You are on page 1of 16


Vol. 39, No. 5, SeptemberOctober 2009, pp. 460475 doi 10.1287/inte.1090.0450

issn 0092-2102  eissn 1526-551X  09  3905  0460 2009 INFORMS

An Integrated Outbound Logistics Model for

Frito-Lay: Coordinating Aggregate-Level
Production and Distribution Decisions
Sla etinkaya, Halit ster
Department of Industrial and Systems Engineering, Texas A&M University, College Station, Texas 77843

Gopalakrishnan Easwaran
Department of Engineering, School of Science, Engineering and Technology, St. Marys University,
San Antonio, Texas 78228,

Burcu Baris Keskin

Department of Information Systems, Statistics, and Management Science, University of Alabama,
Tuscaloosa, Alabama 35487,

In this paper, we describe research to improve Frito-Lays outbound supply chain activities by simultaneously
optimizing its inventory and transportation decisions. Motivated by Frito-Lays practice, we rst develop a
mixed-integer programming formulation from which we develop a large-scale, integrated multiproduct inven-
tory lot-sizing and vehicle-routing model with explicit (1) inventory holding costs, truck loading and dispatch
costs, and mileage costs; (2) production, storage, and truck capacity limitations; and (3) direct (plant-to-store)
and interplant (plant-to-plant) delivery considerations. Second, we present an iterative solution approach in
which we decompose the problem into inventory and routing components. The results demonstrate the impact
of direct deliveries on distribution costs and show that direct deliveries and efcient inventory and routing
decisions can provide signicant savings opportunities over two benchmark models, one of which represents
the existing Frito-Lay system. We implemented our models using an application that allows strategy evaluation,
analysis of output les, and technology transfer. This application was particularly useful in evaluating potential
direct-delivery locations and inventory reductions throughout the supply chain.
Key words: coordinated logistics; integrated inventory; transportation decisions; vendor-managed inventory
and delivery; inventory lot sizing; vehicle routing.
History: Published online in Articles in Advance June 10, 2009.

F rito-Lay North America (FLNA) operates a large

and complex supply chain. The company, a divi-
sion of PepsiCo, employs more than 40,000 people
33 production facilities, 192 DCs, and 1,535 bins in the
United States as well as additional plants, DCs, and
bins in Canada. Consequently, FLNA faces complex
and produces a wide variety of snack foods. It owns outbound logistics planning issues.
a private eet, which consists of more than 1,400 FLNA participates in a vendor-managed inven-
over-the-road (OTR) trucks and 20,000 route-delivery tory and delivery (VMI/D) program. Under a typical
(RD) trucks of various capacities, for the outbound VMI/D program, a vendor is empowered to con-
transport of these products to many delivery loca- trol its resupply timing and quantity at downstream
tions; these include distribution centers (DCs), bins, locations (etinkaya and Lee 2000). Because effec-
and customers (stores). Its overall distribution oper- tively utilizing transportation resources is imperative,
ation also includes interplant (plant-to-plant) ship- a vendor is more likely to dispatch full-vehicle out-
ments because each plant does not produce each bound loads to achieve economies of scale. Using
product group. FLNAs offerings comprise 16 differ- this program also makes coordinating inventory and
ent groups of produced and nonproduced items, with outbound transportation decisions easier. We devel-
thousands of stock-keeping units (SKUs). It operates oped a model to allow simultaneous optimization of
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
Interfaces 39(5), pp. 460475, 2009 INFORMS 461

these decisions to improve outbound supply chain that trucks dispatched from the factory warehouse
efciency and, thus, enhance the success of FLNAs must visit. In this subproblem, we utilize the weekly
VMI/D program. replenishment and shipment quantities, which we
We focused on the outbound supply chain of a obtain from solving the inventory subproblem, as dis-
major FLNA plant in Irving, Texas, and developed tribution requirements at the delivery locations. The
an integrated cost-minimization model that links pro- objective of the routing subproblem is to minimize
duction output and shipment quantities destined to loading and routing costs and maximize truck uti-
delivery locations; thus, it allows demand to be sat- lization. Clearly, convergence might occur to a local
ised in a cost-effective and timely manner. The optimum, and it is impossible to benchmark solution
model is a mixed-integer programming formulation quality relative to the global optimum. To demon-
that seeks to achieve the following: strate the effectiveness of our optimization model and
(1) Inventory optimization and reduction at the proposed solution approach, we compare our results
Irving plants warehouse (called the factory ware- to two practical benchmarks, the chase policy and the
house henceforth); cube-out policy (see descriptions in Benchmark Models
(2) Load optimization for transportation from the and Benets Realized). The former is of interest to the
factory warehouse to the delivery locations (i.e., DCs, company for policy comparison purposes; the latter
bins, plants, and direct-delivery (DD) customers), represents its usual practice. The solution approach
while allowing direct (plant-to-store) and interplant for the benchmark models relies on the solution that
shipments; and we developed for our model. Using numbers, we
(3) Flow optimization throughout the entire out- show that run times in our solution approach are
bound supply chain. reasonable, and our model provides signicant cost
Our models uniqueness is its ability to coordinate savings.
inventory and shipment quantities with truck load- We organize the remainder of the paper as follows.
ing and routing decisions. The model also considers We present a brief review of the relevant literature
the option that DD customers can receive supplies in Related Literature. In Problem Setting and Formula-
directly from the factory warehouse, DCs, and bins. tion, we develop a mathematical formulation of the
Hence, we also seek to identify the DD customers in problem. In Solution Methodology, we discuss prob-
a customer set and measure the value of direct deliv- lem properties to show the difculties of developing
eries to them. solutions for our problem; we also present our pro-
The problem we address is a large-scale, integrated posed approach. We describe the software application
multiproduct inventory lot-sizing and vehicle-routing that combines the database and the solution approach
problem with explicit inventory holding costs, truck in Application Modules. In Benchmark Models and Bene-
loading and dispatch costs, and mileage costs; pro- ts Realized, we discuss our proposed models perfor-
duction, storage, and truck capacity limitations; and mance and demonstrate that our approach produces
direct and interplant delivery considerations. It is high-quality solutions when compared to the bench-
a challenging generalization of the vehicle-routing mark models. Finally, we summarize our ndings in
problem, which is NP-hard. Therefore, an exact opti- Conclusions.
mal solution is impractical. We also cannot obtain a
solution using existing software and FLNAs real data.
We overcome these difculties by developing a new Related Literature
heuristic that decomposes the overall problem into Our problem deals with integrated logistical planning
inventory and routing subproblems. with a specic emphasis on coordinating inventory
In the inventory subproblem, we determine the and transportation decisions. A large body of quan-
nal set of DD customers and the weekly replenish- titative research exists to address integrated logis-
ment and shipment quantities to the delivery loca- tical planning for the simultaneous optimization of
tions while minimizing relevant costs. In the routing various types of decisions, including facility location
subproblem, we nd the subsets of delivery locations and capacity, production, inventory, and distribution.
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
462 Interfaces 39(5), pp. 460475, 2009 INFORMS

