DataPDF Available

Manifest Destiny: Adaptive Cargo Routing at Southwest Airlines



Southwest Airlines is known for innovations in efficiency. In order to make its cargo operation even speedier than it already was, the company adopted an idea from complexity science. Using a computer-based simulation with agents who represent decision makers, Southwest has tested the effects of simple changes in manifesting and loading strategies.
in Action
About the Authors:
Fred Seibel is Vice President of software
development at Bios Group. Previously,
he managed the software development
teams supporting the Advanced X-ray
Astrophysical Observatory for the
Smithsonian. He received his BS in Physics
from Yale University and his Ph.D. in
Nuclear Physics from Duke University.
Chuck Thomas is Director, Financial
Analysis, for Southwest Airlines Company.
Previously, he was head of Southwest’s
planning and budgeting functions. He holds
a Ph.D. from the University of Texas where
he studied accounting and finance.
Southwest Airlines, as its passengers know,
doesn’t assign seating—and it never has.
Admittedly, the idea was laughable when the
airline first got off the ground in 1971, because
Southwest was transporting only three or four people
per flight. But even on its full planes today, there are
no assigned seats. The company’s reasoning? Speed
and efficiency. Without seat assignments, Southwest
can turn planes more quickly at the gates.
Dallas-based Southwest Airlines is known for many
things: low fares, short-haul and high-frequency
flights, legendary customer service, and—perhaps
most of all—continuous innovations in efficiency. And
the fifth-largest airline in the US applies this mantra of
speed and efficiency not just to the transport of
passengers, but to its cargo business as well.
Speedy as its cargo operations were, however,
Southwest was encountering problems of congestion
in some areas and excess capacity in others. With the
assistance of Bios Group, located in Santa Fe, New
Mexico, the airline analyzed its cargo handling system
through a complex adaptive systems lens. This
complexity science approach helped Southwest to
make changes in its manifesting and loading
strategies—a surprisingly intuitive solution that
yielded dramatic results.
Re: Bin Space and Bottlenecks
Southwest had been scrutinizing its cargo business
over the past six years, and the company’s nagging
suspicion was that this area of the business could be
ManifesT Destiny:
Adaptive Cargo
Routing at
Southwest Airlines
Fred Seibel and Chuck Thomas
Unlocking the Secret of Increasing Returns, pg. 39
in Action
tion to the rest of their workload. So they would then
load the cargo onto another aircraft and send it down
line to someone else. This “hot potato” mind-set was
particularly problematic at the airports that saw the
heaviest cargo traffic, such as Phoenix, Houston, Las
Vegas, and St. Louis.
Chasing Prairie Dogs
Southwest was determined to amend its cargo woes in
three ways: by improving the work life of its ramp
agents, tapping into unused bin space capacity, and
growing the business and increasing its profit base.
In a style typical of the culture at Southwest Airlines, a
brainstorming team was tasked with finding
remedies to the cargo-routing problem. The team
consisted of subject matter experts, ramp agents, and
ramp supervisors who had a good feel for best
practices at the different stations. In addition, the
group established a temporary command center to
monitor day-to-day problems with the movement of
cargo and to experiment with a variety of approaches
to specific problems.
Southwest’s approach was likened to, as one team
member described it, “chasing prairie dogs.” The
modus operandi of the team was to study the local
problems the various stations experienced. From those
local problems, the team would then extrapolate and
devise a global solution, or a series of solutions, that
would solve Southwest’s cargo problems across the
board. But as chasers of prairie dogs know, every time
one prairie dog hole is closed off, another pops open.
expanded. One of the key elements under examina-
tion was the airline’s cargo capacity. Southwest’s
operational data suggested that, across the fifty-six
airports it services, an average of 7 percent (by both
weight and volume) of its bin space was full on any
particular flight—a small percentage, considering that
competitors were filling as much as 35 percent of the
bellies of their aircraft.
A thorough field inspection revealed that bottlenecks
existed throughout Southwest’s cargo-routing and
handling system. Many times, aircraft were scheduled
to carry a large load of freight, but lacked the bin
space to accommodate this volume. In more than a
few cases, the company was taking aircraft departure
delays to accommodate its cargo business—which was
unheard of at Southwest, being fundamentally a
passenger airline.
Furthermore, Southwest’s ramp agents, the people
who actually move the payload, were experiencing
frustration. Focused as they were on speed, their
approach to cargo was “I’ll just throw it on the next
plane and get it out of here.” Because space on the
cargo ramps is tight, and because the packages
themselves specified only the final destination and
not the intermediate stops, there was ample incentive
and opportunity for the ramp agent to offload the
cargo to someone else.
