You are on page 1of 13

informs ®

Vol. 34, No. 1, January–February 2004, pp. 26–38 doi 10.1287/inte.1030.0056


issn 0092-2102  eissn 1526-551X  04  3401  0026 © 2004 INFORMS

Menlo Worldwide Forwarding Optimizes Its


Network Routing
Robert C. Prior, Rob L. Slavens, Jerry Trimarco
Menlo Worldwide Forwarding, One Emery Plaza, Vandalia, Ohio 45377
{prior.robert@menloworldwide.com, slavens.rob@menloworldwide.com, trimarco.jerry@menloworldwide.com}

Vedat Akgun, Edward G. Feitzinger, Chyi-Fu Hong


Menlo Worldwide Technologies, 2055 NW Savier Street, Adtech 11-2, Portland, Oregon 97209
{akgun.vedat@menloworldwide.com, feitzinger.edward@menloworldwide.com, hong.chyifu@menloworldwide.com}

Since 2000, Menlo Worldwide Forwarding (formerly Emery Worldwide) has faced a particularly challenging
business environment that has been exacerbated by a weakened economy, the events of September 11, and a
decreasing demand for airfreight. To meet these challenges, the company and Menlo Worldwide Technologies
developed a network-routing-optimization model to optimize Menlo Worldwide Forwarding’s North American
transportation network. The project team and senior managers have repeatedly identified and applied low
cost solutions to meet the changing and complex network-routing requirements. By maximizing its use of
network capacity, the company has increased profitability and reduced operating costs while maintaining high
service levels. In 2002 alone, Menlo Worldwide Forwarding reduced operating costs by 21 percent, increased
operating margin by 41 percent, and improved financial results by $80 million in the North American aircraft
transportation operation. Moreover, management used the optimization model to facilitate Menlo’s transition
from a heavily asset-based, integrated airfreight company to an asset-light, freight-forwarding business. This
created a flexible operating environment and a competitive advantage for future operations.
Key words: transportation: models, network; industries: transportation, shipping.

M enlo Worldwide is a $2.7 billion company that


employs 15,000 people and provides global
supply-chain services in more than 200 countries.
centers around the world. To meet the varied service
requirements of its customers, Menlo Worldwide For-
warding has built and operates a complex network of
Four major components operate under Menlo World- aircraft and trucks within North America (Figure 1).
wide’s brand name: Menlo Worldwide Forwarding, The network connects 150 logistics centers and eight
Menlo Worldwide Logistics, Vector SCM, and Menlo regional hubs and provides next-day, second-day, and
Worldwide Technologies. Menlo Worldwide Forward- economy service within the continental United States
ing (formerly Emery Worldwide) is a global trans- and selected Mexican and Canadian locations. The
portation and logistics provider. Menlo Worldwide network also connects Menlo’s international freight to
Logistics is a leading third-party contract logistics the North American markets through 12 international
firm. Vector SCM is a joint venture with General gateways.
Motors (GM) that is GM’s lead logistics service The core of Menlo’s North American network is
provider worldwide. Menlo Worldwide Technologies the nightly transportation and sortation of freight that
was formed by combining the information technology takes place at the Dayton hub facility. These opera-
departments of the three operating companies named tions, referred to as “primetime” within Menlo, utilize
above. an integrated network of aircraft and trucks to pro-
Menlo Worldwide Forwarding has more than 55 vide next-day service. This network operates in a
years of experience in transportation, logistics, charter, hub-and-spoke fashion with the inbound and out-
ground, and airfreight operations. Known for its abil- bound aircraft and truck routes forming the spokes
ity to ship heavyweight airfreight (which is defined as and Menlo’s on-airport sortation facility in Dayton,
freight that typically exceeds 75 pounds per shipment Ohio, the central hub. The Dayton hub is a one million
and requires the use of a lift device), Menlo Worldwide square foot complex with a sorting capacity of over
Forwarding ships commercial, industrial, and govern- 1.2 million pounds of freight per hour. It has a 50-slot
ment freight worldwide. The company operates in truck dock and enough ramp space (and ground
229 countries and maintains more than 600 logistics handling equipment) to accommodate 70 aircraft that
26
Prior et al.: Menlo Worldwide Forwarding
Interfaces 34(1), pp. 26–38, © 2004 INFORMS 27

HCI
HPO

HDY

HBN
HCL
HLX
HTX

HMC

Aircraft
Feeder Trucks
Regional Hub
Dayton Hub

Figure 1: This example of Menlo Worldwide Forwarding’s primetime network shows aircraft and truck routes.

can range in size from a Boeing 727-100 to a Boeing truck or aircraft for delivery to the logistics center
747-200. These features enable the simultaneous load- nearest its final destination.
ing and unloading of aircraft and trucks, the break- During the early and mid-1990s, when business
down and buildup of the aircraft and truck shipping within the US was booming, Menlo Worldwide For-
containers and the sorting and processing of the warding’s US and international freight volumes grew
nightly primetime freight, all within a four-hour sort steadily. To serve its customers’ shipping needs and
window. Using its North American network, Menlo take advantage of business opportunities in such
processes several million pounds of freight daily, gen- areas as charter sales and US Postal Service Priority
erating approximately $1.5 billion in revenue each Mail, Menlo Worldwide Forwarding expanded its
transportation-network and hub-sorting capabilities.
year.
It added aircraft to the fleet and expanded and mod-
The primetime network operates as follows: Freight
ernized the Dayton hub. It also created a network
is delivered to one of the North American logistics
of seven regional hubs to facilitate its processing
centers by a customer or by a scheduled Menlo
of regional, second-day, and economy freight. These
pickup. Here the freight is processed for delivery and changes, however, greatly increased the complexity of
loaded into a shipping container. From these logistics the North American transportation network.
centers, the freight is transported either to the Dayton Management within Menlo Worldwide Forward-
hub via an inbound aircraft or truck or to another ing recognized that it needed sophisticated decision-
designated location (such as the nearest air location, support tools to help it run the complicated North
one of the company’s regional hubs, or an interna- American network more cost-effectively. Although
tional gateway) by a connecting feeder truck. When it had invested millions to develop software that
the freight reaches its intermediate destination, it is enhanced its shipment tracking, cost accounting, crew
unloaded, sorted, and reloaded onto an outbound scheduling, and aircraft maintenance, these programs
Prior et al.: Menlo Worldwide Forwarding
28 Interfaces 34(1), pp. 26–38, © 2004 INFORMS