However, despite the growing interest in this topic, research is also abundant (Anily and Federgruen 1993,
our specic problem has not been previously ana- Baita et al. 1998, etinkaya 2004, Federgruen and
lyzed in the literature. Simchi-Levi 1995, Viswanathan and Mathur 1997).
Academic research has successfully documented We note that our problem has several characteris-
the cost savings associated with integrated decision tics of the inventory-routing problem and its exten-
making for logistical planning. A detailed review of sions (Archetti et al. 2007, Bertazzi 2008, Bertazzi et al.
this general topic is beyond our scope. For related 2005, Campbell and Savelsbergh 2004, Fumero and
literature, we refer the reader to Baita et al. (1998), Vercellis 1999, Kleywegt et al. 2002, Lei et al. 2006).
etinkaya (2004), Keskin and ster (2007), Sarmiento The inventory-routing problem, studied by Campbell
and Nagi (1999), and Thomas and Grifn (1996); we and Savelsbergh (2004), is a variation of the vehicle-
restrict our attention to previous research that specif- routing problem in which the vendor can make
ically investigates the interaction between inventory decisions about the timing, sizing, and routing of
and transportation decisions. We note that the tradi- deliveries without allowing any shortages. However,
tional inventory models largely ignore this interac- most existing literature on this problem is method-
tion because the early theory assumes that demands oriented (e.g., large-scale mixed-integer programs); a
should be satised as soon as they arrive and the few exceptions (e.g., Lei et al. 2006) focus on real-life
customer should cover delivery costs. Consequently, applications.
the outbound transportation costs are sunk costs Although the specic application setting in this
(Hadley and Whitin 1963, pp. 1718). However, this paper is similar to the inventory-routing problem, we
is no longer true for private-eet outbound distribu- consider a more general supply chain setting than the
tion in general and for VMI/D in particular, both previous literature does. Our setting includes mul-
of which have received signicant academic atten- tiple product groups and intermediate tiers; these
tion during the past decade (Bertazzi et al. 2005, represent the DCs and bins in which customers do
etinkaya and Lee 2000, Kleywegt et al. 2002, Toptal not carry more than one periods inventory and can
and etinkaya 2006). replenish from the DCs, bins, and directly from the
Specically, two types of research emphasize the factory warehouse.
interaction between inventory and transportation
decisions. The rst type focuses on including trans-
portation costs and (or) capacities in traditional Problem Setting and Formulation
inventory problems without explicitly modeling We consider a nite planning horizon in which each
transportation-related (e.g., routing, consolidation, period represents one week. Multiple DCs, bins,
etc.) decision variables such as classical economic plants (that receive interplant shipments), and cus-
order quantity, dynamic lot sizing, stochastic dynamic tomers constitute the delivery locations on the out-
demand, one-warehouse multiretailer, joint replenish- bound supply chain. We model the tiers and ows
ment, and buyer-vendor coordination problems. It is of the outbound supply chain (Figure 1). The ows
worthwhile to note that the previous work on this include (1) production output at the Irving plant;
topic (Alp et al. 2003, Chan et al. 2002, Diaby and (2) nished-product inventories at the factory ware-
Martel 1993, Lee et al. 2003, Lipmann 1969, Toptal house; (3) replenishment quantities from the factory
and etinkaya 2006, ster et al. 2008, Van Eijs 1994) warehouse to the DCs and bins, plant-to-store ship-
focuses on stylistic supply chain settings (e.g., single ment quantities to DD customers, and interplant ship-
vendor and single buyer, single vendor and multi- ment quantities to the other plants; (4) shipment
buyer, etc.) and neglects the vehicle-routing consider- quantities from the DCs and bins to customers; and
ations that we address. (5) customer demand that can be satised by ship-
The second set addresses the integration of inven- ments from DCs and bins and (or) direct supply from
tory and transportation decisions by explicitly con- the factory warehouse. Our models objective is to
sidering both inventory- and transportation-related minimize the handling, inventory, loading, routing,
decision variables. The literature on this latter type of and delivery costs associated with the ows. Because
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
Interfaces 39(5), pp. 460475, 2009 INFORMS 463

For period t and product j Other DCs (set D)



Rdjt Sdijt


Pjt Idjt

plant Rbjt Customers
Factory Staging/ Sbijt (set II))
warehouse loading


Bins (set B)

Figure 1: The gure illustrates our mathematical programming formulation, which represents the tiers and ows
of the underlying network and characterizes the outbound supply chain of the Irving plant.

our focus is outbound logistics, we do not model the Then, the demand constraints (1) and (2) dictate that
production process in detail, and we omit produc- customer demand is satised by timely shipments
tion costs. However, we explicitly model the produc- to customers from DCs, bins, and the factory ware-
tion capacities. One output of our model is a target house. These constraints also dictate that weekly total
production output plan that is useful in developing shipments to a customer location do not exceed the
detailed production schedules. weekly demand, because FLNAs VMI/D program
We introduce our notation and present the com- ensures that customer locations never receive more
plete mathematical formulation in the appendix. To inventory than their weekly demand species.
explain our formulation, we next examine the tiers of We note that we model the demand for each product
the underlying supply chain from customers to the group at a customer location in a given week using an
Irving plant and introduce the problem constraints. aggregate number of cartons. The actual cartons used
Constraints Related to Inventory Decisions for each product group are characterized by varying
for Tiers 14 sizes, i.e., storage volumes. To represent demand in an
Analyzing the chain in Figure 1 backwards, the aggregate fashion, we dene a cubic carton, i.e., a unit
rst tier consists of the customers who generate the measure for our volume calculations, as the following
demand ows that can be satised by shipping from example shows. Suppose that, for a product group, we
DCs and bins. For some customers, a DD can be sent use two types of cartons with storage capacities of 0.5
from the factory warehouse; they can be served on cubic feet and 1.5 cubic feet, respectively. Furthermore,
the same route with DCs and bins. To determine the let the usage proportions (historical data provided) for
source(s) of supply for each customer, we use the cus- the two carton types be 0.75 and 0.25, respectively. We
tomers DC or bin assignment and DD information. dene a cubic carton for this product group with a
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
464 Interfaces 39(5), pp. 460475, 2009 INFORMS

