You are on page 1of 20

Decision Sciences

Volume 35 Number 2
Spring 2004
Printed in the U.S.A.

Cross-training Decisions in Field


Services with Three Job Types
and Server–Job Mismatch∗
Saligrama R. Agnihothri†
School of Management, Binghamton University, State University of New York,
Binghamton, NY 13902, e-mail: agni@binghamton.edu

Ajay K. Mishra
School of Management, Binghamton University, State University of New York,
Binghamton, NY 13902, e-mail: amishra@binghamton.edu

ABSTRACT
To be cost-effective, field service managers must balance the high cost of machine
downtime with the high cost of cross-training technicians in multiple skills. We study
a field service system with three job types requiring three different skills. Each server
has a primary skill, the cost of which is considered sunk, and up to two secondary skills,
which is a managerial decision. We model two important characteristics that distinguish
field services: server–job mismatch and the ratio of travel time to service time. We
use a queueing framework and simulation to study three cross-training decisions: the
number of servers cross-trained in secondary skills, the number of secondary skills each
server should have, and the efficiency in each secondary skill. We find that complete
cross-training is cost-effective in some field service situations. Typically, efficiency in
secondary skills must be close to 100%, but when the probability of mismatch is high
and the ratio of travel time to service time is high, efficiency in secondary skills must be
less than 100%.

Subject Areas: Cross-Training, Field Services, Service Operations, Simula-


tion, and Workforce/Staff/Manpower Planning.

INTRODUCTION
The quality of after-sales support directly impacts the success of companies pro-
ducing complex equipment. A major part of after-sales support is the installation,
maintenance, and repair of equipment at the customers’ site, which is referred to
as field service. Field service is important for machines in medical electronics,
telecommunications and information technology, industrial automation, building
controls, high-voltage power equipment, and office products. Machine downtime

∗ Our heartfelt gratitude to the associate editor, two anonymous referees, and the editor for extensive
and insightful suggestions that improved the paper immensely. We thank Professor Donald E. Simmons for
useful discussion on an early version of this paper.
† Corresponding author.

239
240 Cross-training Decisions in Field Services with Three Job Types

is usually of uncertain duration and decreases a customer’s productivity. Reducing


machine downtime by providing prompt and effective field services is paramount
for customer satisfaction. However, the timing of requests and the skill required for
field service is uncertain, leading to an inability to have the required field service
resources available when necessary.
An important factor that impacts downtime is the inability of the field ser-
vice department to accurately recognize the skill needed to repair the equipment
upon failure. This results in server–job mismatch. Customers cannot usually diag-
nose the problem or convey crucial information because of the complexities of the
failed machine. Dispatchers working with limited information from customers of-
ten make mistakes in judging the right skills required to repair the failed equipment.
“Call-centers have little incentive or training to diagnose problems correctly. Often,
the first option is to assign a field worker” (EDS, 2003). Such mismatch and the
resulting multiple technician visits reduce productivity and increase downtime and
customer dissatisfaction (Schapiro, 1998).
Another factor that is critical to reduce downtime is the immediate availability
of a technician (or server) with the required skill who can be dispatched to a
customer’s site. One way to reduce the chance of server–job mismatch and increase
server availability is to cross-train workers in multiple skills so that they can be
assigned to a variety of jobs as necessary. Since training is expensive and labor cost
is a major component of the service cost, maintaining an optimum workforce skill
mix is an important decision in field services. All prior studies on field service and
cross-training assume that the skill type required to repair a machine is accurately
determined before an appropriate technician is assigned and that the machine is
repaired in a single visit. To our knowledge, cross-training in field services in the
presence of server–job mismatch has not been studied and is a key distinguishing
factor of our study.
We use simulation to study a field service system with three different job
types requiring three different skills. We assume that all technicians are trained
to diagnose all jobs. In the existing system, each technician has the skill to repair
one job type, termed a primary skill with 100% efficiency, where the efficiency in
a skill is determined by the extent of training in that skill. The cost of this initial
training is a sunk cost for the purpose of cross-training decisions. A dedicated
server has training only in one (primary) skill. When all the servers are dedicated
we call it a dedicated system. A technician may also be cross-trained to acquire
secondary skills in one or both of the remaining two skill types with varying degrees
of efficiency. When all the technicians are cross-trained in all the secondary skills
at 100% efficiency, we refer to it as complete cross-training. We assume that the
cost incurred to develop the secondary skills such as training costs and additional
pay is proportional to the efficiency imparted in that skill and is charged in an
amortized manner. We call this cost the server premium for a cross-trained server.
Machine downtime comprises three phases: queue time (time between the arrival
of the job and the assignment of a technician), travel time for the technician,
and service time (time taken for diagnosis and repair). We use the total cost per
unit time, defined as the sum of service cost and downtime cost per unit time, as
the performance criteria. The total cost per unit time depends on three important
decisions:
Agnihothri and Mishra 241

r How many technicians should be cross-trained?


r In how many secondary skills must a technician be cross-trained?
r What should be the extent of training in secondary skills, measured here
by the efficiency of a technician in that skill?
To answer these questions, we need to evaluate the trade-off between the additional
cost due to cross-training and the savings due to reduced machine downtime. The
objective of this paper is to develop insights that aid field service managers in
answering the above questions in the presence of server–job mismatch.
In the next section we describe our field service model in detail. We subse-
quently review the relevant literature. In the following two sections, we explain
the simulation setup and present the results and analysis. In the final section, we
present the conclusions.

