Professional Documents
Culture Documents
7 分时间段-目的:省钱
7 分时间段-目的:省钱
DOI: 10.1177/1938965509333557
Volume 50, Issue 2 155-167
Scheduling Restaurant
Workers to Minimize
Labor Cost and Meet
Service Standards
by KYUWAN CHOI, JOHYE HWANG, and MYOUNGJU PARK
In labor scheduling, restaurant managers face dual chal- existing schedule, the IP-generated schedule helped the
lenges: overstaffing that means excessive labor costs restaurant reduce overall labor costs while ensuring
and understaffing that invites the opportunity cost of appropriate service levels.
service errors and lost business. Most operators seek to
rely on their personal experience and judgment to deter- Keywords: restaurant scheduling; cost minimiza-
mine schedules that are meant to maintain service qual- tion; full-time and part-time employee
ity and limit labor costs, but an analysis of one Korean ratios; labor scheduling; overstaffing;
restaurant finds that this goal is elusive. The labor sched- service quality; understaffing; Korea;
uling model based on integer programming (IP) is pre- chain restaurants
sented to address the staffing and scheduling issues and
E
service standards by maintaining an appropriate ratio of
part-time to full-time employees. Using appropriate con- ffective labor scheduling has always been a
straints, the model was used to generate a schedule for concern for restaurant managers. This article
a table service restaurant in Seoul, South Korea, that is focuses on the labor-capacity management
affiliated with a global chain. When compared to the problems confronted by a single unit of a multiunit
global chain restaurant located in down- principle that full-time servers have more
town Seoul, South Korea. The labor- expertise). Thus, any schedule should
scheduling model proposed in this article maintain a reasonable proportion of full-
seeks to minimize labor costs while ensur- and part-time employees (Mabert and
ing that customer demand is satisfactorily Raedels 1977; Bechtold 1988).
met. Restaurant managers are only too The restaurant in this study has encoun-
aware of the problems inherent in schedul- tered the same problem of low profitabil-
ing, with its twin dangers of underschedul- ity as have other global chain restaurants
ing and overscheduling. and, thus, must focus carefully on labor as
Scheduling issues complicate the com- part of its cost control strategy. Setting
petitive challenges of the Korean market. appropriate schedules has been recognized
The profitability of casual dining restau- as the most effective way of controlling
rants in global chains has diminished labor costs (Thompson 2003). For this
greatly since 2000, despite increased mar- selected restaurant, we sought to address
keting efforts. For example, the 2006 profit the staffing and scheduling issues over
margin of Korea’s Bennigans was –4.47 both the week (or the number of consecu-
percent; Sizzler, –4.23 percent; and TGI tive days worked, which we call the tour)
Friday’s, –3.08 percent. Those negative and each day (that is, the shift). We further
margins were below the average profit wanted to establish a schedule that would
margin of 6.54 percent for other audited identify the correct number of full- and
Korean restaurant chains for that year and part-time employees, when each type of
far below the 20 percent average profit employee should be scheduled, and the
margins experienced by such restaurants shift to which each should be scheduled.
in the late 1990s (Choi et al. 2007). Those Finally, we ran the resulting schedule in
negative results may result in part from the our test restaurant to assess the cost reduc-
inflexibility of the chains’ operating sys- tions arising from a schedule that adopted
tems. Restaurant companies free from our constraints.
such ironclad contracts generate higher
profits than do those held to such contracts Literature Review
(Choi et al. 2007; Klonowski, Power, and Numerous studies have been conducted
Linton 2008). Additionally, in Korea, on labor scheduling (Beaumont 1997;
global chain restaurants’ low profitability Bechtold, Brusco, and Showalter 1991;
is also influenced by particularly high Bechtold and Jacobs 1990; Brusco and
labor costs due, in part, to an increase in Jacobs 1993, 1998; Easton and Rossin 1991;
employee benefits mandated by Korean Goodale and Tunc 1998; Li, Robinson, and
labor regulations. Mabert 1991; Loucks and Jacobs 1991;
The restaurant industry makes heavy Thompson 1990, 1993, 1995a, 1995b, 2004).