storage volume given by the weighted average of the plan. These target production output levels account
volumes of the carton types available for the product for the loading and delivery schedules explicitly (we
group: 07505 + 02515 = 075 cubic feet. Then, discuss the loading and delivery considerations in the
we compute the total demand for a product group at next paragraph). When we have computed a numer-
a customer location in a given week (in required cubic ical solution for the proposed model, we can use
feet) by multiplying the storage volume of the cubic the resulting target production output levels (of indi-
carton for the product group by the total number of vidual products in successive periods) as translated
actual cartons (data provided). demand values in the existing production planning
Next, considering the second tier of the chain, it and scheduling software. The target production out-
is straightforward to develop the inventory-balance put levels should not violate the aggregate capacity
constraints at the DCs (3) and bins (4). The vari- and product-based constraints (10) and (11).
ables representing the replenishment quantities of the
products sent from the factory warehouse to the DCs Constraints Related to Truck Loading
and bins, in successive periods, link the second (DCs and Routing at Tier 3
and bins) and the third (staging and loading area of Figure 2 illustrates the loading and routing operations
the factory warehouse) tiers of the chain in Figure 1. between the factory warehouse and the DCs, bins, DD
These variables also represent the truck loads that are customers, and other plants. We model these opera-
en route from the factory warehouse to the DCs and tions explicitly using constraints (14)(23).
bins in successive periods. Hence, we also use them First, let us consider the loading and routing com-
to develop additional constraints that represent the ponent of the problem associated with the outbound
loading and routing operations. We note that inven- shipments from the factory warehouse (Figure 2). Our
tory levels at the DCs and bins should not violate objective is to minimize the outbound transportation
the capacity limits; hence, we have the corresponding costs from the factory warehouse and to maximize
storage capacity constraints (5) and (6). truck utilization. The OTR trucks based at the factory
Considering the ow conservation at the staging warehouse are identical.
and loading area of the factory warehouse, we have The link between the routing operations and the
the load balance constraints (7), which consider the ow constraints is as follows. The load balance con-
total load from the factory warehouse to the DCs, straints consider the total loads of the products to be
bins, DD customers, and other plants. It is worth not- shipped in successive periods from the factory ware-
ing that, in these constraints, the interplant replenish- house to the delivery locations. The components of the
ment quantities are model parameters that represent total load shown on the left side of the load balance
the nonproduced item requirements at the other constraints represent the replenishment and shipment
plants. The variables representing the total loads of quantities that should be loaded on trucks that are
the products shipped from the factory warehouse, en route from the factory warehouse to the delivery
in successive periods, link the staging and loading locations. Then, it is straightforward to verify the total
area with the fourth tier, the factory warehouse. Con- load and truck volume constraints (14) and drop-off
sidering the replenishment (production) quantities of constraints (15)(18). Note that the drop-off constraints
the products in successive periods, we also have are associated with the set of delivery locations of the
the corresponding inventory balance constraints at routing component of interest. The consequent truck
the factory warehouse (8). The total inventory at the arrival and departure constraints (constraints 1922)
factory warehouse should not violate the capacity assure that (1) the number of trucks arriving at a deliv-
therein; therefore, we also have the storage capacity ery location is equal to the number of trucks leav-
constraints at the factory warehouse (9). ing that location, and (2) all trucks leaving the factory
Because we model the target production output lev- warehouse eventually return to this location. Any sub-
els of different products in different periods as deci- set of the delivery locations can be on the route of any
sion variables, we say that our formulation translates truck. Hence, we must assure that the total load on a
customer demand into a target production output truck destined to visit any such subset does not exceed
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
Interfaces 39(5), pp. 460475, 2009 INFORMS 465

z12t = 1 z23t = 1 3
zo1t = 1
z34t = 1

z5ot = 1
z45t = 1

Staging and loading Vehicle routing

Figure 2: The models loading and routing component deals with nding the subsets of delivery locations that will
be visited by the private eet of trucks dispatched from the staging and loading area at the factory warehouse.

the truck volume; the ensuing route-based truck vol- The two sets of dispatch capacity constraints (12)
ume constraints (23) follow. and (13) ensure that these requirements are satised.

Constraints Related to Market Area Cost Function

Deliveries at Tier 2 The cost function has ve specic components that
The nal set of constraints is associated with smaller represent (1) touch costs (i.e., product-based per-unit
trucks (RD trucks) delivering indirect shipments to handling costs at DCs and bins) associated with indi-
customers from the DCs and bins. At this level, we rect customer deliveries via DCs and bins, (2) xed
differentiate between the RD trucks based at the DCs delivery costs to market areas from the DCs and
and bins based on their distinct capacities. The sets bins, (3) holding costs at inventory-keeping locations,
of customers that each DC and bin serves are xed. (4) xed costs of loading and dispatching OTR trucks
Each set represents a market area, which consists of at the factory warehouse, and (5) distance-based rout-
customers in close proximity and assigned to a spe- ing costs of OTR trucks leaving the factory ware-
cic DC or bin. Under the VMI/D program, FLNA house. The rst two terms of the objective function
serves all customers in each market area using pre- take into account the touch costs. A xed delivery
set delivery routes. Hence, our main interest at the cost, which is associated with each dispatch from each
customer level is not load or route optimization; it is DC (bin) to its market area, is included in the third
to assure that the truck volume constraints associated and fourth terms of the objective function. The fth,
with deliveries from the DCs and bins are satised. sixth, and seventh terms correspond to the holding
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
466 Interfaces 39(5), pp. 460475, 2009 INFORMS