One station, therefore, would ship cargo down line to
another station. The ramp agents who received that
freight couldn’t understand why they had it, in addi-
Francis Fukuyama on Self-Organization . . . , pg. 74
in Action
operations through simulation. They collected histori-
cal data, which would calibrate the model and
illustrate how Southwest was routing its cargo, as well
as the problems that resulted. In this way, the team
could use either the rules that were generated in the
simulation or the rules they might discover, which
would improve the model—and then apply those rules
to a real-world situation.
Bios Group relied on data gleaned from shipment
descriptions, flight schedules, and freight logs in
creating this model. A record of the payload
Southwest had shipped in 1998 was provided by
weight, by number of pieces, by origination and
destination for each shipment, and by class of service.
(Southwest provides three levels of cargo service:
“Next Flight Guaranteed,”(NFG) in which shipments
are placed on the first available nonstop or direct
flight out; “Priority Rush,” which provides 24- to 48-
hour service, depending on the distance covered; and
“Freight,” in which shipments are transported on a
space-available basis and are not flight-specific). The
team also had access to the flight manifests, which
are instructions on how a particular piece of freight
should reach its final destination, where it should be
transferred from one flight to another, and so forth.
In addition, Bios Group was privy to Southwest’s flight
schedules: the airplanes’ takeoff and landing times,
their flight numbers, and their segments. Furthermore,
they incorporated background data into the model.
This background data, or the freight logs, showed how
much of the bin space was actually available and how
Translation: no one solution seemed to cure all the
problems at all the locations.
While Southwest excelled at knowing what worked at
individual stations, what it needed was the big-picture
view: network optimization. The airline considered
bringing in network design specialists from other
fields that experience similar network problems, such
as the transportation or telecommunications
industries. But, in a move that ran counter to its
low-tech, high-touch culture, Southwest called on the
complex adaptive systems expertise of the scientists
at Bios Group.
Following the Rules
The initial solution Southwest proposed, and where it
sought assistance from Bios Group, lay in the
production of a simple “rulebook.” This rulebook
would sit on each freight agent’s table and serve as a
reference for directing freight away from heavily used
stations to lightly loaded stations. With such a rule-
book, Southwest hoped, freight agents could avoid the
problems that Phoenix and Los Angeles International
Airport (LAX), for example, were experiencing.
In generating this rulebook, the team relied on the
rules of complexity theory: First, understand the
behavior of the smallest elements, cells, or agents of
an organism. Second, discover the properties of those
cells that produce large-scale “emergent” behaviors in
the organism.
Following these rules, the group created an agent-
based model that reproduced Southwest’s current
Southwest Airlines is known for innovations in efficiency. In order to make its cargo operation even
speedier than it already was, the company adopted an idea from complexity science. Using a
computer-based simulation with agents who represent decision makers, Southwest has tested the
effects of simple changes in manifesting and loading strategies.
article abstract
A Case for Eco-Industrial Development, pg. 34
in Action
much was pre-loaded with baggage or mail. (Because
baggage and mail take priority over cargo, an assump-
tion was built into the model that some fraction of the
space needed would already be occupied).
The objects in the system—namely, the shipments and
the flights—were incorporated into the model. Freight
forwarders and ramp agents were also modeled into
the simulation, as were the actions of these agents:
receiving a shipment, assigning a shipment to a flight,
loading the plane, flying the segment, unloading the
plane, transferring freight to the next plane, and
moving cargo to the freight house.
Testing the Rules
In order to reveal manifesting strategies, the team
used Southwest’s actual manifests to calibrate the
data. They then ran the network for a week, moving
packages from one place to another. In addition to
this hard data, Bios Group built anecdotal information
about the ramp operators into the simulation to
produce “probabilistics.” (For example, one probabilis-
tic was that an overworked ramp agent in Phoenix
would simply load a piece of cargo onto the next
plane without paying attention to the manifest).
To test the rules embedded in the simulation, the team
generated the manifests and the ramp operations in
the computer as well. The model was then run in three
modes: the way Southwest thought it was handling its
operations, the way Bios Group believed Southwest
was working, and the way Bios Group thought the
airline should run its cargo-loading operations.
What was measured?
TThe amount of cargo handled at each station. A
certain amount of cargo handling is unavoidable;
cargo must be loaded and unloaded, but what can
be avoided are the transfers in between.
TThe volume of cargo that had to be stored over-
night. For security reasons, freight held overnight
must be kept under lock and key and then trucked
back out to the aircraft in the morning.
TWhether or not the level of service for each pack-
age was met. That is, did the NFG packages, for
instance, arrive as promised?
From the get-go, Bios Group experienced a few diffi-
culties with the simulation when it flew the schedules.
For starters, the manifests regularly called for loading
10,000 pounds of cargo on a flight from LAX to
Phoenix—but the planes can only carry 2,000 pounds.
