Absolutely, Positively Operations Research:

The Federal Express Story

RICHARD O . MASON

Edwin L. COX school of Business

Southern Methodist University

Dallas, Texas 75275
JAMES L . M C K E N N E Y

Harvard Business School

Soldiers Field
Boston, Massachusetts 02163

WALTER CARLSON

216 west miton Drive

Bolder Creek, California 95006

DUNCAN COPELAND

School of Business Administration

University of Western Ontario
London, Ontario N6A 3K7
Canada

Federal Express Corporation has used operations research (OR)

to help make its major business decisions since its overnight
package delivery operations began in 1973. An early failure
pointed out the need for scientific analysis. Subsequently, a successful origin-destination model followed by models to simulate
operations, finances, engine use, personal assignments, and
route structures influenced the conduct of business during periods of substantial growth. There were many false starts between
the successes. CEO and founder Frederick W. Smith played a
central role in the use of OR at the company: he established a
relationship with OR and management science personnel and
this relationship supported the growth and success of the
company.
This company is nothing short of being the logistics arm of a whole new society that is building up
in our economya society that isn't built around automobile and steel production, but that is built
up instead around service industries and high technology endeavors in electronics and optics and
medical science. It is the movement of these support items that Federal Express is all about.
Frederick W. Smith

rederick W. (Fred) Smith developed a

vision of a business, brought together
the key people who would make up his ini-

tial management team, and scraped together enough funds to buy the first aircraft for his fleet: French-made Dassault

Copyright 1997, Institute for Operations Research

and the Management Sciences
0092-2102/97/2702/0017$05.00
This paper was refereed.

TRANSPORTATIONAIR
COMPUTERSSYSTEM DESIGN AND OPERATION

INTFRFACES 27: 2 March-April 1997 (pp. 17-36)

MASON, MCKENNEY, CARLSON, COPELAND

Turboprop planes were too

slow.

ing up and delivering small packages.

The founders were familiar with the
South, and this seemed an appropriate
place to begin their new business. They
chose 11 cities, mostly located in the South
and Southeast, to make up the first route
network, which they would operate in a
hub-and-spoke manner. They would fly
outbound packages into Memphis, sort
them, and then forward them to their inbound delivery destinations. FedEx thinks
of things from the point of view of the
package. So, in FedEx parlance, outbound
refers to traffic moving from the station of
origin to the hub (or towards its destination); inbound from the hub (or any intermediate spot) to the destination.

company was forced to operate as an air

taxi service and use aircraft that had a payload of less than 7,500 lbs. Early studies
had shown that turboprop planes were too
slow to meet tight overnight deadlines. Executive jets were the only aircraft that
would. Compared with other executive jets,
the Falcons were sturdier, had lower operating costs, and featured a GE engine that
was more powerful and considered more
reliable than the Pratt and Whitney or Air
Research engines. Equally important, Dassault had 22 new Falcons sitting in the New
Mexico desert and was willing to deal.
With a 56-foot fuselage and 50-foot wing
span, the Falcons had a range of about
1,500 to 2,000 miles flying at about 540
miles per hour and, when appropriately
modified, could carry a payload of about
three tons. Given Smith's requirements, the
Falcons were about the only satisfactory
aircraft"the best of the worst," as some
one put itfor an overnight service pick-

Initial sales calls had generated a lot of

enthusiasm and large estimates of volume.
Expectations were running high. Smith
even worried that his tiny fleet would be
overbooked. March 12,1973 was set for the
inaugural service. That evening Smith, his
founding team, and a few of his major investors waited anxiously as the Falcons descended into Memphis with the first d a / s
load.
Henry Meers, an investment banker, recalls the evening's events. "1 saw the anguish on their faces as they waited. Most
were very worried about their future. It
was a critical moment for all of them as
they finally crowded around the Falcons
and the cargo doors opened. But as they
gazed inside, there was bitter disappointment. There were only six packages and
one of them was a birthday present Fred
Smith sent to his close aide Irby Tedder"
[Sigafoos and Easson 1988, p. 59). The next
two evenings showed little improvement,
and after only three days of operation, the

INTERFACES 27:2

18

Falcon-20 twin-engine executive jets.

His idea was to provide overnight delivery of small, high-value items, such as
Pharmaceuticals, aerospace components,
and computer parts. An avid pilot. Smith
had chosen the Falcons because they best
satisfied the constraints the business had to
work under. To be successful, the business
had to be free to change its routes and
schedules frequently and readily. Civil
Aeronautics Board (CAB) regulations on air
cargo at the time, however, did not permit
this degree of flexibility. Consequently, the

FEDERAL EXPRESS
company discontinued air delivery service.
To maintain its customers, the company
found alternative ways to transport their
packages while it took stock of its situation.
The First Model

After getting over the initial shock, Fred

Smith appointed a task force headed by
Charles Brandon, a physicist and an ardent
flyer he had met in 1962, when both were
home from college. When Smith was contemplating his new business in 1972, he
asked Brandon, "Could you use a computer to better schedule the aircraft?" "Definitely," was Bronson's reply. Brandon
soon thereafter became an advisor to FedEx
and an early employee. Smith's charge to
Brandon was straightforward: Where have
we failed? What should we do now?
The team members closeted themselves
in a conference room at Little Rock's Worthen Bank to begin their soul-searching,
working 18 to 20 hours a day for 15 days
straight. Their approach was classic. They
began by challenging the assumption that
the cities the founders knew best were best
for doing business. This led them to focus
on what characteristics made a city a good
candidate for their route structure, and
they concluded that they should include
some cities because they generated a large
volume of outbound small packages; others
because they received a lot of small packages. Moreover, the final set of cities
should account for a substantial amount of
the total volume within the system.
Given these guidelines, they next acted
on their intuitive ideas in a systematic way
to find data or indicators they could use to
select cities and to decide on a route system. First, they formulated the problem in
terms of an origin-destination flow model.