costs at the inventory-keeping locations. Next, associ- using the inventory balance constraints (3) and (4).
ated with each OTR truck released from the factory We specify the remaining overall requirements by the
warehouse, we consider a xed loading cost. Finally, shipment quantities of the other customers, as dic-
we keep track of the number of drop-offs at each tated by the demand constraints (2) and the prod-
delivery location of the truck routes via the route- uct requirements of the other plants. We also set the
selection variables in successive periods; thus, we can total load in a similar fashion using the load balance
incorporate the total mileage cost in the last term of constraints (7).
the objective function. Utilizing the weekly DD, replenishment, and ship-
ment quantitiesbased either on the outcome of the
Solution Methodology initialization module discussed above or the solution
of the inventory subproblemin the routing subprob-
As we noted earlier, our solution methodology
lem, we determine the corresponding truck routes
decomposes the overall model into two subproblems
while minimizing the loading and routing costs of
that involve complementary inventory and routing
the OTR trucks based at the factory warehouse. Using
components. We illustrate the relationship between
the solution to the routing subproblem, we calcu-
the subproblems in the owchart of our algorithm in late the route-based setup costs associated with each
the appendix (Figure 8), and we also provide the spe- delivery location of the routing subproblem in each
cic formulations of the subproblems. We solve the period. In the appendix, we describe how we calcu-
subproblems iteratively until we cannot nd a cost- late these costs and consider them as an input while
based improvement for the overall solution or reach solving the inventory subproblem. We now proceed
the limit on maximum number of iterations. with a detailed discussion of the routing and inven-
The inventory subproblem seeks to determine the tory subproblems.
weekly replenishment and shipment quantities at the
DCs, bins, and DD customers, and also considers Routing Subproblem
the requirements at the other plants, the touch costs, The constraints and cost terms related to loading and
inventory holding costs, xed delivery costs to mar- routing operations in the overall model constitute the
ket areas from the DCs and bins, and additional xed routing subproblem. Its main goal is to determine the
costs associated with loading and routing (i.e., the OTR truck delivery routes from the factory warehouse
route-based setup costs), as we describe in detail to the delivery locations given the weekly replen-
below. Given the weekly replenishment and shipment ishment and shipment quantities. A closer examina-
quantities, the routing subproblem species the truck tion of the formulation in the appendix reveals that
routes and minimizes the actual loading and rout- the routing subproblem is separable for each period.
ing costs. Our algorithm begins with an initialization Hence, it sufces to solve a routing problem succes-
module. Its main role is to set the replenishment and sively for each period (week) of the planning horizon.
shipment quantities as input into the routing subprob- We use a savings algorithm, the Clarke and Wright
lem in the rst iteration, as we describe next. algorithm (Clarke and Wright 1964, Chopra and
Meindl 2001) with an additional step for improve-
Initialization ment. When the savings algorithm terminates with a
In the initialization module, in which we address set of routes, we improve each generated route with
each product in each period, we set the shipment more than four delivery locations using a cheapest
quantities for each potential DD customer based on insertion heuristic, a well-known travelling salesman
that customers corresponding demand, to satisfy the problem heuristic.
demand constraints (1). Hence, at this iteration, the Before we solve a routing problem for each period,
DD customers do not receive any shipments from we preprocess the given replenishment and shipment
their respective DCs or bins. We set the replenishment quantities to see if full-truck load (FTL) shipments are
quantities from the factory warehouse to the DCs, needed for a specic location; this might be the case
bins, and other plants based on the remaining over- for major customers, such as Sams Club and Costco,
all requirements of the corresponding locations and and metropolitan area DCs. That is, we rst process
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
Interfaces 39(5), pp. 460475, 2009 INFORMS 467

FTL shipments so that the quantity leftover for each balance, storage capacity, load balance, production
location is less than the truck capacity. We note that capacity, and dispatch capacity from the overall
each FTL shipment constitutes a route with a sin- formulation, and the additional route-based setup
gle destination. We then specify the additional truck constraints. The objective function includes the corre-
routes and consider the leftover quantities, which are sponding route-based setup costs and all cost terms
called less-than-truck load (LTL) shipments (each on of the overall model, except the loading- and routing-
a route with others). related cost parameters considered in the routing sub-
problem. We solve the inventory subproblem using
Inventory Subproblem CPLEX 9.0 (CPLEX is a trademark of ILOG, Inc.).
Traditional inventory models evaluate the trade-offs
between xed ordering and inventory holding costs to
determine replenishment quantities. In our inventory
Application Modules
subproblem, because of the inuence of the routing Our application includes the database application and
subproblem, the trade-off also includes touch costs at solution algorithms, and provides broad functional-
the bins and DCs, xed delivery costs to the mar- ity for optimization, strategy evaluation, and analy-
ket areas, and route-based setup costs dictated by the sis of output les. The database application is useful
loading and routing costs of the OTR trucks based at for importing input data and processing the data
the factory warehouse. We discussed these cost com- to eliminate discrepancies in format and range. The
ponents above, except for the route-based setup costs, application also includes (1) a graphical user inter-
which are an input from the routing subproblem. face, (2) application programming interfaces (APIs)
In each period, depending on its replenishment and for constructing a network to facilitate the genera-
shipment quantity, a particular delivery location can tion of interdistances between the delivery locations
be visited by a truck(s) on a route(s). By sending a based on the address information, and (3) modules
truck on a particular route, we incur a route-based, for performing sensitivity analysis and report gener-
xed-loading total-mileage cost. Each delivery loca- ation (Figure 3). We designed the user interface and
tion on the route is responsible for the corresponding the normalized database tables using Visual Basic for
cost. We allocate this cost to the delivery locations on Applications and MS Access 2003 and implement our
the route when calculating the corresponding route- solution using the C++ programming language, Con-
based setup costs, as we describe in the appendix. cert Technology, and CPLEX 9.0. We perform the runs
Our proposed solution for the routing subproblem is on a computer with a 3 GHz Intel XEON processor
such that the set of routes of a given delivery loca- and 6 GB RAM.
tion, l, includes at most one route in which l shares a The applications information ow begins with
truck with other delivery locations, and possibly other importing data using an interface that has a step-
routes carrying FTL shipments to l (if the require- by-step procedure for performing the required data-
ments at l in a given period exceed the truck capac- import and data-integrity verication. The graphical
ity). To ensure that the right delivery locations incur user interface (Figure 4) allows the user to easily
the route-based setup costs in a given period, we import, effectively manage, and efciently query data
use binary variables. The impact of route-based setup pertaining to the logistics network, capacity, demand,
costs are reected by the last four terms of the objec- and cost. The interface is also useful for selecting a
tive function and the route-based setup constraints set of potential DD customers based on a variety of
of the inventory subproblem. These costs and con- user-specied criteria.
straints establish the link between the inventory and The interdistance data are generated through the
routing subproblems. API for input data, such as network-construction
Except for the addition of the route-based setup data, and are passed to the solution algorithms. The
costs and constraints, the remaining formulation API for solution algorithms provides formatted input,
of the inventory problem is based on our overall executes the algorithms, and reads the output solution
model. That is, the inventory subproblem includes into the database tables for later analysis and report
the constraints associated with the demand, inventory generation.
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
468 Interfaces 39(5), pp. 460475, 2009 INFORMS