FIELD SERVICE MODEL


Model Assumptions
We develop a field service model that is representative of actual field service sys-
tems, with the following assumptions.
a. Call arrival process of an incoming job of type i requiring skill i follows
a stationary Poisson distribution with mean, λi , i = 1, 2, 3.
b. The system has a fixed service capacity of K technicians.
c. The level of training, a managerial decision variable, for technician j in
skill i gives an efficiency ηij (0 ≤ ηij ≤ 1). Actual time for repair is time
for repair with 100% efficiency divided by ηij . Each technician has one
primary skill in which the efficiency is 100% and one or two secondary
skills, with varying degrees of (positive) efficiency. If several jobs are
available, a technician chooses a job using the FCFS criterion. A job is
assigned to an idle technician with the highest efficiency in that skill.
d. Travel times follow a triangular distribution with mean T t . It is independent
and identically distributed across technicians and jobs. Time for diagnosis
and time for repair are each exponentially distributed. The resulting service
time has an Erlang distribution (with mean T s ). The mean travel time
relative to the mean service time (T t /T s ) is denoted by travel ratio.
e. There is a cost C per unit time (assumed nonnegative and identical for all
skills) for a technician’s primary skill. This can be considered as the sunk
cost that leads to the billing rate of a technician based on their training in
a primary skill. In addition, there is a nonnegative premium, P, linear in
efficiency, for training in a secondary skill and is identical for each skill. It
is charged in an amortized manner as Pηij per unit time. For all customers,
we assume a linear downtime cost, d per unit time incurred for the entire
duration a machine is down.
f. The dispatcher makes a mismatch between the job and the required skill
with probability φ. The technician realizes this at the end of the diagnosis
242 Cross-training Decisions in Field Services with Three Job Types

phase and judges that the job is one of the other two types; each have
a probability φ/2. If the assigned technician is cross-trained in the skill
required to complete the job, then the repair is completed. Otherwise, the
job is sent back to the appropriate skill’s queue but with a priority higher
than the jobs arriving for the first time. The next technician assigned to
the job completes the repair without further diagnosis.

The Objective Function


From the Little’s equation, the average number in the system is (λ1 + λ2 + λ3 )W s ,
where W s is the average downtime. The average downtime cost per unit time is
given by d(λ1 + λ2 + λ3 )W s . The cost for technician j per unit time is C + P(η1j +
η2j + η3j −1). Thus the average total cost per unit time is given by:
 
3  K
Z 1 = CK + P ηij − K + d(λ1 + λ2 + λ3 )Ws (1)
i=1 j=1

or,

Z 2 = Z 1 /d = (C − P)K /d + Z 3 (2)

where

3 
K
Z 3 = (P/d) ηi j + (λ1 + λ2 + λ3 )Ws . (3)
i=1 j=1

We divide both sides of (1) by d to obtain (2). Since we assume that the number of
technicians is fixed, (C − P)K/d is a constant, and we minimize Z3 .

Cross-Training Configurations
In our experiments, we assume that there are six technicians (K = 6). We use
the notation Cm-Sn-Ef to denote a configuration, where Cm stands for m servers
cross-trained in at least one secondary skill, Sn implies that each server who is cross-
trained has n secondary skills, and Ef denotes that the training in the secondary
skill is at the (100f )% efficiency level. We consider m = 3 and 6, n = 1 and 2, and
f = 0.6, 0.8, 0.9, and 1.0. We also allocate the technicians to each skill using the
concept of chaining (Jordan & Graves, 1995) which has been shown to be effective
in cross-training (e.g., Brusco & Johns, 1998). We denote the configuration with no
cross-training as C0-S0-E0. Table 1 shows the details of the seventeen cross-training
configurations considered in this paper.

A Cross-Training Index
When there are two job types and all the cross-training is at 100% efficiency, the
fraction of technicians cross-trained in both skills is a good measure of cross-
training (e.g., Shumsky, 1999, and Agnihothri, Mishra, & Simmons, 2003). When
there are three job types we develop a Cross-Training Index (CTI) to measure
the amount of cross-training in the field service workforce and compare several
Agnihothri and Mishra 243

Table 1: Server-skill assignment in cross-training and preferred order for various


configurations. Body of a table gives server identity.
Skill Class
Order 1 2 3
EXISTING SYSTEM
(a) C0-S0-E0
A B C D E F
I: Primary A C E
II: Primary B D F
1 2 3

(b) C6-S2-E1, C6-S2-E0.9, C6-S2-E0.8, C6-S2-E0.6


I: Primary A C E
A B C D E F
II: Primary B D F
III: Secondary F B D
IV: Secondary E A C
V: Secondary D F B 1 2 3

VI: Secondary C E A

(c) C6-S1-E1, C6-S1-E0.9, C6-S1-E0.8, C6-S1-E0.6 A B C D E F


I: Primary A C E
II: Primary B D F
III: Secondary F B D
1 2 3

(d) C3-S2-E1, C3-S2-E0.9, C3-S2-E0.8, C3-S2-E0.6


A B C D E F
I: Primary A C E
II: Primary B D F
III: Secondary F B D
1 2 3
IV: Secondary D F B

(e) C3-S1-E1, C3-S1-E0.9, C3-S1-E0.8, C3-S1-E0.6 A B C D E F


I: Primary A C E
II: Primary B D F
III: Secondary F B D
1 2 3

Server; Skill; Primary Skill; Secondary Skill.

configurations along one dimension. It is a relative measure designed to be 0


(the minimum) when all servers are dedicated and 1 (the maximum) when all
servers are trained in all skills at 100% efficiency.
 