March-April 1997

They evaluated approximately 112 cities

(yielding 12,422 cells = 112 X 112 - 112) to
determine which cities they should include
in the system. They developed data on origins from statistics brought by S. Tucker
Taylor, a former consultant to Smith and
member of the first model team, from the
Civil Aeronautics Board on enplanement
volumes and from other sources on such
factors as population, employment, and
business activity by SIC codes. They adjusted gross enplanement data for each of
the 112 origins by subtracting out heavy
freight traffic and then further massaged it
on the basis of assumptions they made using SIC Codes (revealing types of businesses and, therefore, indicating potential
for small package traffic), size, weight, and
so forth. They used all of this to establish a
coefficient of outbound market potential.
Since no destination data was available,
they assumed that a city's inbound market
potential was proportional to its percentage
of the total population of the 112 cities.
They multiplied the outbound potential coefficient for the first city (the origin) by the
inbound coefficient for each second city
(the destination) to obtain a raw estimate of
the one-way total volume for each traffic
lane. They hung huge sheets of butcher
paper on the walls and, with the aid of a
hand-held HP-35 calculator, one by one
filled out the cells of the matrix by hand.
Next the team went through the Official
Airline Guide and the Air Cargo Guide to de-

termine the availability of day and night air

cargo flights, airline belly cargo, and other
competitive means of transportation. They
devised a simple set of rules and used
them to adjust the raw origin-destination
figures to obtain a FedEx expected "share"

19

MASON, MCKENNEY, CARLSON, COPELAND

of the predicted volume. Again they entered the figures by hand.
The resulting matrix was extremely revealing. It showed that the types of businesses a city had and the nature of its air
cargo competition could affect its attractiveness for FedEx. Rochester, New York,
was an example of a city with a quite vibrant economyKodak, Xerox, and other
volume shippers of small parts were located therethat had poor air cargo service. It proved to be an inviting, economically viable location. The analysis showed
that several other cities had good potential
and poor existing service and that FedEx
could expect to get about 30 percent of
their small package business. Some cities in
the original set of 11, like Jackson, Mississippi and New Orleans, paled in comparison. Delta Airlines, which carried some
small packages on passenger fiights, provided good service to these cities; consequently, FedEx could expect to capture
only about five percent of their market because they lacked small package volume.
As a result of these analyses, they combined the newly identified cities with a few
major market centers, such as New York
City and Chicago, to define a 26-city system, one that was quite different from the
original 11-city system.

lion items to over 200 countries each business day. FedEx employs more than
110,000 people worldwide and operates 500
aircraft and more than 35,000 vehicles in its
integrated system.
The first model and the planning system
that evolved around it have served as the
foundation for the $8 billion plus corporation that is the Federal Express Corporation
today. They also established corporate values for analysis and modeling that have become integral to the way FedEx does business. In the process, founder and CEO Fred
Smith has become a dedicated user and
supporter of operations research and in a
real way an active participant in it.

INTERFACES 27:2

20

Come FLY with Me

A delicate balancing act is embedded in

Smith's business concept. On the one hand,
the company must find the most lucrative
cities with which to do business. On the
other hand, it must pick up, transport, and
deliver packages from and to these cities efficiently under the constraints imposed by
overnight delivery. Soon after operations
began, the need to address the other side of
the equahon was apparent.
Norman B. Waite, a consultant with a degree in mathematics from Johns Hopkins,
joined Charles Brandon to address this
problem. The result was FLY, a deterministic model based on known parameters that
On April 17,1973, a rejuvenated FedEx
simulated the operation of aircraft and fabegan serving its 26-city system. In one of
cilitated the construction of flight schedthe great reversals of fortunes in business
history, the new plan of operations worked. ules. The model drew on a database that
contained information on the characteristics
Twenty-three years later, according to its
of airports throughout the US, the operathome page (www.fedex.com), "FedEx is
ing characteristics of the aircraft, loading
the world's largest express transportafion
times, unloading times, prevailing winds,
company, providing fast and reliable serflying
times, and such other factors as air
vices for important documents, packages,
and freight." It delivers more than two mil- traffic control delay times, taxi delay times.

FEDERAL EXPRESS
and instrument approach times. It simulated the flight of a given aircraft comprehensively from gate to gate. They used the
FLY model to calculate such factors as
flight times and engine cycles. They then
used the results to estimate crew requirements, maintenance schedules, brake component repair schedules, and other operational demands. A few months after the
first successful flight in 1973, Federal Express was renting time on an IBM 360-67
at National CSS Time Share, and Waite
wrote the original code for FLY for the 360
in PL/1.
Shortly after FLY became operational,
Brandon used the model to evaluate FedEx's expansion possibilities. Smith planned
to serve as many as 82 cities with 33 Falcons. The expansion plan, however, required additional capital. The $10 million
seed money Smith had begun with was
running out. Meanwhile, the second round
of flnancing was in jeopardy. FedEx had to
find new investors. But it had to overcome
two major hurdles to get them to sign up.
First, most potential investors had difficulty
with the hub concept. ("Do you mean that
a package sent from New York City to
Newark flies all the way to Memphis and
back?" ) Second, they wanted some proof
that they would get an adequate return on
their investments. Using FLY, Brandon constructed a model of a proposed new structure featuring 82 airports served by 33 Falcons. In a barnstorming tour. Smith, Art
Bass (who became president and chief operating officer in 1975), and others used the
82 airport model to convince investors and
employees alike that the company's basic
concept was sound and that the new system was economically attractive. With the

March-April 1997

aid of the model, the team obtained the

cash it needed and added the new cities
and aircraft to the system. At the time it
was the largest venture-capital start-up
flnancing in US history.
With FLY, the team could ask what-if
questions and test its ideas. In one influential test, Brandon, as director of management information systems, demonstrated that Los Angeles was a "magic"
city, a prime place for the next expansion.
The company's resources were tight, and
Smith posed the question, "If we could
obtain just one new plane, what city
should we add?" They used FLY to perform what economists call a marginal
analysis, and it showed that Los Angeles
drew and pumped substantial volume
from and to other cities already in the
system, creating economies of scale. Los
Angeles had flve to 10 times more positive impact on FedEx than the next highest city. FedEx made plans to include LA
in its route structure.
The Three-Model Planning System
With these successes, modeling became a
way of life for FedEx. Soon it developed a
three-model management planning system:
(1) It used an improved origin-destination
flow model to determine the what, when,
and where of package volumes from and to
actual and potential cities in the system.
(2) It used FLY to produce schedules and
determine resource requirements for these
cities. It also ran PT^Y using actual past volumes to review performance, to test other
options, and to recalibrate its coefflcients.
(3) FedEx created a flnancial planning
model to show the overall economic and
flnancial implications of alternative route
structures and flying schedules.

