Professional Documents
Culture Documents
net/publication/247823764
CITATIONS READS
17 1,187
4 authors, including:
Richard O. Mason
Southern Methodist University
88 PUBLICATIONS 4,727 CITATIONS
SEE PROFILE
All content following this page was uploaded by Richard O. Mason on 04 September 2015.
—Frederick W. Smith
Pres W.(Fred) Smith developed a tial management team, and scraped to-
vision of a business, brought together gether enoughfundsto buythefirst air-
the key people who would makeuphis ini- craft for his fleet: French-made Dassault
INTERFACES 27:2 18
company discontinued air delivery service. They evaluated approximately 112 cities
To maintain its customers, the company (yielding 12,422 cells = 112 x 112 — 112) to
foundalternative ways to transport their determine whichcities they should include
packages while it took stock ofits situation. in the system. They developed data onori-
The First Model gins from statistics brought by S. Tucker
After getting over the initial shock, Fred Taylor, a former consultant to Smith and
Smith appointed a task force headed by memberof the first model team, from the
Charles Brandon, a physicist and an ardent Civil Aeronautics Board on enplanement
flyer he had met in 1962, when both were volumes and from other sources on such
homefrom college. When Smith was con- factors as population, employment, and
templating his new business in 1972, he businessactivity by SIC codes. They ad-
asked Brandon,“Could you use a com- justed gross enplanementdata for each of
puterto better scheduletheaircraft?” “‘Def- the 112 origins by subtracting out heavy
initely,” was Bronson’s reply. Brandon freighttraffic and then further massaged it
soon thereafter became an advisor to FedEx on the basis of assumptions they madeus-
and an early employee. Smith’s charge to ing SIC Codes (revealing types of busi-
Brandon wasstraightforward: Where have nesses and, therefore, indicating potential
wefailed? What should we do now? for small packagetraffic), size, weight, and
The team members closeted themselves so forth. They used all of this to establish a
in a conference room atLittle Rock’s Wor- coefficient of outbound marketpotential.
then Bankto begin their soul-searching, Since no destination data wasavailable,
working18 to 20 hours a day for 15 days they assumedthata city’s inbound market
straight. Their approach wasclassic. They potential was proportionalto its percentage
beganbychallenging the assumption that of the total populationof the 112cities.
the cities the founders knew best were best They multiplied the outboundpotential co-
for doing business. This led them to focus efficientfor thefirst city (the origin) by the
on whatcharacteristics madea city a good inboundcoefficient for each secondcity
candidate for their route structure, and (the destination) to obtain a raw estimate of
they concluded that they should include the one-waytotal volumefor eachtraffic
somecities because they generated large lane. They hung hugesheets of butcher
volumeof outbound small packages; others paper on the walls and, with the aid of a
becausethey received lot of small pack- hand-held HP-35calculator, one by one
ages. Moreover,thefinal set of cities filled out the cells of the matrix by hand.
should account for a substantial amount of Next the team went through the Official
the total volumewithin the system. Airline Guide and the Air Cargo Guide to de-
Giventhese guidelines, they next acted terminethe availability of day and nightair
on their intuitive ideas in a systematic way cargo flights,airline belly cargo, and other
to find data or indicators they could use to competitive meansof transportation. They
select cities and to decide on a route sys- devised a simpleset of rules and used
tem.First, they formulated the problem in them to adjust the raw origin-destination
termsof an origin-destination flow model. figures to obtain a FedEx expected “share”
March-April 1997 19
of the predicted volume. Again they en- lion items to over 200 countries each busi-
tered the figures by hand. ness day. FedEx employs more than
Theresulting matrix was extremely re- 110,000 people worldwide and operates 500
vealing. It showed that the types of busi- aircraft and more than 35,000 vehicles in its
nesses a city had andthe natureofits air integrated system.
cargo competition could affect its attrac- Thefirst model and the planning system
tiveness for FedEx. Rochester, New York, that evolved around it have served as the
was an exampleofa city with a quitevi- foundation for the $8 billion plus corpora-
brant economy—Kodak, Xerox, and other tion that is the Federal Express Corporation
volume shippers of small parts were lo- today. They also established corporate val-
cated there—that had poorair cargoser- ues for analysis and modeling that have be-
vice. It proved to be an inviting, economi- comeintegral to the way FedEx doesbusi-
cally viable location. The analysis showed ness. In the process, founder and CEO Fred
that several other cities had good potential Smith has become a dedicated user and
and poorexisting service and that FedEx supporter of operations research and in a
could expect to get about 30 percent of real way an active participantinit.