 K  n 


 ηij − K 

 
R  j=1 i=1 
CTI = (4)
K 
 nK − K 


 


 

In equation (4) above, n is the number of skills and R is the number of servers who are
cross-trained in at least one secondary skill. The expression (nK − K) measures the
K n
maximum cross-training possible in secondary skills while ( j=1 i=1 ηij − K )
244 Cross-training Decisions in Field Services with Three Job Types

measures the actual cross-training imparted. The spread of cross-training among


more technicians as opposed to fewer having many skills is captured by R/K. This
index is not appropriate for systems where the maximum efficiency for a server
across all skills is less than 1. Table 2 shows CTI for varying server efficiencies
for all the 17 configurations. Prior studies have used a simpler index for measuring
cross-training. For example, Kher and Malhotra (1994) use the
number of skills
while Felan and Fry (2001) use, in our notation, an index Kj=1 i=1 n
ηij /K , where
ηij = 0 or 1.

LITERATURE REVIEW
The key features of our model are (i) server–job mismatch with the implications
for travel, (ii) varying efficiency in secondary skills, and (iii) three skill types
studied using (iv) a queueing framework. We now discuss the relevant literature
that involves each of these key features.
Server–job mismatch has not been studied earlier. With mismatch, travel time
becomes significant. In the absence of mismatch, one way to model the travel time
is to combine it with service time (e.g., Hill, 1992) and then use the insights de-
veloped for high server utilizations. When the travel time is negligible, additional
downtime due to a mismatch is small as well. However, with long travel times and
no cross-training, a mismatch results in markedly increased downtime and lower
server productivity. Travel time without mismatch has been studied in dual resource
constrained (DRC) job shops (Trevelen, 1989; Hottenstein & Bowman, 1998). In
the DRC literature, when a worker should be transferred to a different machine is
a decision variable, and researchers focus on finding optimum decision rules (e.g.,
Gunther, 1979; Park & Bobrowski, 1989). In field services, worker transfer must
occur after every job completion. Kher and Malhotra (1994) consider the impact of
transfer delay on the cross-training decisions in DRC systems and report that
cross-training benefits are the highest when there are no transfer delays. This
result is not significant to field service managers since travel times cannot be
avoided.
Assuming that the demand for skills is deterministic, several authors devel-
oped mathematical programming models for multiskilled workforce planning to
minimize staffing costs (e.g., Brusco & Johns, 1998, and Campbell, 1999). They
did not consider queueing delay and the trade-off between cost of cross-training
and cost of downtime as we do. When there are only two skill types, techni-
cians are either dedicated or cross-trained. Shumsky (1999), Agnihothri (2000),
and Chakravarthy and Agnihothri (2003) used analytical queueing models while
Agnihothri et al. (2003) used simulation to study such systems with queueing de-
lays. In comparison, our paper studies three skill types, efficiency in secondary
skills, and mismatch. The partial cross-training possible with three job types in-
creases the complexity of decision making. Hence, we use simulation like many
others in the literature (see, e.g., Begley, Kainen, & Maggard, 1983; Haugen &
Hill, 1999; and Felan & Fry, 2001).
If the training time is negligible and cost for a cross-trained worker is small,
a manager can cross-train all workers to reduce the downtime cost. In general,
due to the complexity of the machines, the training time is long and cross-trained
Table 2: Cross-training configurations for three skill types and six servers.
Server Efficiency in Each Skill Class
Server A Server B Server C Server D Server E Server F
Skill Class Skill Class Skill Class Skill Class Skill Class Skill Class
Agnihothri and Mishra

Order Configuration Code CTI 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3


1 C6-S2-E1.0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
2 C6-S2-E0.9 0.9 1 0.9 0.9 1 0.9 0.9 0.9 1 0.9 0.9 1 0.9 0.9 0.9 1 0.9 0.9 1
3 C6-S2-E0.8 0.8 1 0.8 0.8 1 0.8 0.8 0.8 1 0.8 0.8 1 0.8 0.8 0.8 1 0.8 0.8 1
4 C6-S2-E0.6 0.6 1 0.6 0.6 1 0.6 0.6 0.6 1 0.6 0.6 1 0.6 0.6 0.6 1 0.6 0.6 1
5 C6-S1-E1.0 0.5 1 1 1 1 1 1 1 1 1 1 1 1
6 C6-S1-E0.9 0.45 1 0.9 1 0.9 0.9 1 1 0.9 0.9 1 0.9 1
7 C6-S1-E0.8 0.4 1 0.8 1 0.8 0.8 1 1 0.8 0.8 1 0.8 1
8 C6-S1-E0.6 0.3 1 0.6 1 0.6 0.6 1 1 0.6 0.6 1 0.6 1
9 C3-S2-E1.0 0.25 1 1 1 1 1 1 1 1 1 1 1 1
10 C3-S2-E0.9 0.225 1 1 0.9 0.9 1 0.9 1 0.9 1 0.9 0.9 1
11 C3-S2-E0.8 0.2 1 1 0.8 0.8 1 0.8 1 0.8 1 0.8 0.8 1
12 C3-S2-E0.6 0.15 1 1 0.6 0.6 1 0.6 1 0.6 1 0.6 0.6 1
13 C3-S1-E1.0 0.125 1 1 1 1 1 1 1 1 1
14 C3-S1-E0.9 0.1125 1 1 0.9 1 1 0.9 1 0.9 1
15 C3-S1-E0.8 0.1 1 1 0.8 1 1 0.8 1 0.8 1
16 C3-S1-E0.6 0.075 1 1 0.6 1 1 0.6 1 0.6 1
17 C0-S0-E0 0 1 1 1 1 1 1
245
246 Cross-training Decisions in Field Services with Three Job Types

workers may not be trained at 100% efficiency in all skills. (In one company we
observed the training time to vary between three months and a year). In the DRC
literature, worker efficiency has been modeled as a function of the learning rate,
an exogenous variable (Felan & Fry, 2001). We model efficiency as an exoge-
nous input and a managerial decision like Begley et al. (1983), Brusco and Johns
(1998), and Campbell (1999). Unlike the last two studies, we study its impact on
downtime.