21

MASON, MCKENNEY, CARLSON, COPELAND

It used these three models in concert to
make ongoing operational decisions as well
as crucial strategic decisions. During the
fall of 1973, for example, the three-model
system more than paid for itself. Heretofore
the price of fuel oil ranged between $2.75
and $3.50 per barrel. But, in October, the
Arab members of OPEC announced a drastic reduction in its production of crude oil
and an embargo directed at the United
States and the Netherlands. World crude
prices nearly quadrupled by the end of the
year. Fuel was rationed in the US, and it
appeared that FedEx would soon be driven
out of business. Convinced that he needed
to protect his company from impending
disaster. Smith headed for Washington,
DCbut not before requesting the OR
team to run its models based on different
assumptions of fuel oil prices and availability.

Economics and Modeling Systems

During the early 1970s, the underlying
economics of FedEx's business became
clearer. FedEx uses high-cost equipment
and facilities intensively for a short period
of time each night. Efflcient asset utilization
is made even more difflcult by a set of
precedence conditions, such as "the first
plane cannot leave until the last package is
sorted." Because the system's activities are
interdependent, speeding up one activity
does not necessarily improve anything else.
The only economical solution is for each activity to carry as heavy a load as possible,
but not so heavy as to exceed the capacity
of the assets allocated to that activity. Consequently, capacity planning is essential,
and operations must be conducted with a
very high load factor.

Brandon, with the help of Bill Arthur, estimated the growth in takeoffs, landings,
and average flight length for the next flveyear period and fed this data into the models to determine fuel consumption and
costs. They added an optimistic fudge factor, just in case. J. Tucker Morse, FedEx's
internal company counsel, immediately relayed the results to Smith in Washington.
Smith's first visit was to Representative
Wilbur Mills, who, after seeing the results,
called Secretary of Energy John Sawhill.
Sawhill reviewed the computer estimates
and approved a special allocation for
FedEx, enough to keep the 82-city network
operating for several years. As it turned
out, because of the team's generous assumptions on consumption, FedEx received
a greater fuel allocation than it actually
needed during the panic.

It is challenging enough to develop one

high-load-factor schedule for a stable system, but the problem Brandon and his
crews faced was even more daunting. Because FedEx was adding customers, cities,
and aircraft constantly, it needed to radically reschedule its routes at least once
every month, while simultaneously insuring that every flight had enough capacity to
keep the system from blowing out and
leaving packages on the ground. Many
common carriers operate at load factors
hovering between 60 and 69 percent. With
the aid of the three-model system, FedEx
was generally able to maintain a load factor
between 82 and 93 percent (its average was
87 percent) while reconflguring its route
structure every month, implementing its
schedule changes within a few days, and
keeping its flnancial performance in line.
This was an accomplishment. As late as
1991, another major air freight carrier took

INTERFACES 27:2

22

FEDERAL EXPRESS
over a month to do the same calculations
and to reschedule its operations.
Building the OR Team: Ponder and
Hinson
Operations research was to be central
and ongoing at FedEx. Brandon sought
help and in 1974 hired Dr. Ron Ponder of
Memphis State as a consultant. Ponder's
student, Joe Hinson, also worked as a consultant during the summer of 1974. The following August, Hinson became a full-time
employee. Ponder joined FedEx as director
of OR in 1977.
In addition to his skills in applying OR,
using GPSS tools for model building, and
his expertise in database technology.
Ponder proved to be a very capable manager. He built an OR team of 25 that became known as one of the finest OR groups
in the world. When Ponder became a senior vice-president in 1980, Hinson became
managing director of OR. He now reports
directly to Fred Smith.
Joe/Engine
The first project Joe Hinson tackled in
1974 was scheduling maintenance on the
General Electric CF-700 turbofan engines
used in the Falcons. Since the engine had
been used primarily in executive jets, it had
not logged the hours or cycles necessary to
get FAA approval for extended intervals
between maintenance sessions, a factor
known as time between overhauls (TBO).
Without a large statistical database from
which to estimate reliability, the FAA set
FedEx's TBO on the Falcons at a stringent
600 hours. Overhauls took about two to
three weeks, and since each plane flew
about six hours a day, overhauls were required every 100 days. This was a very
tight constraint for FedEx's small fleet.

The model showed that there were not

enough engines in the world to fly FedEx's
fleet hours given its forecasted continual
need to replace bluckets. The OR team,
working closely with the M&E Department,
used the model to manage a pool of engines (FedEx rented some additional engines) and averted the crisis by developing
strategies for swapping engines from aircraft to aircraft and by dynamically rescheduling overhauls. With Joe/Engine,

March-April 1997

23

Joe Hinson built a model to develop an

acceptable overhaul schedule. Officially
called "Engine," it was fondly known
throughout FedEx as "Joe/Engine," because of the computer file name that appeared at the top of every printout. Its results were alarming. Given FedEx's
expanding schedule, the model revealed
that the required removal rate would soon
surpass the company's total fleet capacity
and bring operations to a halt.
Further events deepened the crisis. The
FAA and GE notified FedEx's maintenance
and engineering (M&E) department that
GE had used a defective die to cast a critical part of the engine called the "blucket."
As a result, this rotating blade tended to
fail more frequently than expected. It appeared as though a blucket might fail every
50 cycles of use. A cycle is completed every
time an engine is started, idled, powered
up, idled, and then shut down. FedEx's exposure was much greater than most other
users because its planes took off and
landed so many times. A typical FedEx
plane consumed several cycles a day, implying that every plane had to be taken out
of service at least once a month. Worse, a
blucket failure often destroyed the entire
engine.

MASON, MCKENNEY, CARLSON, COPELAND

FedEx kept its fleet in service until GE developed an improved blucket, and the crisis
was averted. Subsequently, EedEx set up an
engine reliability program and compiled a
database of EedEx's actual performance.
Based on this new evidence, the FAA eventually approved a 3,000 hour TBO for
FedEx's CF-700 engines.
How about Pittsburgh?
Hinson determined that FedEx needed a
forecasting model to bridge the gap between the origin-destination model and
FLY and to sharpen its estimates of daily
package counts for each city. Although
FedEx had only about eight months of actual operations data at that time, Hinson
acquired Air Transport Association's gross
data and devised a way to create a seasonal
index and to estimate growth factors. Using
a crude form of exponential smoothing, he
introduced each night's actual counts into
the model every few days or so to keep its
coefficients current. Early in 1975, he translated the FLY model for the newly acquired
Burroughs B-6700 computer and programmed the forecasting model to accompany it.