their small package business. Somecities in Come FLY with Me
the original set of 11, like Jackson, Missis- A delicate balancing act is embedded in
sippi and New Orleans, paled in compari- Smith’s business concept. On the one hand,
son. Delta Airlines, which carried some the company must find the most lucrative
small packages on passengerflights, pro- cities with which to do business. On the
vided goodservice to these cities; conse- other hand,it must pick up, transport, and
quently, FedEx could expect to capture deliver packages from andto thesecities ef-
only aboutfive percent of their market be- ficiently under the constraints imposed by
cause they lacked small package volume. overnight delivery. Soon after operations
As a result of these analyses, they com- began, the need to address the other side of
bined the newly identified cities with a few the equation was apparent.
major market centers, such as New York NormanB. Waite, a consultant with a de-
City and Chicago, to define a 26-city sys- gree in mathematics from Johns Hopkins,
tem, one that was quite different from the joined Charles Brandonto address this
original 11-city system. problem. The result was FLY, a determinis-
On April 17, 1973, a rejuvenated FedEx tic model based on known parameters that
beganserving its 26-city system. In one of simulated the operation of aircraft and fa-
the great reversals of fortunes in business cilitated the construction offlight sched-
history, the new plan of operations worked. ules. The model drew on a databasethat
Twenty-three years later, accordingto its contained information on the characteristics
homepage (www.fedex.com),“FedEx is of airports throughoutthe US,the operat-
the world’s largest express transportation ing characteristics of the aircraft, loading
company, providing fast and reliable ser- times, unloading times, prevailing winds,
vices for important documents, packages, flying times, and such other factors as air
and freight.” It delivers more than two mil- traffic control delay times, taxi delay times,
INTERFACES 27:2 20
and instrument approachtimes.It simu- aid of the model, the team obtained the
lated theflight of a given aircraft compre- cash it needed and added the newcities
hensively from gate to gate. They used the andaircraft to the system. At the timeit
FLY model to calculate such factors as wasthe largest venture-capital start-up
flight times and engine cycles. They then financing in UShistory.
used the results to estimate crew require- With FLY, the team could ask what-if
ments, maintenance schedules, brake com- questionsandtestits ideas. In one influ-
ponent repair schedules, and other opera- ential test, Brandon, as director of man-
tional demands. A few monthsafter the agement information systems, demon-
first successful flight in 1973, Federal Ex- strated that Los Angeles was a “magic”
press wasrenting time on an IBM 360-67 city, a prime place for the next expansion.
at National CSS Time Share, and Waite The company’s resources weretight, and
wrote the original code for FLY for the 360 Smith posed the question, “If we could
in PL/1. obtain just one new plane, whatcity
Shortly after FLY becameoperational, should we add?” They used FLYto per-
Brandon used the modelto evaluate Fed- form whateconomists call a marginal
Ex’s expansion possibilities. Smith planned analysis, and it showed that Los Angeles
to serve as manyas 82 cities with 33 Fal- drew and pumpedsubstantial volume
cons. The expansion plan, however, re- from andto other cities already in the
quired additional capital. The $10 million system, creating economies of scale. Los
seed money Smith had begun with was Angeles had five to 10 times more posi-
running out. Meanwhile, the second round tive impact on FedEx than the next high-
of financing was in jeopardy. FedEx had to est city. FedEx madeplansto include LA
find new investors. But it had to overcome in its route structure.
two major hurdles to get them to sign up. The Three-Model Planning System
First, most potential investors had difficulty With these successes, modeling became a
with the hub concept. (“Do you mean that wayOf life for FedEx. Soon it developed a
a package sent from New York City to three-model managementplanning system:
Newark flies all the way to Memphis and (1) It used an improvedorigin-destination
back?” ) Second, they wanted some proof flow model to determine the what, when,
that they would get an adequate return on and whereof package volumes from and to
their investments. Using FLY, Brandon con- actual and potential cities in the system.
structed a model of a proposed new struc- (2) It used FLY to produce schedules and
ture featuring 82 airports served by 33 Fal- determine resource requirements for these
cons. In a barnstorming tour, Smith, Art cities. It also ran FLY using actual past vol-
Bass (who becamepresident and chief op- umes to review performance, to test other
erating officer in 1975), and others used the options, andto recalibrate its coefficients.