Figure 3: The graphic shows a snapshot of our applications main menu.

Finally, the user can use the analysis module to The chase policy matches the demand by sending
change the network, capacity, demand, and cost data, the exact weekly requirements (including the demand
and to perform scenario analyses to evaluate new and safety stock) to the DCs and bins. Using this
strategies for incorporating DD customers for other policy, we obtain a feasible solution to the inventory
plants. The application requires little user training; subproblem by satisfying the replenishment and ship-
in addition, it provides effective solutions in rea- ment requirements, and then solve the routing sub-
sonable time and facilitates technology transfer for problem. Hence, it provides a basis for measuring the
easy implementation at FLNA plants. FLNA exec- relative gains associated with inventory optimization
utives have used it for distribution policy, strategy and for considering DD customers. Using the cube-
evaluation, and scenario analyses. Their main objec- out policy, FTL shipments, which include the demand
tives have been to assess potential cost savings that and some excess supply for the most popular prod-
would result from (1) direct deliveries for varying uct, potato chips, are sent to the DCs and bins to
demand forecasts and (2) inventory optimization and achieve FTL shipments. This benchmark, which aims
reduction at the factory warehouse. A plant in Perry, to increase truck utilization and routing efciency
Georgia has implemented the application for these without explicitly addressing the role of inventory
purposes. optimization, represents the usual company practice.
We consider 5 product groups that are produced
Benchmark Models and at Irving, 8 FLNA plants located elsewhere to which
Benets Realized interplant shipments are sent, 8 DCs, 74 bins, 8,460
We present our numerical results to demonstrate customers with 85 potential DD customers, and
our approachs effectiveness and benets. First, we 12 weeks of demand data. We observe that our solu-
compare our results to two benchmarks, the chase and tion for the optimization and benchmark models runs
cube-out policies. Neither considers DD customers. in a very reasonable time using real data. We obtain
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
Interfaces 39(5), pp. 460475, 2009 INFORMS 469

Figure 4: The graphic shows a snapshot of the graphical user interface for network data of the factory warehouse.

the chase and cube-out policy solutions in a few Compared to the two benchmark policies, the inven-
minutes and obtain a solution for our model within tory holding costs at the factory warehouse in our
10 minutes. We also obtained similar run-time results model are lower by as much as 76.92 percent (chase
for a relatively larger problem at another FLNA plant policy) and 73.62 percent (cube-out policy). The inven-
in Perry, Georgia. tory holding costs at the DCs are lower by 24.52 per-
In Table 1, we present the overall cost and its main cent and 25.59 percent compared to the chase policy
components for each model. In the last row of this
table, we report the percentage difference between the Our model Chase policy Cube-out
cost using our model and the benchmark models. In ($) ($) policy ($)
the following discussion, we elaborate on the detailed
benets of our model relative to the two benchmark Touch cost 675 864 864
Delivery cost (from DCs and bins) 3377 4517 4517
models and the individual strengths of each model. Inventory holding cost 283 408 434
Loading cost (at the factory 2863 2830 2572
Detailed Analysis of Cost Components warehouse)
Mileage cost (from the factory 2802 2501 2251
In Table 2, we show the cost-component results for warehouse)
our model and for each policy. Relative to the chase Total cost 100 11120 10638
policy and cube-out policy, the inventory levels in our Our model is better by 1120 638
model at all stock-keeping locations are signicantly
Table 1: Examining the costs using our model and each policy indicates
lower because of inventory optimization; the factory that our model outperforms both policies, and the cube-out policy outper-
warehouse achieves the greatest inventory reduction. forms the chase policy.
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
470 Interfaces 39(5), pp. 460475, 2009 INFORMS

Our model Chase policy Cube-out policy One of the most striking results from the data in
($) ($) ($) Table 1 is that, in each model, transportation-related
expenses (i.e., delivery costs associated with ship-
Inventory holding ments from DCs and bins and loading and total
Overall 283 408 434
mileage costs associated with shipments from the fac-
Factory 024 104 091
DCs 157 208 211 tory warehouse) represent approximately 90 percent
Bins 102 097 131 of the cost. In our model, the loading costs at the
Touch factory warehouse are higher by 1.16 percent and
Overall 675 864 864 11.31 percent compared to the chase and cube-out
DCs 450 619 619
Bins 225 246 246 policies, respectively. This is a consequence of the
Deliveries from DCs and bins
increased number of delivery locations served from
Overall 3377 4517 4517 the factory warehouse (because of including DD cus-
DC 2976 4079 4079 tomers). For the same reason, the total mileage cost
Bin 401 437 437
from the factory warehouse is higher by 12.03 percent
Table 2: A detailed cost breakdown of the inventory subproblem for our and 24.48 percent compared to the chase and cube-
model and each policy is helpful in evaluating the impact of positioning out policies, respectively. However, in terms of overall
inventory at the DCs, bins, or factory warehouse. cost, our model is better than the chase and cube-out
policies by $11.20 and $6.38, respectively, because our
and cube-out policy, respectively. The inventory hold- explicit consideration of potential DD customers and
ing costs at the bins decrease by 22.13 percent com- inventory optimization helps to signicantly reduce
pared to the cube-out policy, whereas they increase by the touch costs, inventory holding costs, and delivery
only 4.90 percent compared to the chase policy. costs from DCs and bins.
In terms of touch costs and delivery costs from As we note above, the potential DD customers
the DC and bins, the chase policy and cube-out pol- include 85 locations identied based on demand vol-
icy perform identically; our model outperforms both ume. However, each potential DD customer does not
because we explicitly consider DD customers. Our receive its supplies directly from the factory ware-
model reduces the touch costs as much as 27.30 per- house in each period. The selected number of DD cus-
cent and 8.54 percent from the DCs and bins, respec- tomers varies between 66 and 74, as Figure 5 shows.
tively. Similarly, it reduces the total delivery costs For the specic planning horizon considered, in each
from the DCs and bins by 27.04 percent and 8.54 per-
cent, respectively. We note that higher cost reductions
Potential DD customers Selected DD customers
in the DC-related touch and transportation costs are DD customers receiving FTL shipments
because of DD customer assignments. In general, DCs 90
are responsible for serving large stores that are poten- 80
tial DD customers. Using the chase and cube-out poli-
Number of customers