were largely stand-alone systems that did not provide Menlo Worldwide Technologies
users with a means to conduct what-if routing analy- Although Menlo Worldwide Forwarding had a clear
ses or support network planning. understanding of its air and ground transporta-
To address the shortfall, Menlo Worldwide For- tion networks, its contracted airline operations, and
warding began evaluating existing modeling tools its customers’ needs, it lacked sufficient knowledge
and off-the-shelf software packages. The company of the optimization technology to go it alone.
wanted to reduce network operating costs and replace Menlo Worldwide Technologies offers comprehensive
the manual network-planning process with a comput- information technology and engineering resources for
erized process. The firm’s original plan was to pur- the Menlo Worldwide operating companies and their
chase a commercial software package and contract customers. Its service includes execution information
for any needed changes. Towards that objective, it technology and network design solutions for trans-
evaluated several packages. Some could support opti- portation, logistics, freight forwarding, and global
mization of Menlo’s truck networks but no package supply-chain management.
was found to optimize the air operations or the Menlo Worldwide Technologies’ optimization tech-
nology team focuses on developing business deci-
integrated aircraft and truck primetime network.
sion solutions for internal and external customers by
Most of the airline software packages were designed
using operations research and management science
to support aircraft maintenance, crew scheduling, techniques, including optimization, simulation, statis-
airline passenger operations, or aircraft scheduling— tics, and data-warehouse analysis. Menlo Worldwide
not network optimization. Finding no suitable soft- Technologies’ optimization technology team develops
ware package, Menlo Worldwide Forwarding selected models and business solutions using the best software
the software package that was the closest fit and for the customer—either optimization software built
worked with its developer to create a proof-of-concept in house or off-the-shelf software packages.
demonstration that focused on the optimization of the Moreover, based on its experience in developing
air network. Results of these efforts were uninspir- optimization models, Menlo Worldwide Technologies
ing. Although prototype software was developed to a emphasizes the model-development processes to
functional level, this software failed to show any real ensure that it uses the best technology for the busi-
cost savings potential. The optimized results (albeit ness operating environment. The model-development
preliminary) were no better than those of the existing processes are the following: define the project scope,
manual process. define the issues, collect and process the data, con-
In parallel to the proof-of-concept demonstration, struct and validate the model, conduct what-if analy-
ses, and develop recommendations.
the Menlo Worldwide Forwarding management team
Following these processes, Menlo Worldwide
opened a dialogue with its sister company, Menlo
Technologies systematically proceeds through the
Worldwide Technologies, to assess the feasibility, pit-
network-routing-optimization analysis, sharing its
falls, and potential benefits of developing the needed ongoing results with management. Menlo Worldwide
optimization capability in house. The results of these Technologies’ optimization-technology team works
efforts were more positive. The company discovered closely with management to ensure that the analysis
that by leveraging the operational knowledge and continually addresses those issues of primary impor-
experience within Menlo Worldwide Forwarding, and tance to Menlo Worldwide Forwarding. The team
the software development and optimization technol- can also provide good routing solutions based on
ogy expertise within Menlo Worldwide Technologies, Menlo Worldwide Technologies’ operational experi-
it could form a winning team. The team would com- ence, technological expertise, tools, and resources.
prise a small, select group of individuals from both By combining Menlo Worldwide Technologies’
companies with management oversight from Menlo proven development methodologies, knowledge of
Worldwide Forwarding. This approach would not optimization technology, and employee skills and
only capitalize on the talents resident in both compa- capabilities with the operational expertise and leader-
nies, but it would allow for a low cost, incremental ship within Menlo Worldwide Forwarding, the com-
and time-phased development of the needed opti- pany formed a winning team.
mization software that would focus on the areas of
highest return first. It would also ensure that the soft- Problem Overview
ware would be tailored to Menlo’s operational envi- Until 2001, Menlo Worldwide’s North American
ronment and would allow the company, over time, to transportation network comprised two major compo-
capture all the subtleties and business considerations nents: Menlo Worldwide Forwarding (which was then
unique to Menlo Worldwide Forwarding. Emery Worldwide) and Emery Worldwide Airlines.
Prior et al.: Menlo Worldwide Forwarding
Interfaces 34(1), pp. 26–38, © 2004 INFORMS 29