82 airport model to convince investors and (3) FedEx created a financial planning
employeesalike that the company’s basic modelto show the overall economic and
concept was sound andthat the new sys- financial implications of alternative route
tem was economically attractive. With the structures and flying schedules.
March-April 1997 21
INTERFACES27:2 22
March-April 1997 23
FedEx keptits fleet in service until GE de- business, a bypass hub receives, sorts, and
veloped an improved blucket, and the crisis redistributes packages without forwarding
was averted. Subsequently, FedEx set up an them to the central hub. Thatis, it operates
engine reliability program and compiled a in parallel with the central hub and thus
database of FedEx’s actual performance. functions differently from a set of hubs op-
Based on this new evidence, the FAA even- erating in series in which packagesare
tually approved a 3,000 hour TBO for moved from hub to hub. Working together,
FedEx’s CF-700 engines. Mike andJoe ran various simulations using
HowaboutPittsburgh? the models and determined that a mini-hub
Hinson determined that FedEx needed a in Pittsburgh would indeed be economical,
forecasting modelto bridge the gap be- but only if FedEx could use DC-3s for sup-
tween the origin-destination model and plementallift. The company had no DC-3s
FLY and to sharpenits estimates of daily at the time. Consequently, it let contracts
package counts for each city. Although with local DC-3 operators and soon became
FedEx had only about eight monthsof ac- the largest charter operator of DC-3s in the
tual operations data at that time, Hinson US. Pittsburgh wasthefirst multiple hub to
acquired Air Transport Association’s gross be run in parallel with Memphis, establish-
data and devised a wayto create a seasonal ing an approach to multi-hubbing thatis
index and to estimate growth factors. Using still in use for subhubs.Significantly, as
a crude form of exponential smoothing, he FLY and related models have been ex-
introduced each night’s actual counts into panded, the OR department has used them
the model every few daysor so to keep its to evaluate many other proposals for
coefficients current. Early in 1975, he trans- additional subhubs.
lated the FLY modelfor the newly acquired Autoroute
Burroughs B-6700 computer and pro- The Pittsburgh solution proved to be
grammedthe forecasting model to accom- temporary. FedEx faced an even bigger cri-
panyit. sis, almost immediately. Late in 1975, the
FLY, coupled with the forecasting model, company would need planes with a much
becamea kind of early warning system. In greater payload than the Falconsif it was
March, the models predicted that with a going to handle anticipated increases in
fleet of only 33 Falcons sometime during volumeand earn an acceptable profit, but
the fall of 1975 FedEx would run outoflift, CABregulations still prohibited companies
thatis, its package volume would exceed like FedEx from using them. CAB conse-
its total aircraft capacity. Despite the fact quently turned down FedEx’sinitial re-
that FedEx wasbarely eking out a profit, it quest to buy larger DC-9’s.
had to do something. Theaircraft capacity problem wentto the
Mike Staunton (then head of scheduling, top of Smith’s agenda and becamethe sub-
nowvice-president for global operations, ject of almost daily conversations between
scheduling, and control) suggested that a him, his assistant Tex Weise, Staunton, and
bypass hubin Pittsburgh might reduce Brandon (then vice-president for operations
someof the pressure. In the transportation planning). Reporting to him wereairline
INTERFACES27:2 24
scheduling, headed by Staunton; telecom- runs of the model, making different as-
munications, headed by Jack Cockrill; and sumptions about the composition of the
OR, headed by Ponder. He asked Hinson fleet, usually by substituting combinations
and three other OR personnel to study the of two orthree different types of aircraft at
capacity issue. a time. No matter what combinations they
As has become the custom at FedEx, used, the result was the same: FedEx could
ORwasrepresented in the weekly senior not remain profitable unless it was allowed
management meetings. Duringthis era, to use larger planes.