cies, these large customers are usually served from 60
DCs, thus incurring higher touch and transportation 50
costs. Our model allows such customers to be served 40
directly from the factory warehouse. Next, we pro- 30
ceed with a detailed discussion of additional benets 20
associated with explicitly considering DD customers. 10
Measuring the Value of Direct Deliveries 1 2 3 4 5 6 7 8 9 10 11 12
A closer examination of the results in Tables 1 and 2
is useful for illustrating both the savings resulting
Figure 5: When we consider a total of 85 potential DD customers, both
from optimization and the explicit consideration of the number of selected DD customers and those receiving FTL shipments
DD customers. from the factory warehouse are signicant.
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
Interfaces 39(5), pp. 460475, 2009 INFORMS 471

70.0 90.0
LTL % FTL % 80.0
50.0 70.0
40.0 60.0

20.0 BM1
10.0 BM2
20.0 Our model
0.0 10.0
1 2 3 4 5 6 7 8 9 10 11 12
1 2 3 4 5 6 7 8 9 10 11 12

Figure 6: This illustration of the FTL and LTL shipments from the fac-
Figure 7: Using our model, average truck utilization is comparable to that
tory warehouse to the selected DD customers, as a percentage of their
of the chase policy.
total demand, demonstrates that some customers still receive shipments,
albeit in small amounts, via their respective DCs and bins.

model was whether it could achieve high truck uti-

week, most of the selected DD customers (varying lization. In Figure 7, we address this issue by showing
the truck utilization percentage for each model.
between 66.2 percent in week 2 and 98.6 percent in
As expected, the cube-out policy provides the high-
week 7) receive FTL shipments from the factory ware-
est truck utilization; it ranges between 81.8 and
house. Shipping these large quantities directly, rather
88.3 percent during the planning horizon; however,
than via DCs and (or) bins, helps to save on touch
our model is comparable to the chase policy and per-
costs and inventory holding costs and also on delivery
forms well relative to company expectations. Based
costs from DCs and bins. For example, in week 7, 73
on overall benets, FLNA evaluates the model as
of 74 selected DD customers receive FTL shipments.
providing signicant value to the company. In addi-
The touch, inventory holding, and delivery costs from
tion to its cost advantage (Table 1), the model allows
DCs and bins decrease by sending smaller quantities
FLNA to directly serve DD customers from the fac-
to the DCs and bins.
tory warehouse with either FTL or LTL deliveries.
We note that when a customer is identied as a DD
Neither policy can provide this benet because nei-
customer, that customer has the opportunity to receive
ther can accommodate a route that includes DD cus-
shipments directly from the factory warehouse; how-
tomer locations.
ever, this opportunity does not forbid shipments from
the DCs and bins. In Figure 6, we illustrate the ship-
ments to DD customers from the factory warehouse. Conclusions
Adding the FTL and LTL percentages sent from the Our research resulted in quantiable benets for
factory warehouse, we observe that 79.8 to 89.1 per- FLNA. Both the proposed model and the solution
cent of the DD customer demand is shipped directly approach are innovative because, despite the recent
from the factory warehouse. In Figure 6, the percent- interest of academia and industry in integrating logis-
age of FTL shipments varies between 48.4 and 61.3 tical decisions, no published results exist that address
percent in different periods, and the percentage of LTL the specic problem we discuss in this paper.
shipments varies between 27.8 and 35.5 percent. A major goal of our research was inventory elimi-
nation at the Irving plant. To this end, FLNA had been
Trade-Off Between Truck Utilization investigating the use of a new packing and shipping
and Inventory Optimization technology that would make direct supply from the
We next discuss the results of the trade-off between factory warehouse to DD customers a viable option.
truck utilization, which FLNA considers a success cri- Our model provided input into FLNAs decision mak-
terion for efcient routing, and inventory optimiza- ing by evaluating the impact of DD customers.
tion. As we mentioned above, the company uses this A careful investigation of the models results in
principle for designing the routes by following the comparison to the chase and cube-out policies showed
cube-out policy. One of its main concerns about our that DD options, in addition to making efcient
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
472 Interfaces 39(5), pp. 460475, 2009 INFORMS

