You are on page 1of 13

OMEGA Int. J. of Mgmt Sci., Vol. 16, No. 4, pp. 333-345, 1988 0305-0483/88 $3.00 + 0.

00
Printed in Great Britain Pergamon Press plc

A Decision Support System for the


Courier Vehicle Scheduling Problem
AV HILL
University of Minnesota, Minneapolis, USA

VA MABERT
Indiana University, Bloomington, USA

DW MONTGOMERY
Crowe Chizek & Company, Indianapolis, USA

(Received June 1987)


The timely collection and transportation of checks from branches to the operations center is a primary
concern to bank managers. Transit cheeks need to be cleared quickly to maximize available funds to
the bank's cash managers. However, management must not over-commit resources, such as drivers
and vehicles, to improve the movement of checks from branch banks to the operations center. Bank
managers must make d i ~ t decisions regarding courier fleet size, routing, and scheduling in order
to balance these conflicting goals and to minimize overall system costs. We present a micro-computer
based decision support system (DSS) that helps bank managers make good decisions for their vehicle
courier service. This system, known as Cbexpedite, helps establish the correct number of vehieles, the
appropriate collection routes, the timing of the routes, and a view of the check flow pattern into the
operations center (back oMce check processing center). The system has been implemented at a number
of banks, indicating its value to bank management. In this paper, the logic of Chexpedite and its
database, optimization, and simulation capabilities are presented. A ease study application demon-
strates the system. The final section concludes with a discussion of Cbexpedite's strengths and
weaknesses.

I. INTRODUCTION for proof, encoding, sorting, and distribution.


Processing these checks before Federal Reserve
MANY COMMERCIAL BANKS use courier and other deadlines allows the bank to minimize
(messenger) vehicles to pick up checks and other its lost availability due to transit checks. (Tran-
items, and deliver them to an operations center sit checks are items deposited at the bank but
for encoding, sorting, and further processing. drawn against another bank, out of the local
Courier schedules in these banks can have a geographic area.)
major impact on the bank's float position and The transit checks must be cleared by specific
costs of operation. deadline times to receive credit in their account.
The clearing activities of a typical bank with Failure to meet the cutoff deadline delays the
multiple branches can be viewed as a series of availability of the funds by one or more days.
operations geographically dispersed. During the The impact on the bank's financial planning
normal course of business, checks are deposited could be substantial in today's banking environ-
at branch banks throughout the day. These ment. Most large commercial banks (Citi Corp,
checks will be picked up at certain times during Chemical, Bank of America, Seattle First, etc.)
the day and transported to an operations center normally clear over a billion dollars a day.
333
334 Hill et al.--The Courier Vehicle Scheduling Problem

i0:00 am -- Transit item r e c e i v e d at a branch bank

Ii:00 am -- Transit item waits for a courier at the branch

12:00 noon -- Picked up by a courier

i:00 pm -- -- Delivered to the operations center and begins encoding

2:00 pm -- Begins sorting

3:00 pm -- Transported to c l e a r i n g agent (Federal Reserve, etc.).

4:00 pm -- --

5:00 pm

6:00 p m

7:00 pm ---- Deadline for clearing

Fig. 1. Typical time line for clearing a transit check.

About 50% of these dollars are drawn against majority of time that a check spends in the bank
other banks. If poor clearing procedures (inap- is spent waiting for the courier vehicle. Once a
propriate courier routes and schedules) reduce check is received in the operations center, it is
the available investment funds by just 5%, up to encoded and sorted relatively quickly. If all
$25 million is lost in investment dollars each activities are done by 7: 00 p.m. in this example,
day. At 5% annual interest, the $25 million lost the check will clear that day.
availability equates to $1.25 million in lost Bank management can control the timing of
interest income per year. Although electronic courier schedules and the number of couriers
fund transfer systems are growing in popularity employed to influence the clearing function. An
(especially for large dollar items), paper checks opportunity for improvement in many banks is
and the associated clearing activities are in the area of improved courier schedules and
projected to continue to grow in volume over improved decisions related to determining the
the foreseeable future. number of courier vehicles to employ.
This paper presents a decision support sys- Several individuals have developed ap-
tem, known as Chexpedite, which has been proaches to the courier problem. Svestka [11]
found to be effective in scheduling couriers at proposed and tested a two-phase approach at
several commercial banks. We first present some Cleveland Trust (now known as AmeriTrust
background on courier scheduling decisions in a Company). The dynamic program maximized
commercial banking environment. Next, an the time discounted value of transit checks
overview of the Decision Support System picked up in order to reduce float cost. In the
structure is outlined, with a discussion covering second phase, an exact M x vehicle traveling
the procedures used to "optimize" the courier salesman branch-and-bound algorithm [12] was
schedules and the simulation used to perform applied to develop courier vehicle routes within
the "what-if" analysis. A case study presents each of the five periods. The implementation of
the use of Chexpedite and its impact on oper- this approach at Cleveland Trust did have sub-
ations at one bank. The conclusion discusses stantial savings, but the model fell into disuse
a number of Chexpedite's strengths and after Svestka left the bank.
weaknesses. Hill [6], Hill and McKenzie [7], Hill and
Whybark [8], and Mabert and McKenzie [9]
II. BACKGROUND
reported development of a heuristic algorithm
for the courier scheduling problem that at-
Figure 1 illustrates a typical time line for a tempted to build courier routes one stop at a
check being processed through the many clear- time for each route. Stops are added to each
ing steps. The time line points out that the route in order to maximize "efficiency" where
Omega, VoL 16, No. 4 335