use of part-time employees because of its Most studies have focused on scheduling in
budget constraints and irregular operating other industries, including airline crews
hours (Love and Hoey 1990). Although (Arabeyre et al. 1969; Bodin et al. 1983;
part-time employees help create flexible Gamache and Soumis 1998), nurses in hos-
capacity, they also mean a more compli- pitals (Aickelin and White 2004; Bradley
cated scheduling task. Moreover, the pro- and Martin 1991), and service clerks in call
portion of part-time employees to full-time centers (Mehrotra 1997). We have seen few
employees in a restaurant may interfere studies investigating labor scheduling in hos-
with maintaining service quality (on the pitality services. Among the few, Love and
restaurant work shifts vary from a few experience for setting schedules. Using
hours to as much as eight to ten hours in managers’ judgment regarding necessary
length, restaurants hire more part-time staffing levels, the restaurant used an Excel
employees than other service industries spreadsheet to match employees with
(Love and Hoey 1990), and restaurants shifts. The resulting schedule was posted
must maintain service standards in the on a dry-erase board hung on the wall.
face of those uncertainties. We also wanted What we observed is that these less than
to incorporate the concept raised by Li, scientific methods have resulted in fre-
Robinson, and Mabert (1991) regarding quent overtime among servers, a decline in
scheduling of days on and days off. The servers’ morale, and high labor costs (due
restaurant that is the focus of this study is to payment of overtime). This approach
located in the business district in down- also absorbed considerable time, as the
town Seoul, Korea. As we mentioned at managers attempted to schedule employ-
the outset, the restaurant is a globally ees fairly and sought to reduce the number
branded, multiunit chain restaurant that of complaints by servers about last-minute
has been operating for eight years. schedule changes during the restaurant’s
We consider this restaurant to be repre- busiest hours.
sentative of all the units in the chain. Like
all restaurants, this one requires effective Labor Scheduling Model Using
labor scheduling to control its labor cost Integer Programming
percentage. Standing at 31.56 percent at the In developing a labor scheduling model
time of our study, the restaurant’s labor cost to minimize the study restaurant’s labor
is slightly higher than the average labor cost costs, we particularly sought to identify
percentage of the other chain restaurants which full- and part-time servers should
(29.01 percent) and well within the range of be assigned to each given work shift. This
chain restaurants in Korea, which is 28 to model had to take into consideration work
40 percent (Korea Food Service Information shifts of various lengths of time, and it had
2006). Because it is located in a downtown to account for days on and days off work—
business district, the restaurant has a more all within an overall rubric of customer
variable demand pattern than do similar service standards.
units located in suburban areas. Considering The model we developed uses integer
the challenges of scheduling with variable programming (IP). Although similar to
demand, we determined that this restaurant linear programming, IP considers decision
was a good candidate for receiving the ben- variables to be integers. This makes sense
efit of effective labor scheduling. because the number of employees must be
The restaurant’s workforce consists of an integer (since one cannot schedule 1.5
twenty-six servers (seven full-time and people). IP uses the objective function and
nineteen part-time employees) and twen- the constraints shown in Exhibit 1, as we
ty-three cooks (all full-time). The majority explain next.
of customers during the week are white- The objective (1) in the model expresses
collar employees who work in the sur- the goal of minimizing total labor cost
rounding downtown area, which causes over the scheduling horizon (one week).