business, a bypass hub receives, sorts, and

redistributes packages without forwarding
them to the central hub. That is, it operates
in parallel with the central hub and thus
functions differently from a set of hubs operating in series in which packages are
moved from hub to hub. Working together,
Mike and Joe ran various simulations using
the models and determined that a mini-hub
in Pittsburgh would indeed be economical,
but only if FedEx could use DC-3s for supplemental lift. The company had no DC-3s
at the time. Consequently, it let contracts
with local DC-3 operators and soon became
the largest charter operator of DC-3s in the
US. Pittsburgh was the first multiple hub to
be run in parallel with Memphis, establishing an approach to multi-hubbing that is
still in use for subhubs. Significantly, as
FLY and related models have been expanded, the OR department has used them
to evaluate many other proposals for
additional subhubs.
Autoroute

FLY, coupled with the forecasting model,

became a kind of early warning system. In
March, the models predicted that with a
fleet of only 33 Falcons sometime during
the fall of 1975 FedEx would run out of lift,
that is, its package volume would exceed
its total aircraft capacity. Despite the fact
that FedEx was barely eking out a profit, it
had to do something.
Mike Staunton (then head of scheduling,
now vice-president for global operations,
scheduling, and control) suggested that a
bypass hub in Pittsburgh might reduce
some of the pressure. In the transportation

The Pittsburgh solution proved to be

temporary. FedEx faced an even bigger crisis, almost immediately. Late in 1975, the
company would need planes with a much
greater payload than the Falcons if it was
going to handle anticipated increases in
volume and earn an acceptable profit, but
CAB regulations still prohibited companies
like FedEx from using them. CAB consequently turned down FedEx's initial request to buy larger DC-9's.
The aircraft capacity problem went to the
top of Smith's agenda and became the subject of almost daily conversations between
him, his assistant Tex Weise, Staunton, and
Brandon (then vice-president for operations
planning). Reporting to him were airline

INTERFACES 27:2

24

FEDERAL EXPRESS
scheduling, headed by Staunton; telecommunications, headed by Jack Cockrill; and
OR, headed by Ponder. He asked Hinson
and three other OR personnel to study the
capacity issue.
As has become the custom at FedEx,
OR was represented in the weekly senior
management meetings. During this era,
Brandon attended the meetings, often accompanied by Ponder or Hinson or some
of the other members of the OR team. This
special role was accorded the OR group because Smith, who has a degree in economics from Yale and is a liberal arts and military science buff, had learned about how
OR worked, was impressed with the results, and had come to trust the advice he
received from the OR team. The payload
problem was so hot that Smith met with
the OR group often for extensive discussions between the regular executive sessions.
The team had originally built Autoroute
as a research tool to help people understand the underlying structure of the aircraft scheduling problem. Joe Hinson beHeved that it would provide valuable
background knowledge for solving some of
the company's operations planning problems, and Brandon had encouraged him to
explore the possibilities. Now, they used
the model to determine the extent of the
impending crisis.
Autoroute is a heuristic optimizing,
scheduling model that takes loads, cities
served, and constraints, such as the latest
hour an aircraft can depart a city and the
earliest it can arrive at its destination, and
generates a complete single-hub-system
flight schedule. Over several months,
Hinson and his OR team made repeated

March-April 1997

runs of the model, making different assumptions about the composition of the
fleet, usually by substituting combinations
of two or three different types of aircraft at
a time. No matter what combinations they
used, the result was the same: FedEx could
not remain profitable unless it was allowed
to use larger planes.
Smith was troubled and moved into action. He took an apartment in Georgetown,
leaving COO Art Bass to run operations in
Memphis, and began a series of wide-ranging lobbying rounds in Washington, DC.
Virtually no one connected with airline or
air cargo regulation escaped his attention.
This effort took nearly a year. Armed with
the Autoroute results. Smith became a
strong advocate for an air cargo reform bill.
In March 1977, he testified before the senate aviation subcommittee and asked CAB
for a certificate of public convenience and
necessity. He appeared before the subcommittee again in August, testifying in support of regulatory reform. This modelbased lobbying eventually got results. In
November, the House of Representatives
passed the Air Cargo Reform Bill, and soon
thereafter President Carter signed it into
law. FedEx almost immediately filed for
permission to fly larger aircraft. It used Autoroute to evaluate the use of Boeing 727s,
among others, on FedEx's high volume
routes, and in December 1977, placed an
order for seven 727s.
Human Resource Planning
The 727s brought about new problems
with personnel planning. As FedEx expanded, introduced more hubs, and
launched a more diverse fleet, it had more
difficulty making sure its jobs and pay
scales were in order. Of special concern

25

MASON, MCKENNEY, CARLSON, COPELAND

each city station bought or leased its own
equipment and ran its own phone system.
Moreover, station managers jealously
guarded their control over the process.
"Our customers," a typical comment went,
"want to talk directly to their favorite agent
right here at home." Unfortunately, there
was little call-answering discipline and
there were no established standards. Vital
information was constantly being lost. In
short, it was chaotic. One day Tucker
Taylor walked into Charles Brandon's office and asked "Why can't we have our
own Wizard of Avis?" The "Wizard of
Avis" was a computerized reservation system for car rentals that featured a centralized telephone system for accepting and
routing all reservations throughout Avis's
system. The possibility of a wizard seemed
to Brandon like a question that OR could
answer, and he moved quickly to propose a
project to Smith. At first he ran into a stone
wall. There was a sharp division of opinion
within the company on this matter, and
The model calculated the pay for pilots,
only after much heated discussion did
first officers, and others and was used to
show them what their expected career path Smith approve the development of a prototype dubbed Project Sydney (for Sydney
at FedEx would look like. Since the company was expanding, the future was prom- Tucker Taylor). FedEx formed a project
team consisting of representatives from
ising for most pilots. This result proved
EDP, telecommunications, field operations,
very useful in labor-management negotiations. FedEx also used the model to forecast sales and marketing, and OR. OR began the
initial modeling, which would serve as the
gaps between the supply and demand of
foundation for building the system.
personnel and to indicate what its future
hiring and training needs were. This huSome preliminary model results sugman resource planning became an integral
gested that a central telephone answering
part of the company's overall planning
system was the best solution. A prototype
process.
for a centralized system was programmed
Perhaps We Need a "Wizard!"
for the Burroughs B6700. In late 1977,
FedEx conducted a test in Newark, at the
As FedEx's volumes grew, it became intime the worst-performing city in the syscreasingly difficult to handle requests and
tem on the basis of company statistics and
inquiries from shippers. Until late in 1977,

were its pilots, who had to be qualified on

each new aircraft and who had many job
opportunities elsewhere. The pilots considered unionizing, and the company wanted
to show them what they could reasonably
expect in career earnings and types of aircraft to fly.
The problem was formulated as a multiperiod, multi-stage stochastic (Markov
chain) transition model. The OR team used
fleet and operating requirements for a decade or more in the future to lay out pilot
requirements by aircraft type for each period. The characteristics of the existing
crew complement served as the starting
point. The model then simulated the movement of each pool of pilots in a given class
through these periods. Statistical data was
used to determine the probability that a pilot would fail a medical examination, fail to
upgrade, accept a job with another company, and other events that could affect his
transition from onie state to another.