It was clear that multiple people were manifesting
cargo for a single plane, or set of planes, that couldn’t
handle the volume.
In the simulation, therefore, a computer-generated
manifest redirected pieces of cargo that didn’t fit on
the originally scheduled flight. Bios Group also
allowed some of the ramp agents in the model to
invoke the “hot potato” strategy of loading the freight
onto the next plane, despite the manifest instructions.
Finally, Bios Group compared its simulation results to
the observed measures.
Figure 1
Flight Origin Destination Takeoff Landing Route
Rush Shipment: Albuquerque to Oakland—The Old Way
Technology Watch, pg. 78
7:00 7:55 102
8:20 10:05 102
10:30 11:55 102
12:15 13:15 102
14:15 15:35 91
13:35 14:55 102
15:15 16:30 102
16:50 18:10 102
18:30 19:30 102
19:50 21:05 102
21:30 23:25 102
in Action
Model Results
Because there are no same-plane routes that connect
certain cities, some transfers are inevitable. But by
adapting the same-plane strategy, where possible,
Bios Group saw the opportunity for Southwest to
reduce both cargo transfers and overnight storage.
(See Figure 2.) For example, if Southwest were to
adopt this strategy, Phoenix, which experiences the
heaviest cargo traffic, could reduce the weight trans-
ferred at its station from 160,000 pounds per week to
roughly 50,000 pounds per week.
System-wide, Southwest stored approximately 240,000
pounds of cargo overnight. Bios Group’s optimal
forecast for overnight transfer weight was just over
50,000 pounds per week. And the total weight
handled dropped from about 3,250,000 pounds to
2,500,000 pounds.
While the model results were certainly promising,
Bios Group had two principal concerns: How would
this same-plane strategy fare in Southwest’s culture?
And what impact would this strategy have on the
airline’s service levels? This same-plane strategy did
not jibe with the culture of speed at Southwest, espe-
cially for the ramp agents who previously had no need
or incentive to consult the schedule. What’s more, this
same-plane approach was also counterintuitive to
meeting service levels. By holding the cargo for a later
plane, for instance, the ramp agents might worry that
it would not arrive within the time frame promised.
Striking “Gold” in the Schedule
Rather than relying on the manifest alone, the team
also devised an algorithm to route the cargo. When
running this algorithm, the amount of cargo being
transferred, as well as the amount of cargo being
stored overnight, dropped dramatically. It was, as one
Bios Group scientist remarked, akin to “striking gold in
the schedule.”
What was the “gold” that the schedule held? The
Consider a package that is brought to the airport in
Albuquerque at 9:00 a.m. for Priority Rush shipment
to Oakland—meaning that it must arrive in Oakland
within 24 hours. In 1998, the employee in the freight
house would have looked at the schedule and seen
that there was a flight that went from Albuquerque to
Las Vegas, and from Las Vegas to San Francisco. And
then there was a flight from San Francisco to Oakland.
See (Figure 1).
But the freight agent was ignoring the route that
specific plane flew. He was overlooking the fact that
the original flight—the one from Albuquerque to Las
Vegas, and from Las Vegas to San Francisco—eventu-
ally flew to Oakland as well. Therefore, by leaving the
cargo on the plane and letting it ride down to San
Jose, back to San Francisco, and then on to Oakland,
the need to transfer that cargo from one plane to
another would be eliminated.
Figure 2
Weight Transferred at Station
in Action
exactly how to route the cargo.
How is the same-plane strategy working in the real
world? Southwest has seen a decline in its freight
transfer rate of 50 percent to 85 percent across its six
busiest cargo stations. That translates to a decrease of
roughly 15 percent to 20 percent in the workload for
the ramp agents moving cargo. In line with this
decrease in workload, the total weight handled has
also dropped, easing both the burden and the frustra-
tion of the ramp agents.
A dramatic reduction in overnight transfer has
allowed Southwest to cut back on its cargo storage
facilities, freeing up dollars that previously went
toward renting expensive airport space. In addition to
requiring smaller overnight facilities, this reduction in
overnight transfers calls for less manpower to bring
cargo to the freight house at the end of the day, issue
new manifests, and then transport the freight out to
the airplanes again in the morning. Minimizing those
wage costs is also an advantage to Southwest.
What’s more, Southwest had been examining its
material handling and overnight freight situation to
see where improvements could be made. But because
overnighting has been cut so dramatically, that focus
group has been suspended.