1.6 Mismatched Aircraft Capacity and Routes


1.4 To maximize potential returns and mitigate the high
1.2 fixed cost of aircraft ownership, the company manu-
1 ally developed aircraft routes to best utilize the fleet.
0.8
To offset operating expense, it filled unused aircraft
space with freight that could have been moved more
0.6
cost effectively by truck. As a result, the company’s
0.4
truck network became underutilized and did not com-
0.2 plement the aircraft network, resulting in decreased
0 margins.
Jan-00 Apr-00 Jul-00 Oct-00 Jan-01 Apr-01 Jul-01
Total Weight Average cost per LB Margin per ops. cost
Increasing Crew and Airplane Maintenance Costs
The costs of operating airline crews and maintaining
Figure 2: The graph shows the total weight shipped, the average operating aging aircraft increased over time, which further cut
cost per pound, and the margin per operating cost between January 2000
into profitability.
and August 2001. We normalized the costs and values to January 2000
data.
Solution Strategy
In 2001, the team began analyzing the problems
Emery Worldwide managed the terminal and ground and developing an optimization model to reduce the
operations and the airlines transported the next-day transportation network routing costs. Leveraging the
airfreight within North America. As business grew operational knowledge and experience within Menlo
over the past few decades, the size of the dedicated air Worldwide Forwarding, the team reviewed the chal-
fleet and the number of hubs grew and the transporta- lenges, developed modeling processes, applied new
tion network expanded. As the capacity and complex- optimization technologies and implemented model
ity of the transportation network grew, so did operat- solutions. The team built the optimization model
ing costs. Beginning in the late 1990s, Menlo World- using a combination of business operations knowl-
wide Forwarding’s operating costs steadily increased edge and the most recent optimization technologies.
even as its freight volume and profit margin shrank Menlo management fully understood and supported
(Figure 2). Several factors contributed to the sag- the optimization modeling technologies and their
ging profitability, including overcapacity in the rout- application. The optimization model was not a myste-
ing network, a reliance on high fixed-cost assets, mis- rious black box; instead, management saw the project
matched aircraft capacity and routes, and increasing as an integral part of the business decision-making
costs for crews and aircraft maintenance. process. Moreover, the team tested several modeling
options to find the best fit for the company’s business
Routing Network Overcapacity environment. The modeling technologies were flex-
In the late 1990s, when the economy looked promis- ible enough that the team could modify the model
ing and shipment volume steadily increased, man- to fit the dynamic business operations quickly. The
agers could use their years of experience to deter- team learned from other software companies to avoid
mine network capacity. However, when the economy changing the company’s business operations to fit the
modeling tools.
slowed and shipment volume dropped, they lacked
To develop the modeling tool for the operating
a sophisticated tool to optimize the network. They
environment, the team used a multiphase devel-
faced overcapacity and rising costs for operating the
opment process. In the first phase, a model was
transportation network.
developed to optimize daily aircraft routes and the
associated aircraft fleet to reduce operating costs.
High Asset-Based Operation Although the model was originally designed for use
In an expanding shipping market during the 1990s, with the Emery Worldwide Airlines fleet, the model
the company increased both the size and number of was quickly adapted for use with contractor air-
the aircraft in its fleet as well as the capacity of its craft after the Federal Aviation Administration (FAA)
Dayton sortation facility. This growth provided cus- unexpectedly grounded Emery Worldwide Airlines
tomers with excellent service during the boom, but on August 13, 2001. In the second phase, the model’s
as the economy and airfreight demand weakened this functionality was expanded to optimize the routing
high-fixed-cost network limited the company’s flexi- plans for an integrated aircraft-and-truck logistics net-
bility to reduce operating costs when shipment vol- work to increase utilization of the entire network and
umes decreased. reduce overall operating costs.
Prior et al.: Menlo Worldwide Forwarding
30 Interfaces 34(1), pp. 26–38, © 2004 INFORMS

Solution Development Benders decomposition to solve aircraft-routing and


The objective of the project is to provide optimization crew-scheduling problems simultaneously.
decision-support tools to Menlo Worldwide Forward- The Menlo Worldwide’s aircraft-scheduling model
ing with daily, weekly, and monthly adjustments to solves four of the five problems together: sched-
truck and aircraft operations to minimize their over- ule development, fleet assignment, aircraft routing,
all cost while maintaining high service levels. The and crew scheduling. Because Menlo Worldwide For-
optimization-technology group also makes ad hoc warding currently uses only contract carriers and
modifications to the model to support other require- the contractors manage the crews and the airplane
ments, such as contract negotiations, budget prepara- maintenance, the model can assume that the crews
tion, and commercial carriers. The team worked with and aircraft fleets are always available. This assump-
management to identify the following priorities. tion reduces model complexity so that an integrated
model can be developed for these four problems to
Optimizing Aircraft Scheduling solve them simultaneously.
Optimizing aircraft scheduling included minimiz- The model is intended to provide a solution for
ing operating costs by optimizing aircraft routes, the complicated problem within a minimal optimality
scheduling, carriers, and the type of aircraft on each gap (such as 0.1 percent) in a few minutes. The
route. The company also wanted to increase aircraft- team built several business-knowledge constraints
schedule flexibility and network-capacity utilization into the model to speed up the search processes
while satisfying market service requirements, provide for delivering the optimal solution. The model can
a computerized modeling system with user interface provide weekly schedules in a reasonable time, and
capability to replace the current manual scheduling because the solution computation time was improved,
processes, and develop a tool for quick what-if sce- management has been able to define more what-if sce-
nario analysis. narios and evaluate more options before setting up
the routes.
Optimizing Routing of the Integrated
Aircraft-and-Trucking Network
Expanding the aircraft scheduling optimization capa- Phase 1: Optimizing Aircraft Schedules
bility to the integrated aircraft and trucking network
was also a priority. The model would be required to Objective and Methodologies
determine whether using a plane or a truck would The objective during the first phase was to minimize
provide the most efficient transportation method the cost of weekly aircraft schedules. The single inte-
depending on the service level required (next-day, grated model optimized aircraft operations for the
second-day, or economy freight) and take both week- Tuesday-through-Friday schedule and the Saturday
day and weekend schedules into consideration. The schedule together.
ideal model would provide more cost-effective rout- In Phase 1, we formulated the aircraft-scheduling
ing and improve overall network utilization by incor- problem as an assignment model. Based on projected
porating the aircraft-and-truck network. freight volumes, operations, and service require-
Yu and Yang (1998) listed the literature on airline ments, we generated a robust series of aircraft
planning, modeling, and optimization. In general, air- routes in advance to input into our assignment-
line planning consists of solving five individual prob- optimization model. The feasible aircraft routes
lems: schedule development (determining when and were based on origin-destination pairs, freight vol-
where to fly), fleet assignment (assigning the aircraft umes, aircraft characteristics (types, availability, flying
equipment types), aircraft routing (assigning aircraft range), market times (departure and arrival time win-
to routes), crew scheduling (determining the crew dows), airport limitations (runway length, loading
itineraries or pairings), and crew rostering (assigning and unloading equipment), and other business con-
crews to crew itineraries). Yu et al. (2003) developed straints that were driven by customer service require-
a decision-support system for Continental Airlines ments. We then developed a costing model for calcu-
to generate a global optimal, or near optimal, crew- lating the route costs as they are generated.
recovery model. They described the model and the A route pairing is a combination of an inbound
solutions at the 2002 Edelman Award competition. route from any of Menlo’s logistics centers to the
Airlines currently solve the planning and optimiza- Dayton hub (HDY), such as Seattle-Portland-Dayton
tion problems separately (Klabjan et al. 2002), because (SEA-PDX-HDY), and an outbound route from the
it is not feasible computationally to solve them as Dayton hub to one of company’s logistics centers,
a single problem. Klabjan et al. developed a partial such as Dayton-Portland-Seattle (HDY-PDX-SEA).
integration model for solving three of the five prob- These feasible route pairs have the same beginning
lems together: schedule development, aircraft rout- and ending locations. The route generator provides
ing, and crew scheduling. Cordeau et al. (2001) used inbound and outbound operating hours for each
Prior et al.: Menlo Worldwide Forwarding
Interfaces 34(1), pp. 26–38, © 2004 INFORMS 31