efficiency is defined as the ratio of items picked them to a clearing center. The two-phase heuris-
up to the travel time for the route. This ap- tic begins by developing routes with a modified
proach has been found to be inadequate because assignment algorithm. If a route is not con-
of its myopic approach to building routes. nected to the clearing house or a transfer point,
Routes were often built that cross themselves it must be manually combined with one that is.
many times. This paper extends the original Once a set of feasible routes is developed, a
concept by changing the objective function of dynamic programming algorithm selects and
the building heuristic to reflect the multiple schedules routes so that checks arrive at the
goals inherent in the problem, adding a number clearing house as early in the work day as
of improvement heuristics, imbedding the sys- possible. Although the algorithm appears prom-
tem in an interactive decision support system ising for large banks with many transfer points,
that runs on a personal computer, and adding it requires substantial human intervention and,
several other features such as time-of-day travel therefore, may be difficult to implement. No
time multipliers and mail routes. implementation experience has been reported.
Haas and Zoltners [5] reported a heuristic More recently Davis, Ceto, and Rabb [2], and
approach that selects the next branch bank to be Davis, Kleindorfer, Kochenberger, Reutzel, and
added to a courier route as a function of both Brown [3] have proposed a simulation approach
travel time "savings" and check volumes. Cou- to the problem. This simulation has an imbed-
rier vehicles are "called" back to the clearing ded routing algorithm. The simulation ap-
center whenever the inventory of checks at proach was found to be useful at BancOhio
the clearing center goes below a specified in a study of the consolidation of operations
value. Although the authors used test data centers.
from several banks, no implementation has been The approach presented here is quite different
reported. from previous approaches and is an extension of
Davis and Swanson [4] proposed a computer earlier work. Chexpedite has the following char-
algorithm for the single-vehicle courier sched- acteristics:
uling problem. This approach has since been
extended to the multiple-vehicle courier prob- (1) Provides an interactive Decision Sup-
lem by Davis [1]. The procedure is somewhat port System that runs on a micro-
similar to that of Hass and Zoltners in that it computer that allows the user to build,
applies heuristic rules to select the next branch change, save, and retrieve databases.
bank for building a route and applies other
(2) Optimizes (heuristically) courier sched-
heuristic rules to call a courier vehicle back to ules.
the clearing center. Again, no implementation
has been reported to date. (3) Simulates user-specified courier sched-
Both the Haas and Zoltners [5] and the Davis ules to give the user the ability to
and Swanson [4] approaches have two major interactively change the "optimized"
shortcomings. First, both approaches require schedules and evaluate the results.
that several non-intuitive management control
(4) Includes time-of-day travel time multi-
parameters be set. Second, after finishing a
pliers to adjust travel times for traffic
route, both approaches require that the vehicle
congestion during different times of the
immediately begin a new route. Frequently, a
day between selected locations.
lower cost schedule can be developed if the next
route is delayed until more checks accumulate at (5) Gives the user control over the "opti-
the branch banks. mized" schedules by means of a variety
Muth and Lepman [10] at Bank of America of intuitive management parameters.
reported a heuristic approach for the courier
problem that allows for some branch banks to (6) Develops near-optimal courier sched-
be identified as "transfer" points. Bank of ules for early morning mail delivery to
America sends 45% of its routes to transfer the branches. (These are "mail routes"
points instead of a clearing center. The other that do not pick up checks, but rather
55% of the routes involve picking up checks at deliver mail to some or all branches in
transfer points or branch banks and delivering the morning.)
336 Hill et al.--The Courier Vehicle Scheduling Problem