INTERFACES 27:2

26

FEDERAL EXPRESS
employee performance, and in Memphis.
The prototype system, though limited in
capacity and reliability, was a vast improvement over the existing system. "For
the first time, the people in the stations
could do their own work," one executive
observed, "because they were relieved of
answering the phones and processing the
orders." Customers were also served better.
Other cities immediately demanded that
they too be added to the system. Despite
some known flaws, FedEx rolled the prototype out to 10 or 11 other cities. Ultimately
FedEx handled all customers' incoming
calls in centralized call centers, which then
sent out requests to dispatch centers located in each city FedEx served.
Given the stunning success of Project
Sydney, Smith approved a project to build
a full-scale production real-time system
eventually called Cosmosto serve the entire FedEx system. At the time, there were
many unresolved questions concerning
how many call centers to establish, where
to place them, and how to staff them.
Hinson and Ponder, with the aid of OR analyst Michael Stemad, built a customer service system model to explore the possibilities. The model clearly showed that length
of call was the most important factor in determining call-handling performance.
Queuing analysis further revealed that it
was quite easy to overload the system and
to bring it down. In fact, a sensitivity analysis applied to the model, revealed that the
length of the call was the main cost driver
for the entire system. Slight decreases in
the average length of call resulted in substantial cost savings; slight increases resulted in a substantial cost increases. Further studies suggested other methods for

March-April 1997

reducing the time each call took, such as

improving the agents' answering behavior,
routing calls more efficiently across time
zones, and modifying other procedures.
FedEx used these results to develop an
overall business plan for both the telephone
system and the computer systems and to
determine the optimum locations for the
call centers. The call center study results
served as the fundament for the subsequent
development of Cosmos.
Ultimately, FedEx decided to build more
call centers than the model indicated because its executives believed that the additional centers provided more flexibility,
avoided possible regulatory issues, and
would serve as a positive labor relations
moveall factors not fully accounted for in
the original formulation, but which
emerged during intense discussions with
Smith and others. Such give-and-take discussions, in which model results are pitted
against executive understanding, have
become common practice at FedEx.
Hubba, Hubba

Operations research influenced one decision that had an enormous impact on

FedEx and set its course for over 15 years.
This hotly debated decision was to build a
SuperHub, and as Charles Brandon observed, "OR was applied effectively here
and reversed the four-hub serial design decision in favor of a single SuperHub. It saved
the company!"
When the network was small, its volumes low, and the Falcons the only aircraft,
a single hub in Memphis was clearly the
best choice (although the Pittsburgh experience had shown the value of parallel hubs).
However, the dramatic growth of the business and the advent of the 727stogether

27

MASON, MCKENNEY, CARLSON, COPELAND

with the possibility of even larger aircraft
in the futureraised questions as to
whether this strategy was still optimal.
Indeed, several key executives were convinced that a series, multihub system
most favored fourwas the way to go.

Four hubs appeared to be a

done deaL
They reasoned that FedEx could use small
aircraft to carry packages to each regional
hub and that it could use larger planes to
haul packages from hub to hub. A great
deal of momentum and enthusiasm built
up within the company as people maneuvered for a four-hub network and, as one
executive put it, "Four hubs appeared to be
a done deal."
Hinson, Ponder, and Brandon were not
convinced. OR had not yet studied the
problem in any depth. And so, as the front
office moved toward a four-hub system,
they began to model the situation. Their
first task was to expand their modeling approach. Autoroute, it turned out, could be
used to estimate the performance of feeders
into a regional hub, but it was incapable of
determining whether or not a given hub
should remain in the final system network.
Since the final solution must contain a discrete number of hubs, they needed an integer linear programming model (ILP). Consequently, they constructed a two-phase
model consisting of Autoroute to perform
within hub analyses and an ILP to choose
from among the candidate hubs. They
made exploratory runs assuming between
one and eight hubs and also addressed fleet
size and composition issues.

INTERFACES 27:2

It was standard practice to discuss even

very preliminary OR results in senior management committee meetingsFred Smith
was an especially eager participant. This dialogue served to change assumptions in the
model and to suggest other approaches.
The first model runs indicated that a single hub was still the preferred solution.
Every time this solution was presented,
however, a heated debate ensued. Smith
and others continually came up with new
ideas about the cost of aircraft, the nature
of the facilities used, and the like. With
these new considerations in hand, the OR
team would head back to its quarters to
modify the model and to run it a few more
times. This continual dialogue, pitting executive assumptions against their implications as derived by a model, lasted for
about a year. Throughout this time, the single hub solution generally dominated in
the model results but not, it turned out, in
many executives' minds.
In the end. Smith decided on the singlehub systemnow dubbed the SuperHub.
The model revealed the underlying economics clearly. With high cost aircraft and
facilities used intensively for compressed
periods of time under conditions that demanded near perfect performancethat is,
failure to service a plane or to pick up a
batch of packages could not be tolerated
the single hub worked best. Given the technology and volumes at the time, the
SuperHub design was more flexible; it required only one big sort, reducing total
sorting time, and it made crew domiciling
easier. The four-hub approach had three
major flaws: it caused low utilization of
high cost assets and, hence, had a poor ROI
(in some cases negative), it had a low toler-