Brandon attended the meetings, often ac- Smith was troubled and movedinto ac-
companied by Ponder or Hinson or some tion. He took an apartment in Georgetown,
of the other members of the OR team. This leaving COO Art Bass to run operations in
special role was accorded the OR group be- Memphis, and begana series of wide-rang-
cause Smith, who has a degree in econom- ing lobbying rounds in Washington, DC.
ics from Yale andis a liberal arts and mili- Virtually no one connected with airline or
tary science buff, had learned about how air cargo regulation escaped his attention.
OR worked, was impressed with the re- This effort took nearly a year. Armed with
sults, and had cometo trust the advice he the Autoroute results, Smith became a
received from the OR team. The payload strong advocate for an air cargo reform bill.
problem was so hot that Smith met with In March 1977, he testified before the sen-
the OR group often for extensive discus- ate aviation subcommittee and asked CAB
sions between the regular executive ses- for a certificate of public convenience and
sions. necessity. He appeared before the subcom-
The team hadoriginally built Autoroute mittee again in August,testifying in sup-
as a research tool to help people under- port of regulatory reform. This model-
stand the underlying structure of the air- based lobbying eventually got results. In
craft scheduling problem. Joe Hinson be- November, the House of Representatives
lieved that it would provide valuable passed the Air Cargo Reform Bill, and soon
background knowledge for solving some of thereafter President Carter signed it into
the company’s operations planning prob- law. FedEx almost immediately filed for
lems, and Brandon had encouraged him to permission to fly larger aircraft. It used Au-
explore the possibilities. Now, they used toroute to evaluate the use of Boeing 727s,
the model to determine the extent of the amongothers, on FedEx’s high volume
impendingcrisis. routes, and in December 1977, placed an
Autorouteis a heuristic optimizing, order for seven 727s.
scheduling model that takes loads, cities Human Resource Planning
served, and constraints, such as the latest The 727s brought about new problems
houran aircraft can depart a city and the with personnel planning. As FedEx ex-
earliest it can arrive at its destination, and panded, introduced more hubs, and
generates a complete single-hub-system launched a morediversefleet, it had more
flight schedule. Over several months, difficulty making sure its jobs and pay
Hinson and his OR team maderepeated scales werein order. Of special concern
March-April 1997 25
wereits pilots, who had to be qualified on each city station bought or leased its own
each new aircraft and who had manyjob equipmentand ran its own phone system.
opportunities elsewhere. The pilots consid- Moreover, station managersjealously
ered unionizing, and the company wanted guardedtheir control over the process.
to show them what they could reasonably “Our customers,” a typical comment went,
expect in career earnings and typesofair- “want to talk directly to their favorite agent
craft to fly. right here at home.” Unfortunately, there
The problem was formulated as a multi- waslittle call-answering discipline and
period, multi-stage stochastic (Markov there were no established standards. Vital
chain) transition model. The OR team used information was constantly beinglost. In
fleet and operating requirements for a de- short, it was chaotic. One day Tucker
cade or morein the future to lay out pilot Taylor walked into Charles Brandon’s of-
requirements by aircraft type for each pe- fice and asked ‘‘Whycan’t we have our
riod. The characteristics of the existing own Wizard of Avis?” The “Wizard of
crew complementserved as the starting Avis” was a computerized reservation sys-
point. The model then simulated the move- tem for car rentals that featured a central-
mentof each pool ofpilots in a given class ized telephone system for accepting and
through these periods. Statistical data was routing all reservations throughout Avis’s
used to determine the probability that a pi- system. The possibility of a wizard seemed
lot would fail a medical examination, fail to to Brandonlike a question that OR could
upgrade, accept a job with another com- answer, and he moved quickly to propose a
pany, and other events that could affect his project to Smith. At first he ran into a stone
transition from onestate to another. wall. There was a sharp division of opinion
The modelcalculated the payfor pilots, within the company on this matter, and
first officers, and others and was used to only after much heated discussion did
show them whattheir expected career path Smith approve the developmentof a proto-
at FedEx would looklike. Since the com- type dubbed Project Sydney (for Sydney
pany wasexpanding, the future was prom- Tucker Taylor). FedEx formed a project
ising for most pilots. This result proved team consisting of representatives from
very useful in labor-management negotia- EDP, telecommunications,field operations,
tions. FedEx also used the model to forecast sales and marketing, and OR. OR began the
gaps between the supply and demandof initial modeling, which would serve as the
personnel andto indicate what its future foundation for building the system.