Operations have become more efficient, without
damaging Southwest’s customer-service levels. As Bios
Group’s simulation suggested, shipments at all
airports are arriving earlier. This improved efficiency
is due, in part, to the elimination of offloading (when
But when Bios Group checked its model, it found that
NFG service actually improved. Why? Although they
didn’t analyze this result in detail, the feeling was that
NFG improved because the airplanes weren’t being
loaded, in the words of one Bios Group scientist,
“chock-a-block” full. By waiting for the planes that
were going to the right places, the model actually
freed up more space for NFG packages and got them
to their destinations by the appointed time. While
Priority Rush service levels did decline by approxi-
mately 1 percent, the criterion for freight service was
set at three days, which was acceptable.
(See Figure 3.)
Real-World Results
To implement this same-plane strategy, Southwest sent
teams out to the Phoenix, Houston, Las Vegas, St.
Louis, Midway, and Kansas City airports. These teams
spent three weeks saturating each of those stations
with the ins and outs of the same-plane strategy—from
helping the ramp and freight agents change the routing
they would normally select, to fighting the intuitive
response of “let’s just get this cargo off the ramp.”
Southwest is now rolling out this strategy to twenty
more airports, without the luxury of automation
support. Because Southwest has Y2K issues (much like
every other company at this stage in the millennium),
the company’s systems personnel have not been free
to automate these changes. Instead, Southwest’s
employees are working from the rulebook the
company initially proposed: a simple set of rules,
spelled out in a three-page document that tells them
Figure 3
Percentage of Lots Delivered Within
Service Contract Requirements
Toward the Adaptive Enterprise, pg. 49 A Case for Eco-Industrial Development, pg. 34
Routing Method
in Action
ramp agents simply shipped cargo down the line,
regardless of its destination) and the resultant
backtracking this offloading practice induced.
In some cases, particularly NFG and Priority Rush
shipments, cargo is arriving one to four hours earlier
than it was with Southwest’s previous mode of opera-
tion. With shipments arriving earlier, customers are
seeing greater differences in the airline’s service
levels, which has a key impact on revenue enhance-
ment for Southwest. And by freeing up additional bin
space (by waiting to load the cargo onto the appropri-
ate planes), Southwest is able to offer more customers
NFG or Priority Rush service, thereby selling
customers higher service levels.
Needless to say, the key sponsors at Southwest are
very pleased with the results.
An Intuitive Solution, in Hindsight
Given the networked nature of the transport system
itself, coupled with the unpredictable nature of deliv-
ery capability, Southwest needed a big-picture view to
solve its cargo-routing problem. Bios Group’s agent-
based model discovered rules of behavior, specific to
the airline’s manifesting and loading strategies, that
could vastly improve Southwest’s cargo handling
operation—and recommended a solution that seems
remarkably intuitive in hindsight.
And intuitive it was. In their brainstorming sessions,
Southwest personnel had actually suggested the same-
plane approach as a possible remedy to their cargo
woes. Without a high-altitude view, however, it was
unclear whether proposed solutions would correct the
problem for Southwest’s entire cargo-routing network,
or if they would only serve to seal off, if you will, one
or two prairie dog holes.
In the end, Bios Group’s discovery of the same-plane
strategy, combined with a simple three-page rulebook,
has provided Southwest Airlines with a network-
optimization solution that alleviates the frustration of
its ramp and freight agents. This gives Southwest
clearance to focus on steps two and three: tapping
into unused bin capacity and growing the business.
A Case for Eco-Industrial Development, pg. 34
... The techniques have been applied to real and theoretical situations within management. Southwest Airlines have used an agent-based model to improve the operational efficiency of airfreight routing (Seibel & Thomas, 2000); excess inventories of Proctor & Gamble have been investigated by introducing an agent-based supply chain model (Siebel & Kellam, 2003) while researchers from France Télécom are using agent-based models to simulate the interaction of the behavior of their customers (Ben Said, et al., 2001), while Axelrod, et al. (1995) have used agent-based models to investigate the formation of alliances. On the theoretical side, agent-based models have been used to study adaptive strategies on rugged fitness landscapes (Rivkin, 2001), Rivkin's agent-based model is predicated on Kauffman's (1993, 1995) NK model to generate 'tunable' fitness landscapes. ...
Full-text available
This chapter introduces agent-based models and their application to manage- ment. The use of agent-based models is especially beneficial in situations where there are a number of inter-connected agents forming a complex system, be they firms, employees, customers, or other entities. We discuss the concept of agent- based modeling, building on an example model of interacting firms and custom- ers. We discuss how this model was developed using the agent-based modeling toolkit RePast. We describe not only the advantages but also the disadvantages of agent-based models over more traditional techniques. We introduce several other agent-based models from the more general social science literature, discuss the level of complexity that should be introduced into an agent-based model, and discuss how such a technique may make an impact on the manager or organiza- tion whose strategy is to endeavor to manage the complex.
ResearchGate has not been able to resolve any references for this publication.