28

FEDERAL EXPRESS
ance for faults because it did not have the
built-in redundancy that the single hub did,
and it consumed almost all of the 12 hours
FedEx had available to process work to
meet its commitments. Hence, it did not allow much slack nor did it provide much
room for growth. Memphis remained the
single hub in the system, and in July 1979,
FedEx broke ground for the massive,
highly automated SuperHub.
Soon after the SuperHub was opened,
the model revealed additional advantages
to the concept. As a competitive move.
Smith wanted to change the committed delivery time from 12:00 noon to 10:30 am.
The model showed that the FedEx system
could deliver on that promise. Furthermore, a competitive analysis application of
the model revealed that UPS, FedEx's major competitor, could not effectively meet
this deadline. In late 1982, Tom Oliver,
head of FedEx's marketing, announced,
"We will offer 10:30 delivery, more service
options, Saturday pickups, package tracing,
and call-backs to shippers informing them
that the packages have been delivered,"
and he did this knowing that his company
could deliver on this promise and that, for
the foreseeable future, UPS could not. All
of this was consistent with Smith's strategic
emphasis on quality and reliability of
performance.
The technological infrastructure of the
company was now pretty much in place: a
mixed fleet of small planes providing service to the smaller stations and larger aircraft on the higher volume runs funneling
through a SuperHub plus centralized call
centers for answering incoming phone
calls, which were in turn dispatched to local stations for response. This infrastruc-

March-April 1997

ture, which was predicated on OR results,

called for a much different approach to
FedEx's organizational design. Smith imderstood this, and in April 1978, he reorganized the company from its previous decentralized form into a highly centralized
company, providing as an explanation that
"various approaches were modeled against
our present and future tasks."
Tracking toward the COSMOS

As volume growth accelerated, the problem of making the stations work better and
servicing customers quicker and in a more
friendly manner surfaced. The excellent results obtained from the Project Sydney prototype and the success that the airlines had
had with computerized reservations systems suggested an approach. With this in
mind, Brandon, by now senior vice-president for planning and information systems
(including electronic data processing),
sought out and hired Howard Bedford, an
expert in airline reservation systems, to develop a comprehensive system. Bedford
hired James ToUefson, who had experience
at IBM and Avis; Henry Howell; and a
crew that soon reached about 150 people,
most of whom were also pioneers in highspeed, on-line transaction processing systems. The group was ensconced in the Advanced Systems Development Center
(ASD) located in Colorado Springs. The result was COSMOS.
COSMOS (Customer, Operations, Service, Management Operating System) is a
mainframe-based order and dispatching
system that FedEx has constantly upgraded
to cope with increasing volumes, an expanded route structure, and an ever increasing emphasis on tracking packages
from origin to final destination. Version 1

29

MASON, MCKENNEY, CARLSON, COPELAND

began operafions early in 1979. It was designed with the idea that the system eventually would be used to keep track of every
single package throughout its life in the
system. The source of this vision, of course,
was Smith.
From the very beginning, Fred Smith
wanted to maintain constant surveillance
on every package that entered the FedEx
system. "Absolutely, Positively, Overnight"
was his business battle cry and still is. Thus
the continual tracking of packages as they
move through the FedEx system is a fundamental, critical success factor of the business.
Early in 1974, Smith had come across a
gas station machine that convinced him
that a package tracking system might be
technologically feasible. FedEx designed a
system called "Star" using a gas station imprinter and multi-part forms. Unfortunately
an experiment run in Chicago was an abject
failure, largely because of the amount of
paper and paper shuffling required to handle each package. The Star experiment was
squelched; but the dream of 100-percent
tracking lived on.
By 1980, with COSMOS and the
SuperHub system in place. Smith renewed
his demands for tracking, and ASD went to
work on it. The result was COSMOS IIA
the first truly positive tracking and status
information systemwhich went on
stream in February 1981.
Version IIBa real-time, point-of-sale to
point-of-delivery system with distributed
process controlfollowed in January 1986.
Today, with COSMOS IIB, a courier uses a
handheld computer (the SuperTracker)
with a wand to scan a bar-coded label
when a package is picked up, and the vital

information is uploaded to COSMOS. This

scanning is repeated five or more times at
various stages of handling until the package is finally delivered. The last wave of
the wand records the delivery to its intended recipient. A companion system
called DADS (Digitally Assisted Dispatch
System) receives shipping order information from COSMOS, sends it to a city's dispatch center computer, and routes it to the
appropriate courier. The request is displayed on a small DADS video screen in
the courier's van or on a portable unit the
size of a small briefcase. COSMOS, and to
some extent DADS, are predicated on the
results that came out of early OR studies.

INTERFACES 27:2

30

Releasing the "Butterfly"

Fred Smith characterizes his company as

having gone through three entomological
phases. About four years beginning in 1973
mark its caterpillar phase during which its
real potential and future were fomenting
within itself. A short 11 months beginning
in January 1977 constitute its chrysalis period, when Smithwith the aid of data obtained from operations research studies
struggled with legislative committees and
government agencies in Washington to rid
the company of its regulatory constraints.
As with any lepidopteran, there was a great
deal of activity going on inside the cocoonin this case considerable OR and
marketing planning, but little of it could be
seen from the outside. Finally, there is the
current butterfly era, which began in November 1977 when the Air Cargo Reform
Bill was signed into law, and FedEx was set
free.
The butterfly period really began to take
off early in 1979, and it presented many
new challenges and opportunities for oper-

FEDERAL EXPRESS
ations research. The existing package-handling system, young as it was, was already
coming to the end of its useful life. Nightly
package counts exceeded 50,000 and were
growing rapidly. How long would the
SuperHub concept remain optimal? What
kinds of air cargo carrying capacity would
FedEx need in the future? How fast must
planes fly to meet their tight deadlines?
Joe Hinson and his group set out to answer these questions. Using a multi-period
model of the entire FedEx system and looking five years out into the future, they first
estimated the system's carrying capacity,
the maximum amount of freight that the
system would have to carry each night.
One million pounds of total capacity appeared adequate and even provided a little
breathing room. (This seemingly conservative maximum load factor was exceeded
earlier than anticipated.) They developed
several scenarios by making different assumptions about package counts, their distribution, transportation modes (types of
aircraft and trucks), and alternative huband-spoke structures. They calculated costs
and performance factors for every activity:
the original shipper request, package
pickup, local transport, outbound station
sort, outbound line haul (air or ground),
hub sort, inbound line haul, inbound station sort, local transport, and final delivery.
Once every few weeks, Hinson would trek
across to headquarters and present his
team's preliminary results to Smith and the
executive committee. Some of the executivesalmost always Smithwould think
of some new possibilities. "What if this
were the case?" or "Couldn't we do that?"
Hinson dutifully took notes, and as each
session came to a close, he would return to

his team to begin a new set of analyses.