congestion during lunch hours. Weekend The first constraint (2) ensures sufficient
business is oriented more toward families. staffing for each day; the next constraint
As with most Korean restaurants, this (3) ensures meeting the minimum staffing
restaurant relies heavily on managers’ requirements for each day; and the final
Exhibit 1:
Labor Scheduling Model Developed Using Integer Programming
(1) Min
iS , jT
ci , j xi , j d i , j yi , j ei zi
iS
subject to
(2)
iPk , jDl
( xi , j yi , j al zi ) max{bk ,l , m} for k P, l T
(3) (1 r )
iSk , jDl
xi , j r
iSk , jDl
( yi , j al zi ) for k P, l T
where
constraint (4) is nonnegative and integral for example, Thompson (1997) used a
numbers. range of ratios for full- and part-time
The model makes the following assump- employees. While many restaurants have
tions. The ratio of full- to part-time employ- excellent part-time servers, most part-
ees is 6:4, as determined by management timers consider the job to be temporary
based on their observation of the most effec- and do not have the professional experi-
tive mix needed to ensure high-quality ence of full timers. Our model thus assumes
service. We agree that this ratio could be that the full-time employees have greater
considered subjective, but other research productivity levels than do the part-time
supports the idea that it is desirable to con- employees and are paid higher wages. It is
trol the proportion of part-time employees also assumed that employees are homoge-
in the mix of labor to provide high-quality neous within their job categories.
service (Mabert and Raedels 1977; With regard to employee tours, our
Bechtold 1988). In his scheduling study, model also assumes that each employee
Exhibit 2:
Shift Type and Service Periods
Time (hour) 8 12 14 21 23
Shift I (8~14)
Shift II (12~21)
Service periods
P1 P2 P3 P4
Exhibit 3:
Maximum Hourly Customer Counts during Each Service Period (Unit: Count)
works five consecutive days. For example, (1) forecasting demand, (2) translating the
an employee whose tour starts on Monday demand forecasts into employee require-
works through Friday for five consecutive ments, (3) comparing alternate labor sched-
days, but we note that not all employees ules, and (4) implementing and evaluating
who work on Monday started their tour on the new schedule.
that day. They could have started as early
as Thursday, for instance.
Forecasting Demand
Our model divides the day into the
(Number of Customers)
following four nonsymmetical service peri-
ods: P1 (8:00^12:00), P2 (12:00^14:00), For the first step in labor scheduling, we
P3 (14:00^21:00), and P4 (21:00^23:00) identified the number of customers to serve
(Exhibit 2). Finally, the minimum number of in the scheduling month based on Point of
employees permitted on any shift is two. Sale (POS) data. Because January repre-
sents a typical business cycle for this res-
Labor Scheduling in taurant, we analyzed four weeks of sales
the Restaurant data from January 2007. We then calcu-
This study follows the framework lated the maximum number of customers
suggested by Thompson (1998a, 1998b, served each hour for each service period
1999a, 1999b). As noted, the labor sched- each day (Exhibit 3). While the peak hour
uling framework is composed of four steps: for lunch (P2) is busy enough, the peak
hour for the afternoon shift (P3) is even schedule would look like based on the
busier, particularly late in the week. We customer count per server, without the IP
identified these peak hours on the assump- parameter and its various constraints.
tion that if our schedule can satisfy demand (Needless to say, the CCS approach does
during the peak hour of a service period, not schedule fractional employees, but the
then the schedule can maintain service IP approach adds constraints not found in
standards in all time periods. Using the the CCS approach.) We can then compare
data of the number of customers served that schedule with an IP schedule with the
(customer count) during each service goal of minimizing labor costs while also
period, we calculated the number of satisfying the constraint of applying the
employees required during each service appropriate ratio of full- and part-time
period within each tour day. servers on each work shift. After creating
these two new schedules (one based on the
Identification of Requirements CCS and one applying the IP constraints)
(Number of Employees Required) we compared them to each other and to the
restaurant’s original schedule.