hiring and training needs were. This hu- Somepreliminary model results sug-
man resource planning became an integral gested that a central telephone answering
part of the company’s overall planning system wasthe best solution. A prototype
process. for a centralized system was programmed
Perhaps We Need a “Wizard!” for the Burroughs B6700. In late 1977,
As FedEx’s volumes grew,it became in- FedEx conducted a test in Newark, at the
creasingly difficult to handle requests and time the worst-performingcity in the sys-
inquiries from shippers. Until late in 1977, tem on the basis of companystatistics and
INTERFACES27:2 26
employee performance, and in Memphis. reducing the time each call took, such as
The prototype system, though limited in improving the agents’ answering behavior,
capacity andreliability, was a vast im- routing calls moreefficiently across time
provementoverthe existing system. ‘For zones, and modifying other procedures.
the first time, the people in the stations FedEx used theseresults to develop an
could do their own work,” one executive overall business plan for both the telephone
observed,‘because they were relieved of system and the computer systems and to
answering the phones and processing the determine the optimum locations for the
orders.” Customers werealso servedbetter. call centers. Thecall center studyresults
Othercities immediately demandedthat served as the fundamentfor the subsequent
they too be added to the system. Despite development of Cosmos.
some knownflaws, FedEx rolled the proto- Ultimately, FedEx decided to build more
type out to 10 or 11 othercities. Ultimately call centers than the model indicated be-
FedEx handled all customers’ incoming cause its executives believed that the addi-
calls in centralized call centers, which then tional centers provided moreflexibility,
sent out requests to dispatch centers lo- avoided possible regulatory issues, and
cated in each city FedEx served. would serveas a positive laborrelations
Given the stunning success of Project move—allfactors not fully accounted for in
Sydney, Smith approved a project to build the original formulation, but which
a full-scale production real-time system— emerged during intense discussions with
eventually called Cosmos—to serve the en- Smith and others. Such give-and-take dis-
tire FedEx system. Atthe time, there were cussions, in which modelresults are pitted
many unresolved questions concerning against executive understanding, have
how manycall centers to establish, where become commonpractice at FedEx.
to place them, and how tostaff them. Hubba, Hubba
Hinson and Ponder, with the aid of OR an- Operations researchinfluenced one deci-
alyst Michael Sternad, built a customerser- sion that had an enormous impact on
vice system model to explore the possibili- FedExandset its course for over 15 years.
ties. The model clearly showed thatlength This hotly debated decision wasto build a
of call was the most important factor in de- SuperHub, and as Charles Brandon ob-
termining call-handling performance. served, “OR wasapplied effectively here
Queuinganalysis further revealed thatit and reversed the four-hub serial design deci-
was quite easy to overload the system and sion in favor of a single SuperHub. It saved
to bring it down.In fact, a sensitivity analy- the company!”
sis applied to the model, revealed that the When the network was small, its vol-
length of the call was the main cost driver umes low, and the Falcons the only aircraft,
for the entire system. Slight decreases in a single hub in Memphis wasclearly the
the averagelength ofcall resulted in sub- best choice (although the Pittsburgh experi-
stantial cost savings; slight increases re- ence had shownthe valueof parallel hubs).
sulted in a substantial cost increases. Fur- However, the dramatic growthof the busi-
ther studies suggested other methods for ness and the advent of the 727s—together
March-April 1997 27
with the possibility of even larger aircraft It was standard practice to discuss even
in the future—raised questions as to very preliminary ORresults in senior man-
whetherthis strategy wasstill optimal. agement committee meetings—Fred Smith
Indeed, several key executives were con- wasan especially eager participant. This di-
vinced that a series, multihub system— alogue served to change assumptionsin the
most favored four—was the way to go. model and to suggest other approaches.
Thefirst model runs indicated that a sin-
gle hub wasstill the preferred solution.