Location Data includes the branch bank lo-


Database
cation ID, name, X - Y coordinates, minimum
Creation and time between stops, maximum number of stops,
Maintenance a "yes" or "no" to indicate if the location is
in the time-of-day travel time multiplier list
(to be discussed later). The X - Y coordinates
can be used to define base travel times
Route with modifications made because of unique
Optimization
features. Locations can be branch banks or
other banking facilities that require courier
stops.
Availability Schedule Data gives the number
What-lf of items (checks, etc.) and the average transit
Analysis
dollars per item by half-hour period by banking
location. This allows for individual character-
Fig. 2. Phases in the use of the Chexpedite decision support istics of branches to be maintained.
system. Item Requirements Schedule Data specifies the
number of items desired for processing at the
HI. THE DECISION SUPPORT SYSTEM operations center in each 15-minute period of
the day. This data typically reflects the actual
A Chexpedite analysis typically has three item processing capacity and the desired output
phases as illustrated in Fig. 2. The three phases necessary to improve the current float position.
are: Travel Time Data allows for user-specified
travel times to override the travel times calcu-
(1) Database creation and maintenance. lated from the X - Y coordinates. Chexpedite
default travel times are based on X - Y coordi-
(2) Route optimization. nates. These travel times can be replaced with
user-specified travel times. Travel times to or
(3) What-if analysis. from selected locations can be inflated or
deflated by the time-of-day multipliers to allow
Each of these three phases will be discussed in for traffic congestion. This requires only a list of
the following sections. locations that are in congested areas and a list
of time-of-day travel time multipliers for each
Database creation and maintenance
15 minute period of the day.
The Chexpedite algorithms are imbedded in a For trips between two locations that are in
decision support system that gives the user the the list of locations affected by the travel time
ability to load databases, create databases, multipliers, the normal travel time (either user-
change databases, save databases, optimize specified or default X-Y travel time) is inflated
schedules, simulate schedules, and print reports. or deflated by the multiplier that is on the
Chexpedite allows the user to create and main- approximate midpoint time of the trip. If nei-
tain a database that describes the banking oper- ther location is in the time-of-day travel time
ations environment. The databases is divided multiplier location list, the multiplier is always
into five sets of data: set to 1.0. If only one of the locations is
(1) Branch bank location data. in the time-of-day travel time multiplier location
list, the multiplier is (1 + m)/2 where m is the
(2) Availability schedule data. user-specified multiplier.
Parameter Data includes the average speed,
(3) Item requirements schedule data (at the map scale factor, operations center location,
operations center). operations center delivery time, earliest time
locations open, earliest time vehicles available,
(4) Travel time data. time all vehicles must be done for the day,
maximum number of stops desired per route,
(5) Parameter data. minimum number of stops desired per route,
Omega, Vol. 16, No. 4 337

maximum time desired per route, minimum item begins at the time that the first vehicle is avail-
volume desired per route, help option is turned able and ends at the time the first items are
on or off, mail routes turned " o n " or " o f f " , and required by the item requirements schedule. The
the time mail routes done. In the optimization second route then has a time window bounded
these parameters allow for more accurate mod- by the time the second vehicle is available and
elling of the operating environment. With the the run out time for the items from the first
simulation option (what-if option), these pa- route.
rameters are used to check a user-defined route Each route is built in the time window with
to be sure that travel times are reasonable and a "greedy" heuristic that finds the best insertion
that no constraints (e.g. maximum number of of any available branch bank location into the
stops per route) are violated. route to maximize the objective function value
Z (defined later). The process is repeated until
Route optimization no further insertions are possible without vio-
Building courier routes involves many lating the time window. Routes are always
conflicting goals. These goals include: constrained to end at the end of the route time
window.
(1) Supply enough items to the operations After the greedy heuristic has built a route,
center to keep the p r o o f area supplied the route is improved with several additional
with work. heuristics:
1. Exchange heuristic.
(2) Bring transit dollars into the operations
center as early as possible in the day in
2. Drop and add heuristic.
order to minimize float opportunity
cost.
3. Drop heuristic.
(3) Minimize transportation cost (fixed ve-
4. Reverse heuristic.
hicle cost and incremental cost of miles
traveled).
5. Greedy heuristic.
(4) Satisfy practical constraints on the
6. Move heuristic.
maximum number of stops at a branch
bank, the maximum number of stops on The exchange heuristic attempts to improve
a route, the maximum route time, and the route within the time window by trying all
the minimum number of items picked possible pair-wise exchanges of locations on the
up on a route. route. The drop and add heuristic pulls a lo-
cation out of the route and tries to find a better
Chexpedite applies several heuristics in order insertion for it in the same route. The drop
to find near optimal schedules. (Note that the heuristic attempts to improve the route by drop-
term "optimal" is used here to mean optimal ping a location from the route. The reverse
seeking. The imbedded algorithms are heuristics heuristic tries to improve the route by reversing
that do not guarantee mathematically optimal the direction of the route. If improvement is
schedules.) Chexpedite applies heuristics to first found by any of the other improvement heuris-
build and then improve courier routes. Routes tics, the greedy heuristic is applied again to
are built one route at a time using the "max- attempt to add another stop to further improve
imum delay" principal which states that routes the route. If the greedy heuristic finds some
should end as late as possible in the time improvement, all of the other improvement
window. The longer a route is delayed, the more heuristics are applied again. The move heuristic
items (and transit dollars) are picked up on each moves the route to the beginning of the time
route. The time window for a route begins at the window if all of the stops on the route are "after
time the vehicle becomes available and ends at closing" stops. (The "maximum delay" prin-
the time the operations center runs out of items ciple is no longer valid after closing.) The move
to process. heuristic is always applied after all other im-
In building the first route, the time window provement heuristics have been applied.