route. The feasible routes include dummy routes, such Phase 2 Optimization Model
as Dayton-Dayton-Dayton (HDY-HDY-HDY), which We developed a time-space network to represent the
represents spare aircraft for emergency use. routing problem and used column-generation tech-
We produced two sets of feasible aircraft route pair- nology to iteratively solve the model. The nodes of the
ings: Tuesday-through-Friday routes, and Saturday time-space network represented the logistics centers.
routes. On Saturdays, Menlo has low operation The arcs represented either aircraft or truck routes
volumes and large market time windows, and it does used to move shipments between the nodes. The
not serve all logistics centers. Therefore, we have the column generator identified feasible routes on the net-
flexibility to park aircraft to reduce operating costs. work for moving a shipment from its origin to its des-
The assignment model consists of all feasible air- tination within the required service time. Based on the
craft route pairings, aircraft capacities, performance shipment opportunity costs from the previous itera-
capabilities, and a variety of other operating con- tion, the column generator derived a new set of routes
straints and costs for optimizing the aircraft sched- for the model until the solution met the optimality
ules and the utilization of network capacity. The criteria.
other operating constraints and costs include ACMI The model’s objective function is to minimize the
(aircraft, crew, maintenance, and insurance), fuel, total operating costs of the aircraft and truck net-
ground-handling, extra-hour, extra-cycle, and extra- works, including fuel costs, stop charges, daily fixed
crew costs. Extra-hour cost is incurred when the charges, block-hour costs, aircraft extra-hour costs,
monthly operational hours for each aircraft type aircraft extra-cycle costs, aircraft extra-crew costs, and
exceed the contracted hours. Similarly, extra-cycle cost airport loading and unloading costs. The model incor-
is paid when the number of cycles (each flight seg- porated critical operating constraints of the air and
ment is considered one cycle) for each aircraft type trucking network along with the service-time require-
exceeds the allowable number of cycles specified in ment and shipment-balance constraints at each node.
the contract. Extra-crew costs are incurred if the num- The nonlinear transportation costs and the complex-
ber of crews, as determined by FAA duty and rest ity of the business operating constraints of both the
requirements, exceeds the allotted number of crews
air and trucking networks complicated the Phase 2
permitted by the contract.
model. The model optimizes the network routing and
The objective function minimizes the sum of all
the associated transportation modes (that is, either
of the operating costs above as well as the costs
air or truck on each leg). By implementing the solu-
of contracting with third-party carriers. The con-
tion, Menlo Worldwide Forwarding can increase the
straints of the model are aircraft availability, logis-
cost effectiveness of the integrated network opera-
tics center coverage, extra-cost constraints, container
tions, improve the utilization of the truck network,
compatibility, connectivity constraints, minimum air-
and reduce overall network operating costs.
craft use, spare-aircraft requirements, specified-route
Realizing the complexity of the integrated opti-
requirements, and fixed-flight-segment requirements.
mization model, we developed a heuristics algorithm
In addition, the optimization-technology group built
a crew-scheduling model, which is embedded in the to speed up the computation. The algorithm deter-
assignment-optimization model to optimize the num- mines the shipment volume split between the aircraft
ber of crews needed on the network (Appendix). network and the truck network and then optimizes
the aircraft-network routing and truck-network rout-
ing separately. After that, the column generator iter-
Phase 2: Optimizing Routing of the atively generates air and truck routes to improve
Integrated Aircraft-and-Truck Network the solution of the integrated network model until it
The second phase of the model provides systemwide, meets the specified optimality.
optimal routing solutions for the integrated aircraft-
and-truck network (Appendix). To account for service
requirements and operating constraints, we formu- Data Collection and Processing
lated the problem as a multiperiod network-routing We assembled and examined a variety of data for the
optimization model for both transportation modes. model. The new Menlo Worldwide Forwarding enter-
The model traded off the costs and capacities of prise system, E2K, provides most of the shipping and
various transportation modes (aircraft or truck) at routing data. Menlo Worldwide Forwarding offices
each leg to optimize the overall routing plans. It and customers around the world use E2K (the com-
generated two sets of routing solutions for weekday pany’s multimodal communications and operations
and weekend operation to meet the different operat- network that serves as the company’s central repos-
ing constraints for shipping freight between the origin itory for shipping and financial information). Based
and destination logistics centers. on the seasonality of shipment trends, we developed
Prior et al.: Menlo Worldwide Forwarding
32 Interfaces 34(1), pp. 26–38, © 2004 INFORMS

a model for forecasting shipping volume. Using aver- Reporting and User Interface
age container weight derived from historical oper- The optimization-technology group developed a user
ational data, we converted the total freight volume interface so that users can easily input data, modify
to the equivalent number of aircraft containers. We model parameters and constraints, develop scenarios,
also determined the allowable departure-and-arrival and output routes on a North American map
time windows for each logistics center. This time win- (Figure 3).
dow is typically called the market time. It has two
important elements: the departure cutoff time and the
arrival cutoff time. Inbound to the Dayton hub, the Model Integration and
shipment departure cutoff time is the latest time that System Architecture
a shipment can arrive at a station to connect via air- The Phase 1 model functionality is embedded in the
craft or ground transportation to a required down- Phase 2 model. Therefore, we can turn off the truck-
stream sortation center. Outbound from the Dayton routing model to make the model behave like the
hub, the arrival cutoff time is the latest time that a Phase 1 model to optimize aircraft schedules sepa-
package can arrive to connect with each station’s local rately. Alternatively, we can use the Phase 2 model
delivery network. To estimate the aircraft block hours, to optimize only the truck network to determine how
we also built an enroute-time model. We based this much freight the truck network can carry. We can
model on historical aircraft performance data, and it use the remaining freight in the aircraft model to
determines the aircraft block hours (that is, block- generate aircraft schedules. The optimization models
out to block-in), taking into account each station’s were developed using the C++ programming lan-
taxi requirements, aircraft routing and directional- guage and CPLEX optimization software. Microsoft
ity, and seasonality, including winds and temperature Excel and Visual Basic were used extensively to pre-
variations. pare scenarios and reporting solutions.