Four hubs appeared to be a Every timethis solution was presented,
donedeal. however, a heated debate ensued. Smith
and others continually came up with new
They reasoned that FedEx could use small ideas aboutthecost of aircraft, the nature
aircraft to carry packages to each regional of the facilities used, and the like. With
hubandthatit could use larger planes to these new considerations in hand, the OR
haul packages from hubto hub. A great team would head back to its quarters to
deal of momentum and enthusiasm built modify the model and to run it a few more
up within the company as people maneu- times. This continual dialogue, pitting exec-
vered for a four-hub network and, as one utive assumptions against their implica-
executive putit, ‘Four hubs appeared to be tions as derived by a model, lasted for
a done deal.” about a year. Throughoutthis time, the sin-
Hinson, Ponder, and Brandon were not gle hub solution generally dominated in
convinced. OR had notyet studied the the model results but not, it turned out, in
problem in any depth. Andso, as the front many executives’ minds.
office moved toward a four-hub system, In the end, Smith decided on the single-
they began to model the situation. Their hub system—now dubbed the SuperHub.
first task was to expand their modeling ap- The model revealed the underlying eco-
proach. Autoroute,it turned out, could be nomics clearly. With high costaircraft and
used to estimate the performanceof feeders facilities used intensively for compressed
into a regional hub, butit was incapable of periodsof time underconditions that de-
determining whether or not a given hub mandednear perfect performance—thatis,
should remain in the final system network. failure to service a plane or to pick up a
Since the final solution must contain a dis- batch of packages could not be tolerated—
crete number of hubs, they needed an inte- the single hub worked best. Given the tech-
ger linear programming model (ILP). Con- nology and volumesat the time, the
sequently, they constructed a two-phase SuperHub design was moreflexible;it re-
model consisting of Autoroute to perform quired only onebig sort, reducing total
within hub analyses and an ILP to choose sorting time, and it made crew domiciling
from among the candidate hubs. They easier. The four-hub approach had three
made exploratory runs assuming between majorflaws: it caused low utilization of
one and eight hubs and also addressedfleet high cost assets and, hence, had a poor ROI
size and composition issues. (in some cases negative), it had a low toler-
INTERFACES27:2 28
ancefor faults because it did not have the ture, which was predicated on ORresults,
built-in redundancy that the single hub did, called for a much different approach to
and it consumedalmostall of the 12 hours FedEx’s organizational design. Smith un-
FedEx had available to process work to derstood this, and in April 1978, he reor-
meet its commitments. Hence, it did not al- ganized the companyfrom its previous de-
low muchslack nor did it provide much centralized form into a highly centralized
room for growth. Memphis remained the company, providing as an explanation that
single hub in the system, and in July 1979, “various approaches were modeled against
FedEx broke ground for the massive, our present and future tasks.”
highly automated SuperHub. Tracking toward the COSMOS
Soon after the SuperHub was opened, As volume growth accelerated, the prob-
the model revealed additional advantages lem of making the stations work better and
to the concept. As a competitive move, servicing customers quicker and in a more
Smith wanted to change the committed de- friendly manner surfaced. The excellent re-
livery time from 12:00 noon to 10:30 am. sults obtained from the Project Sydney pro-
The model showed that the FedEx system totype and the successthat the airlines had
could deliver on that promise. Further- had with computerized reservations sys-
more, a competitive analysis application of tems suggested an approach. With this in
the model revealed that UPS, FedEx’s ma- mind, Brandon, by nowseniorvice-presi-
jor competitor, could noteffectively meet dent for planning and information systems
this deadline. In late 1982, Tom Oliver, (including electronic data processing),
head of FedEx’s marketing, announced, sought out and hired Howard Bedford, an
“Wewill offer 10:30 delivery, more service expertin airline reservation systems, to de-
options, Saturday pickups, package tracing, velop a comprehensive system. Bedford
andcail-backs to shippers informing them hired James Tollefson, who had experience
that the packages have been delivered,’”’ at IBM and Avis; Henry Howell; and a
and he did this knowing that his company crew that soon reached about 150 people,
could deliver on this promise and that, for most of whom werealso pioneers in high-
the foreseeable future, UPS could not. AI speed, on-line transaction processing sys-
of this was consistent with Smith’s strategic tems. The group was ensconced in the Ad-
emphasis on quality andreliability of vanced Systems Development Center
performance. (ASD) located in Colorado Springs. There-
The technological infrastructure of the sult was COSMOS.