OME. 16/4~F
338 Hill et al.--The Courier Vehicle Scheduling Problem

In building and improving routes, the objec- maximum route time parameter (MXRTM),
tive function used to evaluate a candidate route and/or the minimum route volume parameter
is as follows: (MNRVL).
D Each time a route is changed it is simulated
Z = - - backwards from the end of the route time
TPIP2P3 window in order to calculate the objective func-
Where, tion value Z. This simulation calculates the stop
time for each location based on travel times
D = Total transit dollars picked up on the
modified by the time-of-day travel time multi-
route.
pliers. Item volumes and transit dollars for each
stop are also calculated.
T = Total travel time required for the route.
Some banks schedule early morning routes to
drop off mail at all locations. In this situation
PI Penalty function for having less than
the user is most interested in getting very
the minimum number of stops
efficient travel time routes. If the user instructs
(MNNSTP) or more than the max-
Chexpedite to build multiple vehicle mail routes
imum number of stops (MXNSTP).
before a certain time in the day, the greedy
This function is defined as:
heuristic is applied to all routes to build routes
for all vehicles simultaneously. These routes are
[max(MNNSTP--NS, NS
then improved with the improvement heuristics.
- - M X N S T P , 0)]2 -~- 1 The objective function here is the same as before
except that before opening time the route transit
where, NS is the number of stops on the dollars is fixed at 1. (In other words we are
route. applying standard multi-vehicle traveling sales-
man heuristics to develop these "mail routes".)
P~= Penalty function for the route being The user can select which branches are to be
greater than the maximum route time included in the mail routes.
(MXRTM). This function is defined as: The user can "optimize" the number of vehi-
cles by optimizing the schedules with 20 vehi-
max[(T--MXRTM)/IO.,O.] + 1 cles, 19 vehicles, 18 vehicles, etc. until a satis-
factory balance between the number of vehicles
where T is the route time in hours and and operations center idle capacity is found.
M X R T M is the maximum route time in One interesting study is to run Optimize Routes
hours. with an unlimited number of courier vehicles
(e.g. 99) to get an idea of the maximum number
P~= Penalty function for the route having of items that can be brought into the operations
less than the minimum number of items center.
(MNRVL). This function is defined as: In selecting a schedule, the user will have to
make many economic and practical tradeoffs.
max[(MNRVL-VL)/25.,0.] + 1 Economic issues include vehicle costs (lease, gas,
maintenance), driver costs, proof costs, and
where VL is the total item volume for float-related costs. Although Chexpedite can
the route and M N R V L is the user develop "optimized" schedules that deal with a
specified minimum desired route vol- large number of practical issues (maximum
ume. number of stops per location, etc.), it does not
explicitly evaluate vehicle costs, proof costs, or
The D/T ratio reflects our desire to have float costs. It was judged that the data collection
maximum "efficiency" routes that provide the effort required to accurately measure these costs
maximum number of transit dollars per unit was practically impossible. For example, float
travel time. The penalty functions P~, P2, and P3 costs require information on the sending point
serve to deflate the objective function value mix and dollar value of items by branch bank
whenever a route violates the minimum number by time of day. We find that the pragmatic
of stops per route parameter (MNNSTP), the approach is to simply evaluate schedules on a
Omega, Vol. 16, No. 4 339