SIMULATION EXPERIMENTS
We built a simulation model using Arena 3.0 (Kelton et al. 1998). The inputs,
decision variables, and performance measures are explained in Figure 1. For each
parameter in the simulation study, we chose values that matched real systems
modeled in the papers published earlier. Table 3 shows the parameters and relevant
citations to justify those values. Since there is no exact expression for determining
the server utilization for our system, we obtain the desired server utilization by
adjusting the arrival rates in our simulation. To isolate the impact of cross-training
on downtime we select an arrival rate corresponding to the desired utilization for
the existing no cross-training system (C0-S0-E0) and denote it as U0 . Thereafter,

Figure 1: Inputs, decision variables, and performance measures for the simulation
model.

Performance Measure
Exogenous Inputs
Inter-Arrival Time Decision Variables Cost Ratio (P/d)
Service Time Number of Servers Cross-trained (C) Minimum Cost
Number of Secondary Skills (S)
Downtime Configuration
Number of Servers
Probability of Mismatch Efficiency in Secondary Skills (E)
Travel Time

Table 3: Values of the parameters used in the simulation.


Parameter Value(s) Relevant Citations
Number of technicians, K 6 Hill, March, Nachtsheim, &
Shanker, (1992), Haugen &
Hill (1999), Smith (1979)
Diagnosis Time Exponential
(mean 1 hour, max 6 hours) Watson et al. (1998), Haugen &
Repair Time Exponential Hill (1999)
(mean 1 hour, max 6 hours)
Utilization for no Mean Hill et al. (1992), Haugen & Hill
cross-training System U0 Inter-arrival Time (1999), Felan & Fry (2001)
(C0-S0-E0) 0.6 2.45 hours
0.8 1.83 hours
0.9 1.63 hours
Agnihothri and Mishra 247

Table 4(a): Three sets of scenarios for simulation experiments.


Scenario 1 No Mismatch: φ = 0
(No Mismatch, Low Travel Ratio: Travel ratio = 0.47.
Vary
Utilization) (Travel ∼ Triangular (0.6,1.0,1.2) with mean 0.93 hours and
variance 0.015)
Medium to High Utilization: U0 = 0.6, 0.8, and 0.9.
Scenario 2 Low to High Mismatch: φ = 10%, 20%, 30%, and 50%
(Vary Probability Low Travel Ratio: (as in Scenario 1)
of Mismatch) High Utilization: U0 = 0.8
Scenario 3 Medium Mismatch: φ = 20%
(With Mismatch, Medium to High Travel Ratio: Travel ratio = 0.93, 1.87
Vary Travel Mean Inter-
Ratio) Mean arrival time Travel
Travel time Distribution (hours) Variance (hours) Ratio
Triangular(1.53,1.95,2.12) 1.867 0.015 2.41 0.93
Triangular(3.39, 3.83, 3.98) 3.733 0.015 3.58 1.87
High Utilization: U0 = 0.8

we use the same arrival rate for different cross-training configurations. We assume
the arrival rates of different customer types to be equal.
We conduct simulation experiments for three sets of scenarios defined in
Table 4(a). We used synchronized common random numbers to reduce variability,
a conservative warm-up length, and the batch means approach to conduct output
analysis for the system in the steady state. Pair-wise comparisons using Tukey’s
honestly significant difference test revealed most of the differences significant at
the 0.01 level. Further details on the simulation methodology are available from
the authors.

RESULTS AND ANALYSIS


We determined the downtime (W s ) for different cross-training configurations at
various values of the parameters using simulation. Figure 2 shows the downtimes
for some scenarios. The optimum configuration balances downtime cost and cost
of cross-training. The cost of flexibility (cross-training premium) relative to the
downtime cost, P/d (Agnihothri, 2000), is termed the cost ratio. This cost ratio
is less than 0.5 for many field service systems. It is possible that this cost ratio
might be close to 1 or higher for machines with complex technologies requiring
expensive training with high installed base but low downtime cost per unit time
(e.g., office products). Using equation (3), we determined the optimum configura-
tion for various values of P/d. Figure 3 shows the optimum configuration for the
experiments in Scenario 2 defined in Table 4(a). We summarize some important
conclusions in Table 4(b) to provide approximate guidelines for the practitioners
(at the risk of some loss in precision) while the optimum configurations for all our
experiments are shown in Table 5.
248 Cross-training Decisions in Field Services with Three Job Types

Figure 2: Downtimes for all cross-training configurations for U0 = 0.8 with vary-
ing travel times and mismatch.

25
Mean Downtime (hours)

20

15

10

0
0 0.08 0.1 0.11 0.13 0.15 0.2 0.23 0.25 0.3 0.4 0.45 0.5 0.6 0.8 0.9 1
CTI

Tt/Ts = 0.47, No Mismatch Tt/Ts = 0.47, Mismatch 20%


Tt/Ts = 0.93, Mismatch 20% Tt/Ts = 1.87, Mismatch 20%
Tt/Ts = 0.47, Mismatch 50%

Figure 3: Comparing systems with mismatch probabilities of 0.1, 0.2, 0.3, and 0.5
with the no mismatch case using CTI for the minimum-cost configurations at each
cost ratio with U0 = 0.8.