company was now pretty muchin place: a COSMOS(Customer, Operations, Ser-
mixed fleet of small planes providing ser- vice, Management Operating System)is a
vice to the smallerstations andlargerair- mainframe-basedorder and dispatching
craft on the higher volumeruns funneling system that FedEx has constantly upgraded
through a SuperHubplus centralized call to cope with increasing volumes, an ex-
centers for answering incoming phone panded route structure, and an everin-
calls, which were in turn dispatchedto lo- creasing emphasis on tracking packages
cal stations for response. This infrastruc- from origin to final destination. Version 1
March-April 1997 29
INTERFACES27:2 30
ations research. The existing package-han- his team to begin a new set of analyses.
dling system, youngas it was, was already This give-and-take dialogue lasted about
comingto the end of its useful life. Nightly six months during late 1979. Slowly, a few
package counts exceeded 50,000 and were crucial yet robust results emerged. One was
growing rapidly. How long would the that the company could save on fuel expen-
SuperHub concept remain optimal? What ditures if it employed DC-10s on high-
kinds of air cargo carrying capacity would volume routes to such cities as Boston, Chi-
FedEx need in the future? How fast must cago, Los Angeles, and Newark.Atfirst,
planesfly to meet their tight deadlines? Smith balked at this solution. For once the
Joe Hinson and his group set out to an- company wasfinancially sound enough to
swerthese questions. Using a multi-period make the investment, but he thought the
model of the entire FedEx system and look- price was too high. One DC-10 cost nearly
ing five years out into the future, theyfirst
estimated the system’s carrying capacity,
the maximum amountoffreight that the
Everythingis tied to
system would haveto carry each night. everything else.
One million poundsof total capacity ap-
peared adequate and even provideda little as muchashis entire fleet of Falcons had
breathing room. (This seemingly conserva- cost. The OR team changedits approach.
tive maximum load factor was exceeded Using an updated version of Autoroute, the
earlier than anticipated.) They developed team calculated the price point at which
several scenarios by making differentas- FedEx should be indifferent between pur-
sumptions about package counts, their dis- chasing DC-10s and another bundle of air-
tribution, transportation modes (types of craft. The results rekindled Smith’s interest,
aircraft and trucks), and alternative hub- and, armed with this new information, he
and-spokestructures. They calculated costs madean offer for four DC-10-10CFs. The
and performancefactors for every activity: seller initially demurred. But, because he
the original shipper request, package understood the economics of the deal very
pickup, local transport, outboundstation well, Smith held tight throughout the nego-
sort, outboundline haul(air or ground), tiations and finally got a purchase price the
hubsort, inbound line haul, inboundsta- model told him he could live with. In
tion sort, local transport, and final delivery. March 1980, FedEx christenedits first
Once every few weeks, Hinson would trek DC-10.
across to headquarters and presenthis Anotherresult of the butterfly explora-
team’s preliminary results to Smith and the tions was even more controversial, because
executive committee. Some of the execu- it reopened old woundsfrom a previous
tives—almost always Smith—would think internal companybattle. Various model
of some new possibilities. ‘Whatif this runs predicted that sometime during 1987
were the case?” or “Couldn't we do that?” the SuperHub concept would finally reach
Hinson dutifully took notes, and as each its practical limit. FedEx would need more
session cameto a close, he would return to hubs. As long as the 727s could handle the
March-April 1997 31
volume, the single hub design remained than the models had indicated would be
superior. The DC-10s extendedits life optimal. However, Smith was convinced.
somewhat but eventually, as volumes ex- He wanted to move ahead quickly and to
panded, the SuperHub would collapse of test the concept for real. It has been suc-
its own weight. cessful.