relative basis (e.g. in our current environment Reports selection and change any route with the
we will have $2,000,000 of transit items pro- Change Routes Data selection. For example,
cessed by 1:00p.m. and in our "optimized" after Optimize Routes has been run the user
environment we will have $10,000,000 of transit may want to change the order of two locations
items processed by the same time with no in a route. The user should select Change
additional vehicle cost). Routes Data option to change the routes, What-
If the bank has multiple operations centers, if to simulate the new alternative, and Reports
each location must be assigned to a single to print the results for evaluation.
operations center. Chexpedite then handles each Chexpedite allows the user to save a Chex-
problem separately. If the bank has multiple pedite data file with a route structure. The user
collection points that are later picked up by may test a large number of alternative scenarios
couriers to be delivered to an operation center, with the What-if option, save each one in a
each collection point is treated as a separate Chexpedite data file, and then retrieve and print
Chexpedite operation center to develop a sched- the file with the best solution.
ule. A " m a c r o " Chexpedite schedule is then
solved to pick up items from the collection IV. AN E X A M P L E
points.
Sometimes we find that the user-specified item Chexpedite is designed around a decision
requirements schedule demands more items support system framework to facilitate ease of
than can possibly be brought in from the branch use. To demonstrate this system, we wish to
bank locations. When this occurs the time win- present a short case study based upon our
dow for a route becomes so small that not even experience with commercial banks dealing with
one stop will fit into the time window. In this multiple branch check clearing. Many of the
case the time window heuristic will schedule exhibits presented during this discussion come
fifteen minutes of out-of-work time to make the directly from the Chexpedite reporting system.
time window larger. The route building heuris- Middleton National Bank is a small commer-
tics are again applied to attempt to build a route cial bank with a main office and fourteen
in this new enlarged time window. The time branches. Management is concerned about
window is made larger until the minimum clearing the maximum transit dollars each
number of stops (MNNSTP) will fit on the day and is considering changes to the trans-
route. portation department (adding another vehicle)
and/or adjusting capacity within the Check Pro-
What-if analysis cessing Department (CPD) to improve the
The What-if analysis is a deterministic simu- bank's clearings each day.
lation of the bank's courier system. In order to To address these types of banking decisions,
use this option, the user defines the courier it is important that a database be developed
schedules with the help of the interactive Chex- containing the appropriate information on lo-
pedite DSS. Chexpedite will suggest travel times cation of branches, traffic patterns within
to help the user build a schedule. Once the branches, transit dollars collected in branches,
schedule has been defined, Chexpedite will sim- travel time between branches, etc. Such a data-
ulate in order to estimate the number of items base presents the heart of Chexpedite. Travel
and the transit dollars picked up at each stop time data can be one of the more difficult items
as well as the projected item inventories and to collect. To facilitate travel time data col-
out of work times at the operations center. lection, one can take a standard street map and
Caution messages are printed if the user- plot the locations of the branches as shown in
specified travel times are unreasonable or if any Figs. 3 and 4. Using scaling factors that convert
of the maximum or minimum parameters such map distance to approximate travel times, the
as the maximum number of stops per location initial travel times can be estimated between all
are violated. locations. Occasionally, direct travel time esti-
The What-if and Optimize Routes selections mates may be necessary to reflect unique travel
can work together. After the routes have been conditions (one way streets, highways, etc.). The
"optimized" with the Optimize Routes selec- user can specify these travel time pairs directly.
tion, the user can study the routes with the The normal travel time (not modified by the
340 Hill et al.--The Courier Vehicle Scheduling Problem

11.5

Troy
#
11.0 Birmingham

i
RoseviLte
~armington Madison
St CLair
HiLLs Shores
LSouthfieLd
o 10.5
I

Livonia
HighLand
Redford
Township ~ r a ! c k
10.0

Detroit
Dearborn
WestLand Gardencity Height~
Dearborn
9.5 Windsor j
j • / ]
// "1 I I I
5.0 5.5 6.0 6.5 7.0 7.5
X-AxTs
Fig. 3. Branch locations.

travel time multipliers) for this example is for collection during the day at each branch
shown in Fig. 5. Branch bank locations in bank location. Figure 6 lists the estimated ar-
highly congested areas will have incoming and rival pattern (Availability Schedule) for one
outgoing travel times inflated or deflated by the branch (Location L1), plus a summary by half
time-of-day travel time multipliers. Figure 4 hour for all fifteen branches. Most banks do not
indicates which locations were judged to be find it necessary to collect detailed information
affected by the travel time multipliers. on the transit dollars per item by half hour and
Data on arrival patterns of items and transit simply set the transit dollars per item to $1.00.
dollars are required to estimate available work In addition to branch information, Chex-

MULTI-
MAXIMUM MINIMUM PLIERS
LOCATION NUMBER TIME AFFECT
LOCATION COORDINATES STOP OF STOPS BETWEEN TRAVEL
ID NAME -X- -Y- TIME PER DAY STOPS TIMES

LI SMITH STREET 6.70 10.20 7 MIN 5 STOPS 30 MIN Yes


L2 MAIN BRANCH 6.50 9.90 7 MIN 5 STOPS 40 MIN Yes
L3 NORTH SIDE 7.30 10.40 5 MIN I0 STOPS 30 MIN Yes
1,4 HILL STREET 7.10 10.20 4 MIN I0 STOPS 30 MIN Yes
L5 LJC STREET 6.90 10.40 5 MIN i0 STOPS 30 MI N No
L6 HYLAND CREEK ROAD 6.50 i0.40 5 MIN i0 STOPS 30 MIN Yes
L7 94TH STREET 7.00 i0.00 5 MIN i0 STOPS 30 MIN Yes
L8 NORMANDALE BLVD. 7.20 i0.i0 5 MIN i0 STOPS 30 MIN Yes
L9 HIGHWAY I00 6.70 11.20 5 MIN i0 STOPS 30 MIN Yes
LIO HIGHWAY 494 - SOUTH 6.90 10.50 5 MIN i0 STOPS 30 MIN Yes
LII PENN AVENUE 6.70 10.90 5 MIN i0 STOPS 30 MIN Yes
LI2 FRANCE AVENUE 6.90 10.20 5 MIN i0 STOPS 30 MIN Yes
LI3 HUMBOLDT AVENUE 6.70 i0.i0 5 MIN i0 STOPS 30 MIN Yes
LI4 XERXES AVENUE 5.20 9.90 5 MIN i0 STOPS 30 MIN No
LI5 MCKNICHT ROAD 4.80 I0.i0 5 MIN I0 STOPS 30 MIN No