routing and inventory decisions, eliminated inventory Parameters Related to Production, Inventory, and
at the factory warehouse. FLNA could also realize sig- Transportation Capacities
nicant cost savings in comparison to the benchmark dccapd : storage capacity at DC d.
bincapd : storage capacity at bin b.
models, which did not allow direct deliveries. Further-
ocap: storage capacity at o.
more, we were able to evaluate the criteria for specify- pcapjt : production capacity (in cases) for product j in
ing potential DD customers, the impact of positioning period t.
inventory at the DCs and bins versus the factory ware- pj : production time coefcient of each case of prod-
house, and the trade-offs between truck utilization uct j.
and inventory optimization. An additional goal was apcapt : aggregate production capacity in t (in hours or
to evaluate the cube-out policy as implemented for
tcap: volume capacity of trucks based at o.
outbound shipments from plants to DCs. Our results tcapd : volume capacity of trucks based at DC d.
revealed that DDs from the Irving plant, when sup- tcapb : volume capacity of trucks based at bin b.
ported by efcient routing and inventory replenish-
ment decisions, presented a signicant opportunity for Cost Parameters
cost savings compared to an estimate for the exist- sdj : per-unit touch cost of product j at DC d.
sdj : per-unit touch cost of product j at bin b.
ing system. When we implemented our model and
ad : xed delivery/dispatch cost of each truck leaving
the two benchmark models for the FLNA plant in DC d.
Perry, Georgia, we saw similar results. The study ab : xed delivery/dispatch cost of each truck leaving
and results received enthusiastic support from FLNA, bin b.
and our collaboration continued with a closer exam- hdj : per-unit-per-period inventory holding cost at DC d.
ination of the implications of operational (i.e., daily hbj : per-unit-per-period inventory holding cost at bin b.
hoj : per-unit-per-period inventory holding cost at o.
delivery) requirements of the selected DD customers. A: xed loading cost of each truck leaving o.
Additional research, which we will document in a gmn : mileage cost between m and n for m n V .
future publication, is in progress.
Decision Variables
Sdijt : shipment quantity from DC d to customer i IDd of
Notation product j in period t.
We use the following notation in our model formulation. Sbijt : shipment quantity from bin b to customer i IBb of
product j in period t.
Sets and Indices Soijt : DD quantity (from o) to customer i Io of product j
t: time index, t = 1 2     T .
in period t.
J: set of product groups, j J .
Idjt : beginning inventory of product j in period t at DC d.
D: set of DCs, d D.
Ibjt : beginning inventory of product j in period t at bin b.
B: set of bins, b B.
Rdjt : replenishment quantity of product j in period t sent
P: set of other plants receiving interplant ship-
to DC d from o.
ments, p P .
Rbjt : replenishment quantity of product j in period t sent
I: set of customers, i I.
to bin b from o.
Di : set of DCs that serve customer i; i.e., Di  D.
Rojt : total load of product j destined to the DCs and bins
Bi : set of bins that serve customer i; i.e., Bi  B.
from o.
o: factory warehouse.
Iojt : beginning inventory of product j in period t at o.
Io : set of potential DD customers; i.e., Io  I.
IDd : set of customers served by DC d; i.e., IDd  I. Pjt : target replenishment quantity of product j in period t
IBb : set of customers served by bin b; i.e., IBb  I. sent to o.
V: o D B P Io . Ct : number of dispatches in period t from o.
E: links connecting the locations in V . ydt : number of drop-offs at DC d in period t.
GV  E: road network that consists of vertices (locations) ybt : number of drop-offs at bin b in period t.
V and edges (links) E. ypt : number of drop-offs at plant p in period t.
yit : number of drop-offs at customer i Io in period t by
Demand Parameters trucks originating at o.
Dijt : demand (in cubic cartons) at customer i for product zmnt : number of vehicles traveling from location m V to
j in period t. location n V .
Rpjt : product requirement (shipment quantity) of plant p xdt : number of trucks leaving DC d in period t.
for product j in period t sent from o. xbt : number of trucks leaving bin b in period t.
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
Interfaces 39(5), pp. 460475, 2009 INFORMS 473

Additional Notation Related to Subproblems Ibjt bincapb b B t
Kit : route-based setup cost at DD customer i Io in jJ

period t. (Storage cap. at bins), (6)

Kdt : route-based setup cost at DC d in period t.    
Kbt : route-based setup cost at bin b in period t. Soijt + Rdjt + Rbjt + Rpjt = Rojt
iIo dD bB pP
Kpt : route-based setup cost at plant p in period t.
wit : binary indicator of route-based setup at DD customer j J t (Load balance), (7)
i Io in period t.
wdt : binary indicator of route-based setup at DC d in Rojt = Iojt + Pjt Iojt+1 j J t
period t. (Inv. bal. at o), (8)
wbt : binary indicator of route-based setup at bin b in 
period t. Iojt ocap t (Storage cap. at o), (9)
wpt : binary indicator of route-based setup at plant p at jJ
time t. 
pj Pjt apcapt t
M: a large number. jJ

Model (Aggregate prodn. cap.), (10)

We formulate the overall problem as a mixed-integer pro-
Pjt pcapjt j J t
gram as follows.
(Product-based prodn. cap.), (11)

min sj Sdijt + sj Sbijt iIDd jJ Sdijt
t=1 jJ dD iIDd t=1 jJ bB iIBb xdt d D t
(Touch costs)
(Dispatch cap. for DCs), (12)

+ ad xdt + ab xbt iIBb jJ Sbijt
xbt b B t
t=1 dD t=1 bB tcapb
(Delivery and dispatch costs) (Dispatch cap. for bins), (13)

jJ Rojt
+ hdj Idjt + hbj Ibjt + hoj Iojt Ct t
t=1 jJ dD t=1 jJ bB t=1 jJ tcap
(Inventory costs) (Total load and truck volume), (14)


d D t
+A Ct + gmn zmnt tcap
t=1 t=1 nV mV
(Drop-offs at DCs), (15)
(Loading and routing costs) 
  jJ Rbjt
subject to Dijt = Sdijt + Sbijt +Soijt i Io  j J  t ybt b B t
dDi bBi

(Demand), (1) (Drop-offs at bins), (16)

Dijt = Sdijt + Sbijt i I\Io  j J  t ypt
p P t
dDi bBi tcap
(Demand), (2) (Drop-offs at other plants), (17)
Sdijt = Idjt + Rdjt Idjt+1 d D j J  t jJ Soijt
yit i Io  t
iIDd tcap
(Inv. bal. at DCs), (3) (Drop-offs at DD customers), (18)
Sbijt = Ibjt + Rbjt Ibjt+1 b B j J  t zmnt = ynt n V \o t
iIBb mV

(Inv. bal. at bins), (4) (Arrivals to Tier 2), (19)

Idjt dccapd d D t zmnt = ymt m V \o t
jJ nV

(Storage cap. at DCs), (5) (Departures from Tier 2), (20)

etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
474 Interfaces 39(5), pp. 460475, 2009 INFORMS

zmot = Ct t (Arrivals to o), (21) period t. Also, let rs  denote the number of delivery loca-
mV tions on route  RouteSetlt . Then, the route-based setup
 cost of delivery location l in period t is
zont = Ct t (Departures from o), (22)
   A + n m gmn zmnt
 mS jJ Rmjt Klt = 
zmnt S V \o t RouteSet
mS nS
We explicitly consider it in the subsequent inventory
(Route-based truck volume). (23)