1
0.9
0.8
0.7
0.6
CTI

0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2

P /d

Mismatch 10% Mismatch 20% Mismatch 30%


Mismatch 50% No Mismatch

No Server–Job Mismatch (Scenario 1)


Most of the cross-training models in the field service and DRC literature fall
under Scenario 1. Our simulation experiments for this scenario primarily repli-
cate prior results (except on efficiency in secondary skills) and are used to
Agnihothri and Mishra 249

Table 4(b): Managerial summary of cross-training decisions for field service sys-
tems with high utilization. C denotes number of servers cross-trained, S denotes
the number of secondary skills, and E denotes the efficiency in secondary skills.
No Mismatch Medium/High Mismatch Medium Mismatch
(φ = 0) (φ ≥ 20%) (φ ≈ 20%)
Low Travel Ratio Low Travel Ratio High Travel Ratio
Cost Ratio (Tt /Ts ≈ 0.5) (Tt /Ts ≈ 0.5) (Tt /Ts ≈ 2.0)
Low C: High/Medium C: High C: High
(P/d ∼ 0.1–0.5) S: One S: One/Two S: Two
E: Very High E: Very High E: High
Medium C: Medium C: Medium C: High/Medium
(P/d ∼ 0.5–1.0) S: One S: One S: One
E: High E: Very High E: High
High No Cross-training C: Medium C: Medium
(P/d ∼ 1.0–2.0) S: One S: One
E: High E: High/Medium

compare cross-training decisions in the presence of mismatch as modeled in


Scenarios 2 and 3.
Cost ratio and server utilization are important factors in making cross-training
decisions. When the cost ratio is close to zero, the managerial decision is trivial:
cross-train all technicians in all skills at 100% efficiency, that is, complete cross-
training is optimal. When the cost-ratio is not zero, an important and well-replicated
finding from the literature is that limited cross-training yields most of the benefits
of complete cross-training (Fryer, 1974; Jordan & Graves, 1995; Brusco & Johns,
1998; Felan & Fry, 2001). In addition, potential benefits of cross-training increase
with increasing server utilization (e.g., Agnihothri et al., 2003). Our findings from
Scenario 1 attest to these prior conclusions. We observe from Table 4(b) and Table 5
that it is best to cross-train all servers in one secondary skill at very high efficiency
when server utilization is high (U0 ≥ 0.8) and cost ratio is low (P/d ≤ 0.5). At
medium values of the cost ratio (0.6 ≤ P/d ≤ 1.0) some of the servers must be
cross-trained in one secondary skill at high efficiency. At high cost ratio (P/d >
1.0), no cross-training is the best choice. When server utilization is medium or
low (U0 ≤ 0.6), cross-training is not recommended except at low values of the
cost ratio. The reasoning is that when the utilization is low, the average downtime
is already low due to the availability of servers and the additional cost of cross-
training does not justify the further reduction in downtime cost. At high utilization
downtime tends to be very high and depending on the relative cost of cross-training,
more technicians should be cross-trained to reduce the downtime cost. Hence, in
Scenarios 2 and 3 we study only high utilization systems (U0 = 0.8).
From Table 5, we also note that training in two secondary skills is not cost-
effective except when the cost ratio is zero. This agrees with the findings of Park and
Bobrowski (1989) for DRC systems. We also noticed that at high utilization, cross-
training the same number of servers in one secondary skill at 100% efficiency
is better than (lower cost and equal or lower downtime) cross-training them in
two secondary skills at a lower efficiency. A drawback of cross-training at lower
Table 5: Minimum cost configuration for each cost ratio (P/d) for the three scenarios. The last row shows the % increase in downtime for 250
the no cross-training configuration as compared to complete cross-training.
(Scenario 1) (Scenario 2) (Scenario 3)
Tt /Ts = 0.47 Tt /Ts = 0.93 Tt /Ts = 1.87
U0 = 0.6 U0 = 0.8 U0 = 0.9 U0 = 0.8
P/d No Mismatch Mismatch 10% Mismatch 20% Mismatch 30% Mismatch 50% Mismatch 20%
0 C6-S2-E1 C6-S2-E1 C6-S2-E1 C6-S2-E1 C6-S2-E1 C6-S2-E1 C6-S2-E1 C6-S2-E1 C6-S2-E1
0.1 C3-S1-E1 C6-S1-E1 C6-S1-E1 C6-S1-E1 ,, ,, ,, ,, ,,
0.2 C3-S1-E0.8 ,, ,, ,, C6-S1-E1 C6-S1-E1 ,, C6-S1-E1 C6-S2-E0.8
0.3 C0-S0-E0 C3-S1-E1 ,, ,, ,, ,, ,, ,, ,,
0.4 ,, ,, ,, C3-S1-E1 ,, ,, C6-S1-E1 ,, C6-S1-E0.9
0.5 ,, ,, C3-S1-E1 ,, C3-S1-E1 ,, ,, ,, ,,
0.6 ,, ,, ,, ,, ,, C3-S1-E1 ,, C3-S1-E1 C6-S1-E0.8
0.7 ,, C3-S1-E0.9 ,, ,, ,, ,, ,, ,, ,,
0.8 ,, ,, ,, C3-S1-E0.9 ,, ,, ,, ,, C3-S1-E0.9
0.9 ,, C3-S1-E0.8 ,, ,, ,, ,, ,, C3-S1-E0.9 ,,
1.0 ,, ,, ,, ,, ,, ,, ,, ,, C3-S1-E0.8
1.2 ,, C0-S0-E0 ,, C3-S1-E0.8 C3-S1-E0.9 ,, C3-S1-E1 C3-S1-E0.8 ,,
1.4 ,, ,, ,, ,, C3-S1-E0.8 C3-S1-E0.9 ,, ,, ,,
1.6 ,, ,, ,, C0-S0-E0 ,, ,, ,, ,, C3-S1-E0.6
1.8 ,, ,, ,, ,, ,, C3-S1-E0.8 ,, ,, ,,
2.0 ,, ,, ,, ,, C0-S0-E0 ,, ,, C3-S1-E0.6 ,,
2.2 ,, ,, ,, ,, ,, ,, ,, ,, ,,
2.4 ,, ,, C3-S1-E0.9 ,, ,, ,, ,, ,, ,,
2.6 ,, ,, ,, ,, ,, ,, C3-S1-E0.9 ,, ,,
2.8 ,, ,, ,, ,, ,, C0-S0-E0 ,, C0-S0-E0 ,,
3.0 ,, ,, C0-S0-E0 ,, ,, ,, ,, ,, ,,
..
. ,, ,, ,, ,, ,, ,, ,, ,, ,,
4.4 ,, ,, ,, ,, ,, ,, ,, ,, C0-S0-E0
Downtime Increase 29% 66% 106% 90% 120% 163% 339% 154% 209%
Cross-training Decisions in Field Services with Three Job Types
Agnihothri and Mishra 251