To provide high-quality service, our Our first comparison is the CCS sched-
restaurant needed to maintain an appropri- ule to the existing schedule (Exhibits 4
ate level of employees, given the customer and 5). In the CCS-generated schedule, the
count. To establish the correct ratio of cus- number of employees required for service
tomers and employees, we first calculated period 1 on Monday is two, but the man-
the customer count per server (CCS) dur- agement has been scheduling five people
ing the peak hour, simply by dividing the in this breakfast period. The reverse is
number of customers served in that peak occurring on Tuesday’s lunches (service
hour by the number of employees. For the period 2), when the CCS schedule brings
remainder of the schedule, we set a CCS in eleven servers, but the existing schedule
standard equal to 95 percent of the peak- has eight people. This pattern continues
hour CCS. Based on the average of this generally throughout the week. The sched-
CCS calculation for the study period, the ule based on the CCS model identified
restaurant’s standard CCS is 5.5. The for- overstaffing during service periods 1 and 4
mula used to calculate the standard CCS is and understaffing during service periods 2
as follows: and 3. As we discuss more below, these
! " occurrences of both understaffing and
Y
$$4 E DJ >TJ 6 overstaffing accounted for a significant
JE number of service failures that translated
into additional costs (both in labor and
where Ci(Si) is the number of customers product) to the restaurant.
(servers) during the peak hour in day i and We gained even further refinement by
A is the service level. applying the integer programming conven-
tion. The differences between the two
Comparison of Alternatives (IP Model schedules resulted from differential con-
versus Existing Model) straints applied by each method. The IP
method takes into consideration all the
As we said at the outset, our final model requirements of the various employee cat-
will apply an integer programming egories, as well as all the constraints that
approach, but first let us examine what the we mentioned above (e.g., consecutive
Exhibit 4:
Comparison of the Existing Schedule with the Customer Count per Server (CCS)–
Generated and Integer Programming (IP)–Generated Schedules
Existing schedule
P1(8:00^12:00) 5 4 4 5 6 5 4
P2(12:00^14:00) 9 8 9 11 11 10 10
P3(14:00^21:00) 9 10 11 10 10 11 11
P4(21:00^23:00) 5 6 6 4 5 6 6
Schedule based on CCS
P1(8:00^12:00) 2 3 3 2 2 1 1
P2(12:00^14:00) 8 11 11 11 12 9 7
P3(14:00^21:00) 7 10 11 12 12 12 10
P4(21:00^23:00) 3 3 5 5 7 4 3
Schedule based on
integer programming:
total number of
employees (full-time/
part-time)
P1(8:00^12:00) 4 (3/1) 4 (3/1) 5 (3/2) 5 (3/2) 6 (4/2) 3 (2/1) 3 (2/1)
P2(12:00^14:00) 8 (5/3) 11 (7/4) 11 (7/4) 11 (7/4) 12 (8/4) 9 (6/3) 8 (5/3)
P3(14:00^21:00) 7 (5/2) 10 (7/3) 11 (7/4) 12 (8/4) 13 (9/4) 12 (8/4) 10 (6/4)
P4(21:00^23:00) 3 (3/0) 3 (3/0) 5 (3/2) 6 (4/2) 7 (5/2) 6 (4/2) 5 (3/2)
five-day schedule for each employee, ratio example, service period 2 on Monday
of full- and part-time servers). As a result, requires eight employees—five full-time
the IP-generated model produced a more and three part-time, which meets the con-
effective schedule than did the model straint of the proportion of part-time
based on only CCS (Exhibits 4 and 5), not employees to the total number of employ-
because the IP model uses fewer employ- ees during each shift. Without this con-
ees than CCS, but because the IP schedule straint, by contrast, a schedule focused
maintains service standards and also takes only on minimizing labor cost (while still
into account employee tours. maintaining CCS standards) would bring
In the IP-generated schedule, Monday, in no full-time employees.
