

27
Improvement by Simulation

• Average speed of answer in Scenario
  3

28
A Framework For The DSS Environment
29
Conclusion

• The aim of this study is to develop an easy to
  use environment by ARENA and MS-Excel that will
  support the existing tool and enhance its
  solution quality by
• generating valid results
• allowing real time assessment of numerous
  alternatives
• It is shown that integrating system simulation
  and linear programming models is a promising
  approach for solving shift design and staff
  allocation problems in call center management.
• As a future research, proposed DSS environment
  can be enhanced by user friendly interfaces for
  data input, model access and output analysis.
• Both models can be improved to allow more
  constraints like absenteeism skill based
  routing, part time workers, transportation of
  agents.