This give-and-take dialogue lasted about
six months during late 1979. Slowly, a few
crucial yet robust results emerged. One was
that the company could save on fuel expenditures if it employed DC-lOs on highvolume routes to such cities as Boston, Chicago, Los Angeles, and Newark. At first.
Smith balked at this solution. For once the
company was financially sound enough to
make the investment, but he thought the
price was too high. One DC-10 cost nearly

March-April 1997

31

Everything is tied to
everything else.
as much as his entire fleet of Falcons had
cost. The OR team changed its approach.
Using an updated version of Autoroute, the
team calculated the price point at which
FedEx should be indifferent between purchasing DC-lOs and another bundle of aircraft. The results rekindled Smith's interest,
and, armed with this new information, he
made an offer for four DC-10-lOCFs. The
seller initially demurred. But, because he
understood the economics of the deal very
well. Smith held tight throughout the negotiations and finally got a purchase price the
model told him he could live with. In
March 1980, FedEx christened its first
DC-10.
Another result of the butterfly explorations was even more controversial, because
it reopened old wounds from a previous
internal company battle. Various model
runs predicted that sometime during 1987
the SuperHub concept would finally reach
its practical limit. FedEx would need more
hubs. As long as the 727s could handle the

MASON, MCKENNEY, CARLSON, COPELAND

volume, the single hub design remained
superior. The DC-lOs extended its life
somewhat but eventually, as volumes expanded, the SuperHub would collapse of
its own weight.
Multi-hubs were in FedEx's future, but
when, where, and how should they be operated? The models showed that the most
efficient approach was somewhat different
from that proposed during the first corporate debate. The original serial approach
called for collecting all of a region's packages in a regional hub, fully sorting them
there, and then sending packages to destinations within the region while routing all
other packages to the appropriate other
hubs. This placed a heavy sorting burden
on the system. A more efficient approach, it
turned out, was for each pickup station to
make a simple binary decision. Sending
packages for redistribution within the region to the regional hub and sending all
others to the Memphis SuperHub. This approachcalled the overlay hub method
distributed the sorting burden and permitted it to take place in parallel rather than
serially.
The team then used economic location
models to determine the best sites for the
overlay hubs. Early results showed that the
West Coast was a good candidate for an
overlay hub, the center of gravity resting
just north of Los Angeles. Autoroute runs,
however, modified this conclusion. Because
of the flight times from Seattle, Portland,
and other Pacific Northwest cities and a
few other operating considerations, Autoroute indicated that the San Francisco Bay
area was a better spot. In June 1986, FedEx
opened the first overlay hub in Oakland,
California. This was at least a year earlier

INTEREACES 27:2

than the models had indicated would be

optimal. However, Smith was convinced.
He wanted to move ahead quickly and to
test the concept for real. It has been successful.
Locating the Indianapolis overlay hub
was similar. The location model favored
Rockford, Illinois. This assumed that FedEx
would have to build its own facilities for
each of its overlay hubs as it had done in
Oakland. In August 1987, however, a fully
furnished airport facility became available
in Indianapolis. A few quick runs of the
models showed that the Indianapolis site
was economically viable. Smith bought the
Indianapolis facility and began operations
in October 1988.
The transportation models uncovered another advantage of the overlay hub concept. The economics of hauling show that
for short distances trucks are the most efficient carrier. As the distance increases a
switch-over point is reached at which air
cargo becomes less costly. The original distribution and configuration of stations favored air cargo. Nevertheless, Smith and
his associates had been looking forward to
using trucks in high volume areas. Autoroute showed that FedEx had reached that
time; the density of traffic around the new
overlay hubs made trucks more economical
for certain short hauls. FedEx initiated
truck service to replace other modes on
these routes. Today FedEx is one of the nation's largest integrated, multi-conveyance
freight carriers.
Some Current Projects

After extensive studies by the OR team

in July 1991, FedEx ordered 25 A300-600
Airbus freighters. The first deliveries were
in April 1994. An updated version of Auto-

32

EEDERAL EXPRESS
route shows that the faster cycle time,
higher payload, longer range, and capacity
for intensive operations should make this
aircraft effective on high volume and international routes.
Meanwhile, the OR team is focusing on
the least-studied part of the total system
ground pickup and delivery in the field.
This activity accounts for about 55 to 65
percent of the total cost of packagehandling services.
An experimental courier route planning
system project is underway. FedEx is using
geo-coding and geo-positioning technology
to plot a courier's pickups and deliveries
on a map route by longitude and latitude.
One use of the map is to show the courier
the exact route she followed and to suggest
possibilities for more efficient sequencing.
Soon to follow are shortest time path estimates. These will help station managers to
plan the work of couriers under their supervision. Preliminary estimates show that
a five to 10 percent productivity improvement is possible, and almost all of it goes to
the bottom line.
Another application involves plotting
an entire station's deliveries by route
number on the station's service area map.
This will reveal operating problems in the
route structure. Since customer volumes
are always changing, and new customers
appear and old ones disappear, route balancing is a constant challenge. Turfs, as
couriers' routes are called, are currently
plotted by hand, but when the geo-coding
experiment is completed, it is likely that a
computer program can be devised to do
the task more effectively. Ultimately, the
map information may be integrated into
DADS.

The Federal Express Corporation has

succeeded by applying the scientific
method to its operations. Models and analysis have informed many of its crucial,
business-shaping decisions. When the company didn't use OR, beginning with its
very first trial run, it performed poorly.
Why is this? What explains FedEx's success? What lessons can we learn?
OR has been so successful at FedEx for
five major reasons:
(1) The airline and air cargo industries are
whole systems businesses with high levels
of interdependencies among their parts.
They are not easily broken into little pieces
to be managed separately. In effect, everything is tied to everything else. These interdependencies are exacerbated by the demands of "Absolutely, Positively,
Overnight" delivery. Because of the underlying systemic structure of the business, its
route structures and operations are amenable to modeling and analysis.
(2) Fred Smith, the founder and CEO, is
devoted to the use of scientific inquiry in
his business. In the beginning, he sought
out Charles Brandon, listened to him, and
gave him responsibility. Brandon became
the maestro who translated Smith's vision
into OR studies and information and communications systems. With Smith's approval, Brandon hired and initially directed
Ponder, Hinson, and others who eventually
became part of a world-class team. Ultimately, several senior corporate vice-presidents came up through OR.
Throughout the company's history.
Smith has supported the OR effort. The
acid test came early on. In 1973, the company had calamitous cash-flow problems. It