Fig. 4. Location data.


Omega, Vol. 16, No. 4 341

An asterisk (*) indicates a user-specified travel time. Other travel times


are calculated from X-Y coordinates, average speed, and map scale factor.

FROM LOCATION L1 SMITH STREET

L1 0 MIN L2 10 MIN* L3 13 MIN 1.4 8 MIN L5 6 MIN

L6 6 MIN L7 i0 MIN* L8 I0 MIN L9 20 MIN LI0 7 MIN

LII 14 MIN LI2 5 MIN* LI3 5 MIN* LIA 31 MIN LI5 38 NIN

FROM LOCATION L2 MAIN BRANCH

LI 7 MIN L2 0 MIN L3 19 MIN L4 13 MIN L5 13 MIN

L6 i0 MIN L7 i0 MIN L8 15 MIN L9 26 MIN LI0 14 MIN

LII 20 MIN LI2 i0 MIN LI3 6 MIN LI4 26 MIN LI5 34 MIN

FROM LOCATION L3 NORTH SIDE

LI 13 MIN L2 19 MIN L3 0 MIN 1.4 6 MIN L5 8 MIN

L6 16 MIN L7 I0 MIN L8 7 MIN* L9 20 MIN LIO 8 MIN

LII 16 MIN LI2 9 MIN LI3 13 MIN LI4 43 MIN LI5 50 MIN

FROM LOCATION 1.4 HILL STREET

L1 8 MIN L2 13 MIN L3 6 MIN L4 0 MIN L5 6 NIN

L6 13 MIN L7 4 MIN L8 8 MIN* L9 22 MIN LI0 7 MIN

LII 16 MIN LI2 4 MIN LI3 8 MIN LI4 38 MIN LI5 46 MIN

FROM LOCATION L5 LJC STREET

LI 6 MIN L2 13 MIN L3 8 MIN 1.4 6 MIN L5 0 NIN

L6 8 MIN L7 8 MIN L8 8 MIN L9 16 MIN LI0 2 MIN

LII Ii MIN LI2 4 MIN LI3 7 MIN LIA 35 MIN LI5 42 MIN

Fig. 5. Travel time data.

pedite needs an Item Requirements Schedule. Middleton National Bank uses three vehicles
This schedule reflects the bank management's to travel between the main office, which has the
desired average flow pattern into the operations Check Clearing Department, and the fourteen
center. Figure 7 provides the current hourly and branches. Each day messengers pick up checks
cumulative Item Requirements Schedule for the at the branches and deliver them to the central
Middleton National Bank CPD as well as the office for clearing. Like all commercial banks,
newly proposed Item Requirements Schedule Middleton is sensitive to the deadline time es-
for this department. The current Item Require- tablished for clearing transit checks. Given their
ments Schedule has the CPD's staff equally past experience, all checks received by 6: 00 p.m.
scheduled during the day from 12:45p.m. to at the main office will normally be processed
8:45p.m. for a total of 64,000 items. Since through the check clearing stages and be ready
much of the check volume does not arrive until for transmission. Using Chexpedite to evaluate
later in the afternoon, a new staff schedule has the current Item Requirements Schedule for the
been proposed that keeps the current total item CPD and three vehicles, one sees (Fig. 8) that
requirements schedule to 64,000 items but short- the maximum transit dollars that can be re-
ens the overall work day to 7:30p.m., with a ceived by 6:00 p.m. are $16.11 million. (Figure
larger staff present during the late afternoon. 9 provides a sample of two routes that
It was hoped that this new schedule would have created the arrival of checks presented in
improve clearings of the department. Fig. 8).
342 Hill et al.--The Courier Vehicle Scheduling Problem

u]

ooooooo
o o o o o
o o o o o o
o ~ ~ ~~ ~~ ~ ~~
,.le. o ~ [ J
~ = o ~ ~ ~ ~ ~ ~
~od~n

oi! 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ~0
0000
000
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
o3

.C

o ~ ~ ~ ~ o o o o o ~ ~ o o o o o

~o ooooooOoOoOOoOoOOOOoOOOo~oOoOOOoOOO~ oo

~ 0 ~ 0 ~ 0 ~ 0 ~ 0 ~ 0 ~ 0 ~ 0 ~

o ~ o ~ o ~ o ~ o ~ o ~ o ~ o ~ o ~ o ~ o ~ o ~ o ~ o ~ o ~ o ~ o
o ~ o ~ o ~ o ~ o ~ o ~ o ~ o ~ o ~

<<<<<<<<<~
~ooooo~o~ o~ o o o o~o o 0 o ~ o

"0

..C

;_T,

;>
<

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

oooooooooooooooo ~ o o o o o o o o o o o ~ a ~ ~
Omega, Vol. 16, No. 4 343

-../).