efficiency is that when a technician is working on a job using her low-efficiency


secondary skill (and taking longer time to do it) another job requiring her primary
skill may arrive and either wait for a server or be assigned to a server with less than
100% efficiency.
In Table 5, we also find that for Scenario 1 the recommended efficiency in
secondary skills is close to 100%, except when the cost ratio is high. Even when the
cost ratio is high, an efficiency of 60% is never recommended. Cross-training with
60% efficiency produced among the worst average downtimes (see Figure 2, CTI
of 0.075, 0.15, 0.3, and 0.6), irrespective of the number of servers cross-trained
or number of secondary skills. These observations on efficiency are noteworthy in
light of the previously published studies and we elaborate on it in the next section.

With Server–Job Mismatch (Scenarios 2 and 3)


It is intuitively obvious that the cost-effectiveness of cross-training increases with
an increase in the probability of server–job mismatch and the travel ratio. This
intuition is supported by the values of the percent increase in downtime if no
cross-training configuration is chosen instead of complete cross-training (shown
in the last row of Table 5; also see the downtimes in Figure 2). But the tactical
managerial decisions are not obvious. In addition, as will be shown, the insights
from the previous section and the literature do not carry over entirely. Below, we
list important findings of our study, present the evidence, and discuss their reasons
and relevance for managers. We first identify scenarios in which complete cross-
training is optimal.

Result 1
Complete cross-training is cost-effective if mismatch and travel ratio are high and
the cost ratio is low.

As discussed in the previous section, studies repeatedly have shown that lim-
ited cross-training captures most of the benefits of complete cross-training due to
which the latter is not recommended when cost of cross-training is taken into con-
sideration. However, our study shows that complete cross-training is recommended
more often than previously suggested in the literature.
From Table 5 we can see that complete cross-training is optimal for cost ratios
up to 0.3 when the probability of mismatch is 50%, and for cost ratios up to 0.1
when the probability of mismatch is 20% or 30%. When the mismatch and travel
ratio are high, cross-training all technicians ensures that the service is completed in
a single visit. In the absence of cross-training, if a mismatch occurs the downtime
is very high due to the long travel time. Next, we address the issue of the number
of technicians to be cross-trained, the first of the three questions raised in the first
section.

Result 2
Cross-training a high proportion of technicians is recommended in situations with
(i) low cost ratio and (ii) medium cost ratio coupled with medium to high mismatch
and travel ratio.
252 Cross-training Decisions in Field Services with Three Job Types

The well-known insight from the literature that a limited amount of cross-
training is adequate works in many situations (e.g., see the optimality of C3-S1-E1 in
many settings in Table 5) but it could be detrimental to a field service organization
if used always. A majority of field service situations are characterized by cost ratios
that are low to medium (P/d ≤ 1.0). Result 2 implies that when the base utilization
is high with high mismatch and reasonable travel, a high number of technicians
must be cross-trained in one secondary skill.
We can see from Table 4(b) that the number of servers cross-trained is high
when the cost ratio is low and it is medium when the cost ratio is high. When
the cost ratio values are in between (medium) and the travel ratio is high, the
number of servers cross-trained should be high. These results can also be seen
more precisely from Table 5. When most of the servers have at least two skills,
more people are readily available to serve a job (due to cross-training), thus keeping
the downtime low. When cost ratios are high, the reduction in downtime cost is
not worth the increase in the cost of cross-training all technicians. But in Table 5,
for Scenarios 2 and 3, it is important to note that three out of six technicians are
cross-trained despite very high cost ratios, demonstrating the effectiveness of cross-
training in the presence of mismatch or high travel ratio. The experiments reported
here model six technicians in a territory. We ran similar experiments with three
technicians and found that all were cross-trained even at medium to high cost ratios.
When we modeled nine technicians, we saw that cross-training six and keeping
three dedicated was adequate even when the cost ratio is low. Now we answer the
second question regarding the number of secondary skills for a technician.

Result 3
Cross-training more technicians each in one secondary skill is better than cross-
training a few technicians each in two secondary skills.

A manager could use a fixed cross-training budget to either train more tech-
nicians in one secondary skill or a lesser number in two secondary skills. One may
use the latter approach thinking that the dedicated technicians can focus on jobs
requiring their primary skill and a few technicians, with training in all skills, can
give the benefits of limited cross-training. We do not recommend this approach. We
found that technicians should be cross-trained in two secondary skills only when
complete cross-training is optimal (covered by Result 1). Observe in Table 5 that
only after everyone has been cross-trained in one secondary skill, technicians are
given additional secondary skills. As we discussed in the previous section, training
at medium efficiency (60%) in two secondary skills is counterproductive. On the
other hand, if the efficiency is 100% in all skills, it is not cost-effective due to the
infrequent usage of more than two skills. Several studies in the DRC literature have
shown that cross-training in more than two or three skills is not useful (Hottenstein
& Brown, 1998). Finally, we investigate the optimal efficiency in secondary skills.