which has the lowest customer count, The IP schedule also reveals the effects
requires the lowest number of employees of the constraint of a five-day consecutive
(twenty: thirteen full-time and seven part- tour for each employee. Most notably, in
time employees); and Friday, which has contrast to the original schedule, full-time
the highest customer count, requires the employees do not necessarily start their
highest number of employees (thirty-five: tours on Monday. Among the three full-
twenty-two full-time and thirteen part- time employees on shift 1, for instance,
time employees). Perhaps the greatest two full-time employees’ tours start on
advantage of the IP-generated schedule is Monday, but the other full-time employ-
that it shows a detailed distribution of full- ee’s tour starts on Wednesday and ends on
and part-time employees during each work Sunday (instead of Monday through
shift as well as each service period. For Saturday, as laid out by the restaurant’s
Exhibit 5:
Comparis ons of Overstaffing and Understaffing of the Schedule Based on Integer
Programming (IP), the Old Schedule, and the Schedule Based on Customer Count
per Server (CCS)
weekdays
12
number of employees
10
8
6
4
2
0
service period
weekends
12
number of employees
10
8
6
4
2
0
service period
existing schedule; Exhibit 6). According 5.50 percent decrease in labor costs
to the optimal schedule generated using (Exhibit 7), primarily because the new
the IP method, no full-time employees schedule decreased in number of servers
start their tour on either Tuesday or on from twenty-six to twenty-one. In addition
Thursday. A similar pattern is also observed to the reduction in staff, the restaurant also
for the schedule of part-time employees, saved the opportunity cost of service fail-
with part-time employees’ tours starting ures from understaffing. The cost incurred
on different days. For example, part-time due to overstaffing was calculated by mul-
employees who work shift 1 begin their tiplying the number of hours overstaffed
tours on either Monday or Wednesday. by the hourly wage, which we calculated
Additionally, according to the IP-generated as a weighted average of full- and part-
schedule, separate scheduling for week- time employees’ wages. The weighted
days and weekends is not necessary. average is 5,100 won, calculated as fol-
lows: (5,500 won [the wage for full-time
Evaluation of the System Implemented employees] s 0.6) (4,500 won [the part-
time wage] s 0.4). Overall, the cost
We were able to test this schedule in incurred due to overstaffing dropped by
operation at the restaurant. When the new 3.16 percent.
IP-generated schedule was implemented at Although the cost of understaffing is a
the study restaurant, the restaurant saw a real cost, it is more complicated to calculate
Exhibit 6:
Number of Servers Scheduled Based on the Optimal Schedule Resulting from
Integer Programming
Shift 1 (8:00−14:00)
Full-timer 2 2 2 2 2
1 1 1 1 1
1 1 1 1 1
Part-timer 1 1 1 1 1
1 1 1 1 1
Total 4 4 5 5 6 3 3
Shift 2 (12:00−21:00)
Full-timer 2 2 2 2 2
1 1 1 1 1
1 1 1 1 1
1 1 1 1 1
Part-timer 1 1 1 1 1
1 1 1 1 1
1 1 1 1 1
1 1 1 1 1
Total 4 7 6 6 6 6 5
Shift 3 (14:00−23:00)
Full-timer 1 1 1 1 1
1 1 1 1 1
1 1 1 1 1
1 1 1 1 1
1 1 1 1 1
Part-timer 2 2 2 2 2
Total 3 3 5 6 7 6 5
Total
Full-timer 8 10 10 11 13 10 8
Part-timer 3 4 6 6 6 5 5
Total servers 11 14 16 17 19 15 13
this value. The primary challenge is to method was used by Davis (1991) in his
place a direct monetary value on poor ser- study of waiting cost. (Davis [1991] calcu-
vice. According to Thompson (1997), an lated waiting cost by multiplying the min-
accurate calculation of the cost of under- utes that customers wait by the employee’s
staffing, as a cost of poor service, must wage per minute.) For our study, we calcu-
include a measure of the true impact of lated and compared the costs of under-
staff size changes on customer service, staffing in the original schedule and in the
and not a surrogate measure of the impact. IP-generated schedule using this simple
The simplest way to calculate the cost of calculation method. According to this cal-
understaffing is to multiply the number of culation, the restaurant reduced its costs
hours understaffed compared to the require- due to understaffing by 5.78 percent by
ment and the hourly wage. A version of this using the IP-generated schedule.