Multi-hubs were in FedEx’s future, but Locating the Indianapolis overlay hub
when, where, and how should they be op- wassimilar. The location model favored
erated? The models showed that the most Rockford, Illinois. This assumed that FedEx
efficient approach was somewhatdifferent would haveto build its ownfacilities for
from that proposed duringthe first corpo- each of its overlay hubs as it had done in
rate debate. The original serial approach Oakland. In August 1987, however,a fully
called for collecting all of a region’s pack- furnished airport facility became available
ages in a regional hub, fully sorting them in Indianapolis. A few quick runs of the
there, and then sending packages to desti- models showedthat the Indianapolis site
nations within the region while routingall waseconomically viable. Smith bought the
other packages to the appropriate other Indianapolis facility and began operations
hubs. This placed a heavy sorting burden in October 1988.
on the system. A moreefficient approach,it The transportation models uncovered an-
turned out, was for each pickupstation to other advantageof the overlay hub con-
make a simple binary decision. Sending cept. The economics of hauling show that
packagesfor redistribution within the re- for short distances trucks are the mosteffi-
gion to the regional hub and sendingall cient carrier. As the distance increases a
others to the Memphis SuperHub. This ap- switch-over point is reached at which air
proach—called the overlay hub method— cargo becomesless costly. The original dis-
distributed the sorting burden and permit- tribution and configuration of stationsfa-
ted it to take place in parallel rather than vored air cargo. Nevertheless, Smith and
serially. his associates had been looking forward to
The team then used economic location using trucks in high volume areas. Auto-
models to determine the best sites for the route showed that FedEx had reached that
overlay hubs. Early results showed that the time; the density of traffic around the new
West Coast was a good candidate for an overlay hubs made trucks more economical
overlay hub, the center of gravity resting for certain short hauls. FedEx initiated.
just north of Los Angeles. Autoroute runs, truck service to replace other modes on
however, modified this conclusion. Because these routes. Today FedEx is one of the na-
of theflight times from Seattle, Portland, tion’s largest integrated, multi-conveyance
and other Pacific Northwestcities and a freight carriers.
few other operating considerations, Auto- Some Current Projects
route indicated that the San Francisco Bay After extensive studies by the OR team
area wasa better spot. In June 1986, FedEx in July 1991, FedEx ordered 25 A300-600
openedthefirst overlay hub in Oakland, Airbusfreighters. The first deliveries were
California. This wasat least a year earlier in April 1994, An updated version of Auto-
INTERFACES 27:2 32
March-April 1997 33
INTERFACES27:2 34
March 1973 -Original idea -Founder -Initial 11-city model -11-city schedule
(failure)
April 1973 -Soul searching and -Emplanement -112-city origin- -26-city system
rethinking statistics destination model (satisfactory)
-Employment SIC
-Business
population
-Experience
June 1973 -Cost pressures -Airport data -FLY -Refine schedules for
-Draw onthe -Aircraft -Simulation model moreefficient
results of the characteristics produces expected operations
O/D model -Loading- flight times and
unloading engine cycles
-Ground
operations
-Weather
experience
Early Fall 1973 -Desire to grow -Additional data -FLY and O/D models Successful
-Need for additional on candidate simulate 82-city presentation to
capital cities system with 33 potential investors
-Investor confusion Falcons -Add Los Angeles to
system early
Mid Fall 1973 -Need to add -Accounting data -Financial planning -Refine operations to
financial -Financials model (Fin) improvefinancial
managementto -Cost data performance
operations
management
Late Fall 1973 -OPEC rationing -Fuel consumption -FLY, O/D, Fin models -Obtain fuel allowance
imposed and cost predict five-year fuel during OPEC
estimates requirements for an restrictions
82-city system
Summer 1974 -Need for engine -Flight time data -Joe/Engine model -Keptfleet in service
overhauls -Numberofflights until bluckets
threatens to -Engine cycles problem wassolved
cnpple fleet -Overhaul
requirements
Summer 1975 -Company growing -Expected growth -Forecasting model, FLY, -Capacity alert bypass
faster than ability -Seasonalfactors and other models to hubestablished at
to manage well -Actual load data predict operating Pittsburgh
-ATA gross volume requirements
Winter 1975 -Increased -Load factors -AUTOROUTE -Air Cargo Reform Bill,
complexity of the -Cities served -FedEx orders seven
system -Arrival and 727s
departure times
December -Customercalls -Customercall data -Initial telephone -Need for prototype
1975 overload system -Telephone volume answering model centralized system
-Telephone established
facilities -Project Sydney
-Operating data
March-April 1997 35
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 O. 1995, Waves of Change:
Business Evolution Through Information Technol-
ogy, Harvard Business School Press, Boston,
Massachusetts.
INTERFACES 27:2 36