Figure 3: The system seamlessly displays the routes for each scenario on a North America map. The figure shows
the inbound aircraft routes into the company’s primary hub in Dayton, Ohio for one scenario. Outbound routes,
which are not shown, might differ from inbound routes.
Prior et al.: Menlo Worldwide Forwarding
Interfaces 34(1), pp. 26–38, © 2004 INFORMS 33

periodically performs network optimization. This


Scenario organization is responsible for creating weekly air-
Generation craft and truck schedules. In its weekly planning, it
does not seek wholesale changes but rather tweaks
Raw Data
(E2K)
Optimization Modeling Data
the scheduled routes and aircraft to improve opera-
tions and accommodate any unexpected changes in
aircraft availability, market volumes, and other oper-
Route Generator ational dynamics. It uses the model proactively (for
example, to respond to new markets and logistics
Network Routing Model

center expansions) and reactively (for example, to


cope with an overload). On Monday, the freight
PHASE 2

PHASE 1
Aircraft Routing Model
Truck Routing Model
volumes from the previous week’s operations are
fed into the model along with any aircraft or
market-volume changes. By controlling the model
constraints, the schedule planner can eliminate or
consider changes in aircraft departure or arrival times,
Solution Reporting
routing options, and aircraft assignments. The opti-
mized results identify possible changes in routes and
cost savings and help schedule planners to evalu-
Routing
Database ate their impacts. The planners discuss the results
(E2K)
at schedule-planning meetings on Wednesdays and
identify schedule changes for management approval.
Menlo Worldwide Forwarding can thus minimize
User Interface
changes and rapidly adjust its operations in response
to variations in its customer base or market volumes.
Figure 4: The figure shows the high-level data flow between the optimiza-
tion models and the online shipment database. The functionality of the Planning Monthly and Quarterly Schedules
Phase 2 model includes the functionality of the Phase 1 model.
Menlo Worldwide Forwarding also uses the model
for planning monthly and quarterly schedules. The
process differs from weekly planning. First, the plan-
Information from various databases feed the opti-
ning is intended to create a robust network solu-
mization database. The optimization database pro-
tion that Menlo can fly with little change for a long
cesses the data and generates the final data that the
time. To ensure that the company is implementing a
optimization model will use (Figure 4). Then, the user
low-cost solution, the planners minimize the number
specifies the scenario constraints and runs the model
of constraints. They consider such factors as aircraft-
through the graphic user interface.
base changes, aircraft-fleet size and mix, market clo-
sures or additions, and various market times. Because
Implementation some possible changes may affect other business func-
The model has become an invaluable tool for tions, the planners must include all possible sig-
developing network plans and for analyzing Menlo nificant changes, run model iterations, and obtain
Worldwide Forwarding’s air-and-truck-transportation the approval of marketing managers, field opera-
network. Menlo uses the model to build schedules tions, and corporate management before implemen-
that meet its customers’ needs. The company also uses tation. Accordingly, for both monthly and quarterly
the model to minimize aircraft operating costs and to planning, planners go through iterative processes
use the assets available within the Menlo Worldwide that rely heavily on parametric analysis to resolve
Forwarding network more efficiently. Although we conflicting business interests. They use the model’s
are constantly finding new applications for the model, results to identify the lowest-cost solution within
Menlo Worldwide Forwarding is currently using it specified network parameters and operational con-
for the following tasks: weekly planning, planning straints and to quantify the cost impacts of departing
monthly and quarterly schedules, preparing the bud- from a baseline solution. Because Menlo Worldwide
get, and analyzing contracts. Forwarding relies on contracted aircraft and on other
air-cargo space available through commercial sources,
Weekly Planning the model results help ensure that Menlo contracts for
To take full advantage of the model’s cost- adequate aircraft capacity but not excess capacity. The
saving opportunities, the transportation-and-logistics results also ensure that the types, number, and mix
organization within Menlo Worldwide Forwarding of aircraft match the lane volumes and operational
Prior et al.: Menlo Worldwide Forwarding
34 Interfaces 34(1), pp. 26–38, © 2004 INFORMS

and market requirements. The model also enables the 1.6

planners to adjust the network for expected changes 1.4


in Menlo’s customer base, for expansion or contrac-
tion in the number or location of Menlo’s logistics
1.2

service centers, and for seasonal variations in freight 1

volumes. 0.8

The Budget 0.6

To create a realistic operating budget, Menlo must 0.4


forecast shipment volumes for the coming year and
design air and ground networks to accommodate 0.2

these volumes. Because planners must consider many 0


assumptions and possible network trade-offs, they Jan-00 Apr-00 Jul-00 Oct-00 Jan-01 Apr-01 Jul-01 Oct-01 Jan-02 Apr-02 Jul-02 Oct-02

use an iterative process. The model allows them Total Weight Average cost per LB Margin per ops. cost