March-April 1997

33

Lessons to Be Learned

MASON, MCKENNEY, CARLSON, COPELAND

asked employees not to cash their paychecks and put all vendors on hold.
National CSS Time Share threatened to cancel FedEx's computer time unless it made
an $8,000 payment. Since the OR effort required the computers. Smith came up with
the money. Of equal importance. Smith has
always engaged the OR team in debate on
issues of crucial importance to the firm. OR
has a voice at FedEx.
(3) From the beginning, the OR team focused on issues that were crucial to the
business, such as equipment investment
and route structure. The team always tried
to take the simplest, most straightforward
modeling approach it could, as long as it
could come up with appropriate and actionable answers. It did not get sidetracked
in the esoteric or bogged down in minutia.
(4) The models and their results were considered to be living things, not imperial edifices cast in concrete. The OR team's pride
resided more in the company than in a particular model. Through these continual dialogues, the OR team and executive management jointly arrived at shared understandings
of the problems they faced, the pros and
cons of the models, and the strengths and
weaknesses of the model results. They
made trade-offs and compromises to
achieve FedEx's overall corporate goals.
The models informed these judgments.
Consequently, when Smith and his executives made decisions, they knew why.

tionsthat characterize all of the important

activities of the business: marketing, route
composition, package handling, hauling,
and finances. The OR team is constantly
updating these facts and assumptions as it
discovers new information about technology and operations and new executive
insights and desires.
Acknowledgments
This paper is a product of the Harvard
Business School MIS History Project. The
project is led by James McKenney of Harvard. The team comprises Walter Carlson,
Wchard Canning, George Glaser, and Ted
Withington, who have been associated with
information systems and operations research for nearly 40 years and are currently
consultants to industry. The other academic
members are Richard Mason of Southern
Methodist University and Duncan Copeland of the University of Western Ontario.
Some of the findings of the project are reported in Waves of Change [McKenney,
Copeland, and Mason 1995].

(5) Finally, the OR team conducted itself as

a learning organization, a center of knowledge and inquiry about the most important
dimensions of the business. Captured in
the models and databases are millions of
items of informationfacts and assump-

We based this article on information

gleaned from public sources and from interviews with Jim Barksdale, Charles Brandon,
Harry Dalton, Carolyn Freeman, Joe Hinson,
Dermis Jones, Tom Martin, Jim Moore,
Martha Outlaw, Ron Ponder, Fred Smith,
Mike Staunton, S. Tucker Taylor, Jim
Turpin, Edwin White, and Denise Wood, all
of whom work for or formerly worked for
Federal Express. Carolyn Freeman also
helped schedule interviews and provided
editorial suggestions. Charles Brandon, Joe
Hinson, Dennis Jones, and Fred Smith provided comments on an earlier draft. Nevertheless, the authors accept responsibility for
any errors in the manuscript.

INTERFACES 27:2

34

FEDERAL EXPRESS
APPENDIX: FedEx Summary of Key Events
Date

Situation

Information Source

Model

Decision or Use

March 1973

-Original idea

-Founder

-Initial 11-city model

-11-city schedule
(failure)

April 1973

-Soul searching and

rethinking

-Emplanement
statistics
-Employment SIC
-Business
population
-Experience

-112-city origindestination model

-26-city system
(satisfactory)

June 1973

-Cost pressures
-Draw on the
results of the
O/D model

-Airport data
-Aircraft
characteristics
-LoadingurJoading
-Ground
operations
-Weather
experience

-FLY
-Simulation model
produces expected
flight times and
engine cycles

-Refine schedules for

more efficient
operatior\s

-Additional data
on candidate
cities

-FLY and O/D models

simulate 82-city
system with 33
Falcons

-Successful
presentation to
potential investors
-Add Los Angeles to
system early

Early Fall 1973 -Desire to grow

-Need for additional
capital
-Investor confusion
Mid Fall 1973

-Need to add
financial
management to
operations
management

-Accounting data
-Financials
-Cost data

-Financial planning
model (Fin)

-Refine operations to
improve financial
performance

Late FaU 1973

-OPEC rationing
imposed

-Fuel consumption
and cost
estimates

-FLY, O/D, Fin models

predict five-year fuel
requirements for an
82-city system

-Obtain fuel allowance

during OPEC
restrictions

Summer 1974

-Need for engine

overhauls
threatens to
cripple fleet

-Flight time data

-Number of flights
-Engine cycles
-Overhaul
requirements

-Joe/Engine model

-Kept fleet in service

until bluckets
problem was solved

Summer 1975

-Company growing
faster than ability
to manage well

-Expected growth
-Seasonal factors
-Actual load data
-ATA gross volume

-Forecasting model, FLY, -Capacity alert bypass

and other models to
hub established at
Pittsburgh
predict operating
requirements

Winter 1975

-Increased
complexity of the
system

-Load factors
-Cities served
-Arrival and
departure times

-AUTOROUTE

-Air Cargo Reform Bill,

-FedEx orders seven
727s

December
1975

-Customer calls
overload system

-Customer call data

-Telephone volume
-Telephone
facilities
-Operating data

-Initial telephone
answering model

-Need for prototype

centralized system
established
-Project Sydney

March-April 1997

35

MASON, MCKENNEY, CARLSON, COPELAND

Early 1978

-Prototype
centralized
system

-Customer service model -Decision to initiate

COMOS and
centralize call
centers
-COSMOS Version I
implemented

Early 1979

-Expanded Autoroute
model

-SuperHub decision

March 1980

-Autoroute (and other

models)

-Add DC-lO's to the

fleet

February 1981

-Autoroute (and other

models)

-COSMOS Version IL\

implemented

Winter 1982

-Autoroute (and other

models)

-10:30 AM guaranteed
delivery decision

January 1986

-Autoroute (and other

models)

-COSMOS Version IIB

implemented

June 1986

-Autoroute (and other

models)

-Overlay Hub
established in
Oakland, CA

Spring 1991

-Autoroute (and other

models)

-Additional overlay
hubs
-Trucking justified

April 1994

-Autoroute (and other

models)

-25 A300-600 airbuses

ordered

Spring 1979

-More operating
problems

-System structure
-Cost of aircraft
-Facilities
characteristics

References
Sigafoos, Robert A. and Easson, Roger R. 1988,
Absolutely, Positively, Overnight! The Unofficial
Corporate History of Federal Express, St. Lukes
Press, Memphis, Tennessee.
McKenney, James L.; Copeland, Duncan C; and
Mason, Richard 0.1995, Waves of Change:
Business Evolution Through Information Technology, Harvard Business School Press, Boston,
Massachusetts.

INTERFACES 27:2

36