Exhibit 7:
Improvement in Cost Reduction by Optimal Scheduling Using Integer
Programming
Goodale, J. C., and Enar Tunc. 1998. Tour scheduling with Thompson, Gary M. 1990. Shift scheduling in services
dynamic service rates. International Journal of when employees have limited availability: An L.P.
Service Industry Management 9 (3): 226-47. approach. Journal of Operations Management 9 (3):
Goodale, J. C., R. Verma, and M. E. Pullman. 2003. A 352-70.
market utility-based model for capacity scheduling ———. 1993. Accounting for the multi-period impact of
in mass services. Production and Operations service when determining employee requirements
Management 12 (2): 165-85. for labor scheduling. Journal of Operations
Hueter, Jackie, and William Swart. 1998. An integrated Management 11 (3): 269-87.
labor-management system for Taco Bell. Interfaces ———. 1995a. Improved implicit optimal modeling of the
28 (1): 75-91. labor shift scheduling problem. Management Science
Keith, E. G. 1979. Operator scheduling. AIIE Transactions 41 (4): 595-607.
11 (1): 37–41. ———. 1995b. Labor scheduling using NPV estimates of
Klonowski, Darek, Jaqueline L. Power, and Daniel Linton. the marginal benefit of additional labor capacity.
2008. The development of franchise operations in Journal of Operations Management 13 (1): 67-86.
emerging markets: The case of a Poland-based res- ———. 1997. Labor staffing and scheduling models for
taurant operator. Cornell Hospitality Quarterly 49 controlling service levels. Naval Research Logistics
(4): 436-49. 44:719-40.
Korea Food Service Industry Year Book. 2006. Korea ———. 1998a. Labor scheduling, part 1. Cornell Hotel and
Food Service Information, Co. Ltd: 136-160. Restaurant Administration Quarterly 39 (5): 22-31.
Li, Cheng, E. Powell Robinson, and Vincent A. Mabert. ———. 1998b. Labor scheduling, part 2. Cornell Hotel
1991. An evaluation of tour scheduling heuristics and Restaurant Administration Quarterly 39 (6):
with differences in employee productivity and cost. 26-37.
Decision Sciences 22 (4): 700-718. ———. 1999a. Labor scheduling, part 3. Cornell Hotel
Loucks, John S., and F. Robert Jacobs. 1991. Tour sched- and Restaurant Administration Quarterly 40
uling and task assignment of a heterogeneous work. (1):86-96.
Decision Sciences 22 (4): 719-38. ———. 1999b. Labor scheduling, part 4. Cornell Hotel
Love, Robert R., Jr., and James M. Hoey. 1990. and Restaurant Administration Quarterly 40 (3):
Management science improves fast-food operations. 85-96.
Interfaces 20 (2): 21-29. ———. 2003. Labor scheduling: A commentary. Cornell
Mabert, V. A., and A. Raedels. 1977. The detail scheduling Hotel and Restaurant Administration Quarterly 44
of a part-time work force: A case study of teller staff- (5/6): 149-55.
ing. Decision Sciences 8:109-20. ———. 2004. Planning-interval duration in labor-
Mehrotra, V. 1997. Ringing up big business. OR/MS Today shift scheduling. Cornell Hotel and Restaurant
24 (4): 18-24. Administration Quarterly 45 (2): 145-57.
Kyuwan Choi, Ph.D., is an assistant professor in the College of Hotel and Tourism Management, Kyung
Hee University, Seoul, Korea (kwchoi@khu.ac.kr).
Johye Hwang, Ph.D., is an assistant professor in the Hotel and Restaurant Management program at the
University of Missouri (hwangj@missouri.edu).
Myoungju Park is a doctoral candidate in the department of Industrial Engineering at Seoul National
University (pmj0684@snu.ac.kr).