Result 4a
Cross-training with high or very high efficiency is recommended in most situations,
except when mismatch, travel ratio, and cost ratio are all high.
Agnihothri and Mishra 253

In Table 5, we find that cross-training with 60% efficiency is never rec-


ommended when the travel ratio is low, irrespective of the utilization, mismatch
probability, or cost ratio. Very high (100%) efficiency is optimal when the cost
ratio is low whereas a high efficiency (80% to 90%) is recommended when the cost
ratio is medium. We believe that a majority of field services have low to medium
cost ratio and hence fall into this category. Result 4a provides an important insight
to managers and researchers that cross-training at low efficiency is rarely justified.
An explanation for this result could be that with the usage of a technician with
very low efficiency, the low service cost does not offset the increased downtime
cost due to the larger service time. Furthermore, in high-utilization systems, if a
technician has low efficiency in a secondary skill, sometimes she might be busy
using this skill when she could be putting her primary skill to use on another job.
When the utilization is low, only a limited cross-training is justified and it is better
to be at a high efficiency.
Several papers on cross-training modeled efficiency in secondary skills as a
managerial decision and an exogenous fixed input to the system. Unlike previous
studies, we studied the impact of efficiency on downtime. The minimum efficiency
used in the published studies, including those that used real data, ranges between
0.4 and 0.6 (e.g., Begley et al., 1983; Brusco & Johns, 1998; Campbell, 1999).
Such low efficiency was never optimal in our experiments.

Result 4b
An efficiency of 100% in secondary skills is not optimal when the travel ratio is
very high.

In contrast to Result 4a, we see from the last column of Table 5 that when
the travel ratio is about two, an efficiency of 100% in secondary skills is never
recommended, except for a cost ratio of 0.1 or less. This result is noteworthy. It
implies that training in secondary skills at very high efficiency does not reduce the
downtime cost commensurate with even the relatively low expense in training when
the travel ratio is very high. The money is better spent in training more technicians
and, when cost ratio is low, on multiple secondary skills.

Result 4c
Cross-training with medium efficiency is appropriate when mismatch, travel ratio,
and cost ratio are all high.
In Table 5, we see that for high values of the cost ratio, training half the people
in one secondary skill at 60% efficiency (medium) is recommended. It is interesting
to note that even though 60% efficiency was never optimal in Scenarios 1 and 2, it
was recommended in Scenario 3 when the cost ratio and travel ratio are very high.
No cross-training is the best course only for very large values of cost ratio (P/d ≥
4.4), an unlikely situation for field services. The intuition behind this result is that,
when the travel time is long compared to the service time, if there is a chance of
mismatch, it is better to send a cross-trained person, even if he or she is not very
efficient in the secondary skill, than to send a person with only a primary skill. The
reduction in downtime cost justifies the cross-training.
254 Cross-training Decisions in Field Services with Three Job Types

As the mismatch probability and travel ratio increase, the mean downtime
decreases when more technicians are cross-trained in one or both secondary skills,
even if the efficiency is low (see Figure 2). Furthermore, CTI serves as a better
surrogate for downtime since the cross-training configurations sorted in increasing
order of downtime turn out to be approximately sorted in decreasing order of
CTI too. Thus, with high mismatch and travel times, the number of technicians
cross-trained and the number of secondary skills become more important levers
in reducing downtime than efficiency of cross-training. Moreover, it is adequate
to provide training with close to 100% efficiency. In contrast, cross-training at
100% efficiency is more desirable to reduce downtime at high utilizations with no
mismatch and a low travel ratio (Scenario 1).
In Figure 3, we notice that there is no overlap between the five curves
for various probabilities of mismatch. It leads us to conjecture that a minimum-
cost cross-training configuration for a particular probability of mismatch (φ) and
cost ratio (P/d) will also be the minimum-cost configuration for a higher φ
and P/d.

CONCLUSION
We developed insights on three managerial decisions in cross-training field service
technicians: the number of servers cross-trained, the number of secondary skills
imparted, and the efficiency in secondary skills. When there are more than two job
types, partial cross-training is possible, which increases the number of possible
configurations. We studied a system with three job types and compared various
cross-training configurations to evaluate the trade-off between cost of server cross-
training and cost of downtime, using a queueing framework.
Server–job mismatch is common in field services and it impacts cross-
training decisions. However, server–job mismatch had not been studied yet in
the cross-training literature. Many previous studies showed that when there is even
a small cost for cross-training, complete cross-training is not necessary since a
limited amount of cross-training captures most of the benefits of complete cross-
training. We found that this result holds in the absence of mismatch but com-
plete cross-training is optimal in some situations with mismatch. We also studied
efficiency in secondary skills as a managerial decision and obtained several in-
sights. Two interesting insights are that low efficiency (below 60%) in secondary
skills is rarely justified in most field service situations and, when the travel time
relative to service time is high, efficiency in secondary skills must be at less
than 100%.
This research opens up several questions that could be investigated further.
Researchers could study mismatch using analytical models for two job types.
Our model could be enhanced by incorporating one or more of factors, such
as learning-based efficiency (e.g., Malhotra, Fry, Kher, & Donohue, 1993, Fe-
lan & Fry, 2001, and Pinker & Shumsky, 2000), cross-training existing servers
versus increasing capacity (studied for one example in Begley et al., 1983), non-
linear downtime costs, nonstationary arrival (e.g., Watson, Chawda, McCarthy,
Drevna, & Sadowski, 1998), considering location of service along with the
Agnihothri and Mishra 255

nearest neighbor rule for job assignment, and assigning jobs through scheduling
(Haugen & Hill, 1999) rather than dispatching. [Received: April 2003. Accepted:
January 2004.]