to rapidly perform accurate trades and identify


and evaluate their impacts. For example, in 2002 Figure 5: The graph shows the total weight shipped, the average operating
the model was used for the first time to develop cost per pound, and the margin per operating cost between January 2000
and December 2002. We normalized the costs and values to January 2000
Menlo Worldwide Forwarding’s 2003 air-haul budget. data.
Although this process generated many more network
variations than in previous years, both the number
of man-hours required and the duration of time to cost and increase operating margin. Based on our
complete the budget were reduced. Throughout the analysis of the model results, Menlo Worldwide For-
process, the model proved to be a valuable tool for warding implemented the solutions and dramati-
generating and costing the initial baseline routings cally improved utilization of its network capacity. It
and in iterating subsequent network changes. reduced its cost for air-haul operations and increased
its profitability despite the weak economy (Figure 5).
Vendors and Contracts
By using the model, Menlo improved its utiliza-
Menlo Worldwide Forwarding contracts with third-
tion of the capacity of the North American routing
party air and truck carriers for most of its opera-
by 30 percent, reduced operating costs by 21 per-
tions. Because most of the company’s network costs
cent, and improved operating margins by 41 percent.
and its next-day-delivery performance depend on air-
These reductions in costs accounted for 62 percent of
craft operations, it must choose aircraft sizes, types,
the company’s profitability increase in 2002. Menlo
and vendors carefully. Moreover, provisions in the
also increased its management efficiency, transformed
contracts determine the actual cost of operations. For
its business model, and integrated its truck-and-air-
example, monthly minimums and the costs for extra
routing networks. Finally, Menlo is planning to trans-
hours, extra cycles, and extra crews are based on
the rates and parameters specified in the contracts. port the model to its European routing network.
By running the models for a fixed shipment volume,
Network Capacity Utilization
we can perform parametric analyses. These analyses
Menlo used the optimization model to evaluate all
not only allow us to determine the best complement
possible combinations of routes and aircraft to deter-
of aircraft (that is, the number, type of aircraft, and
mine the best routes and aircraft to fly those routes.
operator) to meet specific network requirements at
the lowest operating costs, but it also allows us to The model helped Menlo Worldwide Forwarding’s
identify the aircraft and routing changes and quan- management to reduce the size of the fleet by using
tify the operating cost differential if something other the right size aircraft for each route and increasing
than the optimal mix is selected. By examining the the use of available capacity. It reduced the number
results, we can also determine how specific contract of aircraft by 48 percent and network capacity by
provisions are influencing cost and aircraft selection. 59 percent. We also increased the utilization of net-
We can then use this information to negotiate more work capacity by 30 percent and reduced average air-
favorable contracts with our aircraft operators. craft size by 20 percent (Table 1).

Operating Costs and Margin


Benefits We used the model results to minimize the network
In September 2001, we began giving management operating costs by optimizing the routes, carriers, air-
what-if routing scenarios to review. Each itera- craft, fuel consumption, and flight-crew assignments.
tion of our modeling exercise resulted in a good Because Menlo implemented results in September
solution that helped management reduce operating 2001, it has reduced average air-haul operating cost
Prior et al.: Menlo Worldwide Forwarding
Interfaces 34(1), pp. 26–38, © 2004 INFORMS 35

Operational Jan 2000 Jan 2001 Jan 2002 Jan 2003 week. The model’s speed enables Menlo management
Metric (%) (%) (%) (%) to evaluate many scenarios before creating schedules.
By analyzing multiple what-if scenarios, management
Aircraft network 100 87 59 54 can identify the root causes of service problems and
shipment volume develop better routing solutions. It also uses the
Number of primetime 100 88 58 52 model to create new routes ad hoc during the week if
aircraft
necessary.
Aircraft network capacity 100 87 47 41
Average aircraft size 100 98 81 80 The optimization model also proved valuable for
Aircraft network 100 101 126 130 creating a new flexible network quickly. When the
capacity utilization FAA unexpectedly grounded Emery Worldwide Air-
lines on August 13, 2001, we used a prototype
Table 1: The North American network metrics for January 2000, 2001,
of the model to develop new routes using third-
2002, and 2003 (values normalized to year 2000 data) show improvement
between January 2000 and 2003. January was selected for the comparison party contractors to replace the company-owned
because it is one of the lowest margin months. planes. Over the weekend, Menlo Worldwide For-
warding switched to the new contracted airlift with
no disruption in service—a remarkable feat by the
per pound by 21 percent, increased its operating mar- business team that rapidly implemented the model’s
gin by 41 percent, and saved $80 million in operating outputs.
costs in 2002. Part of the $80 million was a $12 million
reduction in average fuel costs (Table 2). The New Business Model
In December 2001, Menlo Worldwide Forwarding,
Contribution to Profitability a heavily asset-based business that owned or held
After a staggering loss in 2001, Menlo exceeded long-term leases on aircraft, became a flexible net-
its financial targets in 2002 in large part through work made up of third-party aircraft contractors on
routing optimization. Between 2001 and 2002, Menlo short-term leases. Our technical capabilities made it
Worldwide Forwarding improved its results by $130 easy to modify the model to incorporate the prin-
million. The North American network-routing opti- cipal cost elements for the new fleet of contracted
mization contributed $80 million, or 62 percent, of airlift. We used the model to optimize total aircraft
the increase in profitability. The optimization project network costs by integrating contractual requirements
has also improved the company’s financial prospects. and terms, such as the number of crews, the air-
Prior to implementing the optimization project, Menlo craft cycle ratio, and operating hours, into the model
Worldwide Forwarding’s profit margin was contin- to minimize the overall operating costs. The aircraft-
ually shrinking. After it implemented the model’s cycle ratio, crew costs, and extra hourly charges for
solutions, its operating costs dropped and its margin operating aircraft were crucial new cost elements. We
dramatically increased (Figure 5). also embedded the crew-assignment model in our
model to optimize overall operating costs and simul-
Increased Management Efficiency taneously provide the optimized number of crews
The optimization model has proven to be a great and crew assignments on each route.
decision-support tool that continues to improve the Thanks to the success of our model, Menlo World-
cost-effectiveness of Menlo Worldwide Forwarding’s wide Forwarding moved rapidly from a fixed-cost
business operations by expanding management’s fleet system to a highly dynamic flexible system. Man-
route-planning capabilities. Menlo Worldwide For- agement can take the outputs of the optimization
warding uses the modeling tool and the shipment model and quickly translate them into changes in the
forecasts to develop an optimal routing solution each company’s contracted airlift capacity.