0
. . . . . . o ~

.o

°I o¢

.... o ....... oooooooooooooooo o

0~
L~

I ~, 13, O, i:~ i:~ ~. i~ i:~ ~ ~, i~. 13, ~a, i:~ i~ ~, ~. 0,, ~ i~ ~ ~ ~ ~ I~ I~ I~. ~- ~, I~. I~ I~ O, I~-

, .~ ~; ~ ~ ,:.; ,~ ~ ,.:.;~ ~ ~ ~ ~ ,.:..;,:.:,,:..;~ ~ ~: ~: ~ ,~ ~ ~ ,~ ,~ ,~ ,~ ~ ~ ~ ~ ,~ ,;:; ,:,: ,~ ~


344 Hill et al.--The Courier Vehicle Scheduling Problem

As indicated earlier, Middleton's manage- section presents the experience at National


ment was interested in evaluating whether the Westminster Bank of New York with 140
new proposed Item Requirements Schedule for branches.
the Check Processing Department could im- National Westminster's branches were
prove the clearance of transit dollars. Also grouped into one of six sectors with approxi-
under consideration was the possibility of ad- mately 30 branches per sector. Each sector had
ding another vehicle to pick up checks at the a hub from which the couriers were dispatched
branches. Chexpedite provides a powerful tool and items were consolidated for movement to
to help evaluate these alternatives and assess the check processing center. A study utilizing
their impact on the operations of Middleton. Chexpedite was conducted in four of the six
Using the Optimize Routes capabilities of sectors to improve operations.
Chexpedite, one can evaluate the current and The estimated times to each branch were
proposed CPD Item Requirements Schedule, entered into the model to initialize the database.
plus whether three or four vehicles should Because of the inconsistencies in New York
be employed. The following is a summary of traffic, and because foot messengers, subways,
transit dollars cleared by 6 p.m. for the different helicopters, and cars were being used to deliver
alternatives. items, the travel time data created the major
challenge with using Chexpedite. It was neces-
Transit dollar clearing summary sary to modify the time-of-day multipliers and
Three Four lower the travel times considerably from the
vehicles vehicles initial estimate to obtain reasonable values.
Current item schedule $16,119,297 $16,442,878 With the information in place, the routes were
Proposed item schedule $17,366,208 $18,176,468
optimized using Chexpedite while considering
the minimum and maximum stops per branch,
One can see from the data in the above table minimum number of items to collect on any
that the proposed Item Requirements Schedule, stop, and exchange schedules at the hubs to the
with four vehicles, allows the most transit dol- check processing center. The optimized routes
lars to be available at the CPD by the 6: 00 p.m. were reviewed and revised to meet internal bank
deadline time. Almost a $2 million increase in policies, such as mail delivery times.
transit dollars occurs per day, which increases Chexpedite was one phase of a larger two
investable funds. With this information avail- year work-flow/transportation project. The
able, bank management can now collect the Chexpedite phase, which lasted approximately
other expenses associated with these alterna- one year, had an estimated annual savings of
tives, such as the cost of additional vehicle, an approximately $90,000 in the four regions used.
additional driver, extra maintenance, adjust- The total project had savings of approximately
ments in check processing work schedules, etc., $250,000 resulting from reduced number of
to evaluate the total cost implications. Many drivers, messengers, and mileage. These savings
other alternatives (such as a different Item did not include the better clearance of transit
Requirement Schedule for the CPD) can also checks that also resulted.
be conveniently evaluated by Chexpedite for Discussions with the bank's users indicated
management's consideration. the recognition that travel times did not have to
be exact. Having reasonable approximations
was more than adequate for Chexpedite to build
good messenger routes.
V. IMPLEMENTATION