REFERENCES
Agnihothri, S. R. (2000). Server flexibility in a queuing system with heterogeneous
customers. Working Paper, School of Management, Binghamton University,
Binghamton, NY.
Agnihothri, S. R., Mishra, A. K., & Simmons, D. E. (2003). Workforce cross-
training decisions in field service systems with two job types. Journal of the
Operational Research Society, 54(4), 410–418.
Begley, T., Kainen, T., & Maggard, M. (1983). The impact of alternative staffing
configurations on work unit performance in the service sector. Human Re-
source Planning, 6(4), 219–231.
Brusco, M. J., & Johns, T. R. (1998). Staffing a multiskilled workforce with vary-
ing levels of productivity: An analysis of cross-training policies. Decision
Sciences, 29(2), 499–515.
Campbell, G. (1999). Cross-utilization of workers whose capabilities differ. Man-
agement Science, 45(5), 722–732.
Chakravarthy, S. R., & Agnihothri, S. R. (2003). Optimal workforce mix in service
systems with two types of customers. Working Paper, School of Management,
Binghamton University, Binghamton, NY.
EDS (2003). Field service optimization overview chart. Available at http://www.
eds.com/services offerings/ov chart fso.pdf.
Felan, J. T., & Fry, T. D. (2001). Multi-level heterogeneous worker flexibility in
a dual-resource constrained (DRC) job-shop. International Journal of Pro-
duction Research, 39(14), 3041–3059.
Fryer, J. S. (1974). Labor flexibility in multiechelon dual-constrained job shops.
Management Science, 19(9), 1001–1012.
Gunther, R. E. (1979). Server transfer delays in a dual resource constrained parallel
queueing system, Management Science, 12, 1245–1257.
Haugen, D. L., & Hill, A. V. (1999). Scheduling to improve field service quality.
Decision Sciences, 30(3), 783–804.
Hill, A. V. (1992). An experimental comparison of dispatching rules for field service
support. Decision Sciences, 23(1), 235–250.
Hill, A. V., March, S. T., Nachtsheim, C. J., & Shanker, M. S. (1992). An ap-
proximate model for field service territory planning. IIE Transactions, 24(1),
2–10.
Hottenstein, M. P., & Bowman, S. A. (1998). Cross-training and worker flexibility:
A review of DRC System Research. Journal of High Technology Management
Research, 9(2), 157–174.
256 Cross-training Decisions in Field Services with Three Job Types

Jordan, W. C., & Graves, S. C. (1995). Principles on the benefits of manufacturing


flexibility. Management Science, 41(4), 577–594.
Kelton, W. D., Sadowski, R. P., & Sadowski, D. A. (1998). Simulation with Arena.
New York: McGraw-Hill.
Kher, H. V., & Malhotra, M. K. (1994). Acquiring and operationalizing worker
flexibility in dual resource constrained job shops with worker transfer delays
and learning losses. Omega, 22(5), 521–533.
Malhotra, M. K., Fry, T. D., Kher, H. V., & Donohue, J. M. (1993). The impact of
learning and labor attrition on worker flexibility in dual resource constrained
job shops. Decision Sciences, 24(3), 641–663.
Park, P. S., & Bobrowski, P. M. (1989). Job release and labor flexibility in a dual
resource constrained job shop. Journal of Operations Management, 3, 230–
249.
Pinker, E., & Shumsky, R. A. (2000). The efficiency-quality tradeoff of cross-
trained workers. Manufacturing and Service Operations Management, 2(1),
32–48.
Schapiro, D. (1998). Making the right match: The key to quality service and cus-
tomer retention. AFSM International Journal (April), 21–25.
Shumsky, R. A. (1999). Approximation and analysis of a queueing system with
flexible and specialized servers. Working Paper, University of Rochester,
Rochester, NY.
Smith, S. A. (1979). Estimating service territory size, Management Science, 25(4),
301–311.
Trevelen, M. D. (1989). A review of the dual resource constrained system research.
IIE Transactions, 21(3), 279–287.
Watson, E. F., Chawda, P. P., McCarthy, B., Drevna, M. J., & Sadowski, R. P.
(1998). A simulation metamodel for response-time planning. Decision Sci-
ences, 29(1), 217–241.

Saligrama R. Agnihothri is an associate professor of operations management


at the School of Management, State University of New York at Binghamton. He
holds an MSc degree from Karnatak University, Dharwar, India, and MS and PhD
degrees from the Simon Graduate School of Business Administration, University of
Rochester. His research interests include management of field service operations,
process flexibility and cross-training decisions, and scheduling in hospitals. He
has published in journals such as Operations Research, IIE Transactions, Naval
Research Logistics, European Journal of Operational Research, Interfaces, Journal
of the Operational Research Society and Computers and Operations Research. He
has served as an associate editor of Management Science and is on the editorial
review board of Production and Operations Management.

Ajay K. Mishra is an assistant professor of operations management at the School


of Management, State University of New York, Binghamton. He obtained a PhD in
Business Administration from the Katz Graduate School of Business, University
Agnihothri and Mishra 257

of Pittsburgh, Pittsburgh, and an MTech in industrial and management engineering


from Indian Institute of Technology, Kanpur, India. His research interests are in
component commonality, supply chain coordination, and applications of optimiza-
tion and simulation. His previous research has been published in Production and
Operations Management, European Journal of Operational Research, and Journal
of the Operational Research Society.

You might also like