Cost Metric 2000 (%) 2001 (%) 2002 (%)


Conclusion
In 2002, one year after adopting the network-routing-
Air haul operating cost per pound 100 111 79 optimization model, Menlo Worldwide Forwarding
Margin per operating cost 100 67 141 had increased its utilization of network capacity by
30 percent and reduced operating cost by $80 million.
Table 2: This table of operating costs and margin between 2000 and 2002 Menlo Worldwide Forwarding and Menlo World-
(all values normalized to year 2000 data) demonstrates the model’s ben-
efits after its implementation in September 2001. The costs related to
wide Technologies worked together to develop the
operating both Emery Worldwide Airlines and the contracted carriers were network-routing-optimization model that optimized
embedded in our 2001 cost data; therefore, we used 2000 as our baseline Menlo’s North American transportation network, and
to estimate the total cost saving. it has become essential for daily routing decisions.
Prior et al.: Menlo Worldwide Forwarding
36 Interfaces 34(1), pp. 26–38, © 2004 INFORMS

We are pleased to have had the opportunity to Capa : the capacity of the equipment a where a ∈
apply optimization technology and develop what is AC ∪ TK.
now Menlo Worldwide Forwarding’s most important W T s: shipment size where s ∈ S.
routing tool. The optimization model solved the com-
pany’s aircraft-scheduling, fleet-assignment, aircraft- Variable Definition
routing, and crew-scheduling problems simultane- xdar : equal to 1 if route r is selected at day d; 0 other-
ously. We expect to use our knowledge and apply our wise; where r ∈ AR ∪ T R, d ∈ D, and a ∈ AC ∪ TK.
optimization technology further for other Menlo com- ya : total number of aircraft a in the network where
ponents and our external customers. a ∈ AC.
zha : extra hours for aircraft a where a ∈ AC.
zca : extra cycles (legs) for aircraft a where a ∈ AC.
Appendix: Model Formulation zwa : extra crew of aircraft a where a ∈ AC.
zwdat : extra crews of aircraft a at the block hour time
Sets and Parameters bucket t ∈ T .
N : the set of logistics center locations and hubs.
AR: the set of aircraft routes, including spare and
dummy routes at the Dayton hub or logistics centers.
Phase 1 Model: The Aircraft
T R: the set of truck routes. Scheduling Model
AC: the set of aircrafts.
Phase 1 Model Formulation
TK: the set of trucks.
D: the set of operation days that includes Tuesday-    c c
Minimize Cdar xdar + ha zha + r a za
through-Friday and Saturday operations. d∈D a∈AC r∈AR a∈AC a∈AC
T : the number of time buckets for crew service time. 
+ raw zwa "
M: the set of service types (next day, second day,
a∈AC
and economy).
S: the set of shipments from origin logistics center i Phase 1 Model Constraints
to destination logistics center j at day d service level m (1) Aircraft availability: Total number of aircraft in
where i j ∈ N , d ∈ D, and m ∈ M. the system (flying, spare, and parked) should be less
E: the set of container types. than or equal to available aircraft.
T : number of time intervals to group routes based 
on the block hours. xdar ≤ ya ∀d ∈ D a ∈ AC
r∈AR

Coefficients and Parameters y a ≤ Ma ∀a ∈ AC"


The subscripts r, a, i, and d represent routes, equip-
(2) Minimum aircraft use:
ment, logistics center, and operation day, respectively.
cdar : operating cost of route r for day d where r ∈ 
xdar ≥ Na ∀d ∈ D a ∈ AC"
AR ∪ T R, d ∈ D, and a ∈ AC ∪ TK. r∈AR
bdar : block hours for route r where r ∈ AR, d ∈ D,
and a ∈ AC. (3) Spare aircraft constraints: Specify the mini-
gdar : number of flight legs of route r where r ∈ AR, mum number of spare aircraft required for regular
maintenance.
d ∈ D, and a ∈ AC.
Ma : available aircraft a where a ∈ AC. 
srxdar ≥ Oda ∀a ∈ AC d ∈ D
Nda : minimum number of aircraft a for day d where r∈AR
a ∈ AC and d ∈ D.
Ha : fixed monthly operation hours for aircraft a where sr is an indicator function; sr takes on the
where a ∈ AC. value 1 if r is a spare route and 0 otherwise.
ha : cost of an extra hour for aircraft a where a ∈ AC. (4) Logistics center coverage: Each logistics center
Rca : cycle (flight leg) ratio for aircraft a where a ∈ should be covered only once for inbound and out-
AC. bound flight.
rac : cost of an extra cycle for aircraft a where a ∈ AC. 
IBrLr ixdar = 1 ∀d ∈ D i ∈ N \&HDY (
Rwa : crew ratio for aircraft a where a ∈ AC. a∈AC r∈AR
raw : cost of an extra crew for aircraft a where a ∈ AC. 
DNd : number of operating days per month for OBrLr ixdar = 1 ∀d ∈ D i ∈ N \&HDY (
a∈AC r∈AR
day d where d ∈ D.
Oda : number of spares of aircraft a requirement for where IBr is an indicator function; IBr takes on
day d where a ∈ AC and d ∈ D. the value 1 if r is an inbound route from a logistics
Prior et al.: Menlo Worldwide Forwarding
Interfaces 34(1), pp. 26–38, © 2004 INFORMS 37

center to the Dayton hub and 0 otherwise. OBr is an For example, the constraint does not apply to the
indicator function; OBr takes on the value 1 if r is Dayton hub.
an outbound route from the Dayton hub to a logistics 
center and 0 otherwise. Lr i is an indicator func- M∗ CP e i aOBrLr iXdar
r∈AR
tion; Lr i takes on the value 1 if the route r stops at 
logistics center i and 0 otherwise. ≥ CP e i acIBrLr iXdar
(5) Extra hour constraints: Specify the monthly r∈AR

fixed hours requirement and the extra hours that are ∀i ∈ N  d ∈ D e ∈ E and M is a big number
incurred from our routing plans.
where CP e i a is an indicator function; CP e i a

DNd bdar xdar ≤ Ha ya + zha ∀a ∈ AC" takes on the value 1 if container type e is compatible
d∈D a∈AC r∈AR for aircraft a at logistics center i and 0 otherwise.
(9) Inbound and outbound aircraft-equipment bal-
(6) Extra cycle constraints: A cycle is a flight leg ance constraints: The aircraft equipment at the ori-
from an aircraft’s takeoff to landing. This constraint gins of the inbound routes and the destinations of the
specifies the number of extra cycles that the network outbound routes should be balanced every day. At
used for aircraft a. the Dayton hub, denoted by HDY, the outbound air-
  craft equipment is balanced with the previous day’s
d∈D r∈AR bdar xdar  inbound aircraft equipment. A dummy route that has
≤ gdar xdar + zca ∀a ∈ AC"
Ra
c
d∈D r∈AR the same origin, intermediate, and destination loca-
tions represents the aircraft parked at the logistics
(7) Extra crew constraints: An extra crew member center.
would be added into the network to avoid the total  
flight hours of a crew member exceeding FAA daily Dd r iOBrXdar = Od r iIBrXdar
r∈AR r∈AR
flight hour limitation. By matching a long inbound
route with a short outbound route or a short inbound ∀i ∈ N \&HDY ( d ∈ D a ∈ AC
route with a long outbound route, the network can
keep the crews’ flight hours below the FAA’s limit and 
Dd − 1 r iIBr iXd−1ar
avoid extra crews. The model partitions the FAA daily r∈AR
flight-hour limitation into T time buckets, and each 
time bucket has the same time span (say 15 minutes). = Od r iOBr iXdar
r∈AR


t  ∀i ∈ &HDY ( d ∈ D a ∈ AC
zwdat + IBrT Mr kxdar
k=1 r∈AR where Dd r i is an indicator function that takes on