Since its creation, Chexpedite has been adop- VI. CONCLUSIONS


ted by a number of banks (i.e., First National
Bank of Atlanta, BayBanks (Boston), and First Chexpedite has proven to be an effective DSS
National Bank of Maryland) in the United for several large commercial banking institu-
States. Even though Chexpedite utilizes a gen- tions because it has a number of advantages
eral design to facilitate easy use, there are over current approaches to route development.
always unique situations that are present. This First, it provides a structured approach to ad-
Omega, Vol. 16, No. 4 345

dress a complex problem. The interaction of Chexpedite by its very nature cuts across or-
branch arrival patterns, vehicle travel times, ganization lines of the branch administration,
C P D process capacity, unique restrictions, etc., check processing, and transportation areas. It is
influence the appropriate routes and timing. not clear who should be in charge and who
Second, since Chexpedite is a micro based DSS, should make the tradeoffs between float, trans-
it can develop reasonable routes quickly for user portation, and check processing related costs. It
evaluation. The system provides a user-friendly is clear, however, that Chexpedite can provide a
approach to developing sophisticated models of useful tool for making these decisions.
branch bank check collection systems. Third,
users can insert unique conditions/requirements ACKNOWLEDGEMENTS
and test these conditions. These models include The authors would like to thank the many individuals and
travel times that can be adjusted by the time of organizations who contributed to the development of Chex-
day, multiple vehicles, morning mail routes, pedite over the past ten years. These include Professor Clay
Whybark, Mr K Edward Elverud, Mr Phil McKenzie,
and m a n y other practical issues. And fourth, North Carolina National Bank, Ernst & Whinney, and
Chexpedite's decision logic focuses on both the Crowe Chizek & Company.
available items at branches and the C P D capa-
bility to process. This coordinated decision REFERENCES
making assures that the m a x i m u m dollars are I. Davis SG (1978) Float reduction through the scheduling
cleared. The schedules can be either optimized of commercial bank branch messengers by heuristic
(heuristically) or simulated. programming. Unpublished doctoral dissertation, Syr-
acuse University.
The challenges of using Chexpedite include 2. Davis SG, Ceto N and Rabb JM (1982) A comprehen-
the data collection effort, the modeling effort, sive check processing simulation model. J. Bank Res. 13
and the implementation effort. The data col- (3), 185-194.
3. Davis SG, Kleindorfer GB, Kochenberger GA, Reutzel
lection challenge is significant. The most ET and Brown EW (1986) Strategic planning for bank
difficult data to collect is the item and transit operations with multiple check-processing locations.
dollar availability schedule data for each branch Interfaces 16 (6), 1-12.
4. Davis SG and Swanson LA, (1978) A computerized
bank. Few banks have found it worthwhile to operations scheduling model for the reduction of com-
try to break down the average transit dollars per mercial bank float. J. Opl Res. Soc. 29 (6), 559-564.
item for each half hour of the day. Most banks 5. Haas G and Zoltners AA (1977) A computerized bank
check collection vehicle routing system. J. Bank Res. 8
have found it useful to simply estimate an (3), 148-158.
arrival profile (pattern) in terms of percent of 6. Hill AV (1977) Structuring and solving the bank mes-
total items that come in each half hour and senger vehicle scheduling problem. Unpublished doc-
toral dissertation, Purdue University.
multiply this times the total expected for each 7. Hill AV and McKenzieJP (1977) A computer algorithm
day of the week. Although this approach may for messenger vehicle scheduling at Ohio National
not satisfy the precision of most accounting Bank. Proceedings of the AIIE Systems Engineering
Conference, American Institute of Industrial Engineers,
functions, it serves the operations requirements Norcross, U.S.A.
for this application quite well. The philosophy 8. Hill AV and Whybark DC (1982) CHEXPEDITE: A
behind Chexpedite is to make the data col- Computer-based Approach to the Bank Courier Prob-
lem. Decis. Sci. 13 (2), 251-265.
lection effort as simple as possible. Pareto's 9. Mabert VA and McKenzie JP (1980) Improving bank
80-20 law suggests that 80% of the benefit operations: a case study at BancOhio/Ohio National
can be achieved with only 20% of the data Bank. Omega 8 (3), 345-354.
10. Muth MB and Lepman RT (1980) Achieving prod-
collection effort. uctivity gains in Bank of America's branch network.
The modeling challenges arise when the bank Paper Presented at TIMS/ORSA Joint National Meet-
has multiple operation centers, m a n y non- ing, May. Washington, DC.
1I. Svestka JA (1976) A system model for controlling the
branch activities for couriers, or other unusual operations of check processing in a branch bank net-
situations which must be modeled. We have work. Interfaces 7 (2), 69-79.
found that the optimization can still provide 12. Svestka JA and Huckfeldt VE (1973) Computational
experience with an M. Salesman traveling salesman
satisfactory "first cut" schedules for m a n y of algorithm. Mgmt Sci. 19 (7), 790-799.
these problems. These "first cut" schedules can
then be refined to meet all of the organizational
requirements with the simulation (what-if)
ADDRESS FOR CORRESPONDENCE: Professor VA Mabert,
model. School of Business, Indiana University, Tenth and Fee
The implementation challenges are many. Lane, Bloomington, IN 47405, USA.

You might also like