T  the value 1 if i is the destination of route r starting
≥ OBrT Mr kxdar at day d and 0 otherwise. Od r i is an indicator
k=t+1 r∈AR function that takes on the value 1 if i is the origin of
∀a ∈ AC t = 0 1 2 3 " " "  T − 1 route r starting at day d and 0 otherwise.
 

T 
zwdak ≤ Rwa − 1 xdar + zwa ∀a ∈ AC d ∈ D Phase 2 Model: Master Model for
k=1 r∈AR Aircraft-and-Trucking-Network
where T Mr t is an indicator function for the block Routing Model
hour bucket; T Mr t takes on the value 1 when the The functionality of the Phase 2 model includes all
total block hours of route r belong to the time bucket t the functionality of the Phase 1 model. The formula-
and 0 otherwise. tion of these two models is similar in the aircraft net-
(8) Container compatibility constraints: Most of work. Both aircraft model costs include all principal
the logistics centers used the containers unloaded aircraft operating costs, such as the standard aircaft,
from outbound routes to load shipments. Because the crew, maintenance, and insurance costs (ACMI), block
inbound aircraft may be different from the outbound hour, fuel, airport landing-and-departure, loading-
aircraft, this constraint ensures that the container type and-unloading, extra-cycle, extra-hour, and the extra-
is compatible with the inbound aircraft and the out- crew costs. The trucking routing model and costs
bound aircraft. Container compatibility constraints do are added into the Phase 2 model. A shipment can
not apply to all logistics centers because some logis- be shipped by either truck or air on multimode-
tics centers can handle multiple types of containers. transportation multileg routes. A shipment can be
Prior et al.: Menlo Worldwide Forwarding
38 Interfaces 34(1), pp. 26–38, © 2004 INFORMS

dropped at an intermediate logistics center for trans- (3) Shipment balance constraint at any intermedi-
fer to other routes. The model must ensure that every ate locations or hubs: This constraint ensures that no
shipment is picked up at its origin and delivered shipment stays at any intermediate hub location. A
to its destination. This increases the model size and shipment has to be picked up on the next day when
the complexity of the solution methodologies. The it is dropped at an intermediate hub location.
integrated aircraft-and-truck network became compli- 
cated, and we used column-generation technology to V r s iXa d+1 r
a∈AC∪TK r∈AR∪T R
solve the Phase 2 model.

+ W r s iXadr = 0
Phase 2 Model Formulation a∈AC∪TK r∈AR∪T R
 
Minimize Cdar xdar + ha zha ∀s ∈ S d ∈ D i ∈ N 
d∈D a∈AC∪TK r∈AR∪T R a∈AC
  where i is an intermediate hub that is neither the ori-
+ rac zca + raw zwa "
a∈AC a∈AC gin nor the destination of shipment s.
(4) Route capacity constraint: This constraint en-
Model Constraints sures that the total shipping on a route does not
The master problem of the Phase 2 model includes all exceed its capacity.
the constraint sets except set 4 of the Phase 1 aircraft- 
network model. The Phase 2 model includes the fol- Capa xadr ≥ V r s iW T s ∀r ∈ AR ∪ T R"
lowing extra constraints: Constraints 1 through 3 for s∈S i∈N
ensuring that the shipment travels from its origin to
its destination by either aircraft or trucks. Constraint 4
Acknowledgments
is the weight limitation for each route. The shipment-
John Williford, president and chief executive officer of
service-requirement constraint is included in the col- Menlo Worldwide; Ron Berger, managing director of Menlo
umn generator to make sure the route meets the Worldwide Technologies; and Neal Wagner, director of
service time requirement. Menlo Worldwide Forwarding, have fully supported this
(1) Shipment pickup constraint: This constraint project from day one. We also appreciated input from
ensures that a shipment is picked up from its origin i. Nancy Colvert, Jamie Fenimore, Jeff Link, and Brian Tuttle
 that helped us in polishing this paper.
V rsiXadr = −1 ∀s ∈ S d ∈ D i ∈ N 
a∈AC∪TKr∈AR∪T R
References
where V r s i is an indicator function that takes on
Cordeau, Jean-Francois, Coran Stojkovic, Francois Soumis, Jacques
the value −1 if the route r picks up shipment s at Desrosiers. 2001. Benders decomposition for simultaneous
shipment origin i and 0 otherwise. aircraft routing and crew scheduling. Transportation Sci. 35(4)
(2) Shipment drop-off constraint: This constraint 375–388.
ensures that a shipment is delivered to its destina- Klabjan, Diego, Ellis L. Johnson, George L. Nemhauser, Eric
Gelman, Srini Ramaswamy. 2002. Airline crew scheduling with
tion i.
time windows and plane-count constraints. Transportation Sci.
 36(3) 337–348.
W r s iXadr = 1 ∀s ∈ S d ∈ D i ∈ N  Yu, Gang, Jian Yang. 1998. Optimization application in the airline
a∈AC∪TK r∈AC∪T R industry. Ding-Zhu Du, Panos M. Pardalos, eds. Handbook of Com-
binatorial Optimization. Kluwer Academic Publishers, New York.
where W r s i is an indicator function that takes on Yu, Gang, Michael Arguello, Gao Song, Sandra McCowan, Anna
the value 1 if the route r drops off the shipment s at White. 2003. A new era for crew recovery at Continental Air-
shipment destination i and 0 otherwise. lines. Interfaces 33(1) 5–22.