Nonnegative Integer Variables Inventory Subproblem

Input parameters include the route-based setup costs
Sdijt d D i I j J  t denoted by Kit , Kdt , Kbt , and Kpt for i Io , d D, b B, p P ,
Sbijt b B i I j J  t and t = 1     T . We consider these costs and inventory-
related cost parameters sj , hdj , hbj , and hoj , loading-related
Soijt i Io  j J  t cost and capacity parameters ad , ab , tcapd , and tcapb , cus-
tomer and interplant demand parameters Dijt and Rpjt , and
Rdjt  Idjt  Rbjt  Ibjt d D b B j J  t
initial and end-of-horizon inventory levels at the inventory-
Rojt  Iojt  Pjt j J  t keeping locations Idj0 , Ibj0 , Ioj0 , IdjT +1 , IbjT +1 , and IojT +1 . The
output includes (1) all inventory-related decision variables
xdt d D t
Sdijt , Sbijt , Soijt , Idjt , Ibjt , Iojt , Rdjt , Rbjt , and Rojt ; (2) the number
xbt b B t of trucks required at DCs, xdt , and bins, xbt , for customer
deliveries; and (3) binary route-based setup indicators wit ,
Ct t wdt , wbt , and wpt .
ydt d D t


ybt b B t min sj Sdijt + sj Sbijt

t=1 jJ dD iIDd t=1 jJ bB iIBb
yit i Io  t


zmnt m V  n V  t + ad xdt + ab xbt + hdj Idjt

t=1 dD t=1 bB t=1 jJ dD

Obtaining an exact optimal solution is impractical for the 


overall model because of its size and complexity. We over- + hbj Ibjt + hoj Iojt + Kit wit
t=1 jJ bB t=1 jJ t=1 iIo
come this computational difculty by developing a heuristic
that builds on the decomposition of the above model into 

two subproblems involving inventory and routing compo- + Kdt wdt + Kbt wbt + Kpt wpt
t=1 dD t=1 bB t=1 pP
nents. We rst describe these subproblems and then present
a owchart of our iterative heuristic. subject to constraints (1)(13) and

jJ Soijt
Routing Subproblem wit i Io  t
Input parameters are the inventory-related decision vari-
ables Soijt , Rdjt , Rbjt , and Rojt and input parameters A, gmn , (Route-based setup at DD customers), (24)

Rpjt , and tcap. The output includes (1) routing-related deci- jJ Rdjt
sion variables Ct , ydt , ybt , ypt , yit , and zmnt , and (2) setup wdt d D t
costs Kdt , Kbt , Kpt , and Kit .
(Route-based setup at DCs), (25)

T   jJ Rbjt
min A Ct + gmn zmnt wbt b B t
t=1 t=1 nV mV
(Route-based setup at bins), (26)
subject to constraints (14)(23) with variables Ct , ydt , ybt , 
yit , and zmnt as dened above. jJ Rpjt
wpt p P  t
When we obtain a solution, we calculate the route-based (Route-based setup at plants), (27)
setup costs of the delivery locations on the resulting routes.
For a given delivery location l V \o, let RouteSetlt denote with variables Sdijt , Sbijt , Soijt , Rdjt , Rojt , xdt , and xbt as dened
the set of routes to which delivery location l belongs in above and binary integer variables wnt n V \o t.
etinkaya et al.: An Integrated Outbound Logistics Model for Frito-Lay
Interfaces 39(5), pp. 460475, 2009 INFORMS 475

Campbell, A. M., M. W. P. Savelsbergh. 2004. A decomposition

Initialization approach for the inventory-routing problem. Transportation Sci.
38(4) 488502.
Sojt, Rdjt, Rbjt, Rpjt, Rojt etinkaya, S. 2004. Coordination of inventory and shipment con-
solidation decisions: A review of premises, models, and
Sojt, Rdjt, Rbjt, Rpjt, Rojt
justication. J. Geunes, E. Akcali, P. M. Pardalos, Z.-J. Shen,
Routing subproblem
eds. Applications of Supply Chain Management and E-Commerce
Research in Industry. Kluwer, New York, 351.
Route-based setup costs: Kit , Kdt , Kbt, Kpt
etinkaya, S., C.-Y. Lee. 2000. Stock replenishment and shipment
scheduling for vendor-managed inventory systems. Manage-
Inventory subproblem ment Sci. 46(2) 217232.
Chan, L. M. A., A. Muriel, Z.-J. M. Shen, D. Simchi-Levi, C.-P.
costNew = cost(INV) + cost(VRP) Teo. 2002. Effectivene zero-inventory-ordering polices for the
cost = costNew
single-warehouse multiretailer problem with piecewise linear
iterNo = iterNo + 1
cost structures. Management Sci. 48(11) 14461460.
cost costNew Chopra, S., P. Meindl. 2001. Supply Chain Management: Strategy,
No > 0.01 Yes
costNew Stop Planning, and Operation. Prentice Hall, Upper Saddle River, NJ.
iterNo > MaxIterNo Clarke, G., J. W. Wright. 1964. Scheduling of vehicles from a central
depot to a number of delivery points. Oper. Res. 12(4) 568581.
Diaby, M., A. Martel. 1993. Dynamic lot-sizing for multi-echelon
distribution systems with purchasing and transportation price
Figure 8: The owchart shows the solution algorithm in which cost(VRP) discounts. Oper. Res. 41(1) 4859.
and cost(INV), respectively, represent the corresponding objective func- Federgruen, A., D. Simchi-Levi. 1995. Analytical analysis of vehicle
tion values of these two subproblems. routing and inventory routing problems. M. O. Ball, T. L. Mag-
nanti, C. L. Monma, G. L. Nemhauser, eds. Network Routing.
Elsevier, Amsterdam, 297373.
Solution Algorithm Fumero, F., C. Vercellis. 1999. Synchronized development of pro-
The algorithm involves iteratively solving the routing and duction, inventory, and distribution schedules. Transportation
inventory subproblems described above, which we outline Sci. 33(3) 330340.
in Figure 8. Hadley, G., T. M. Whitin. 1963. Analysis of Inventory Systems. Pren-
tice Hall, Englewood Cliffs, NJ.
Keskin, B. B., H. ster. 2007. Meta-heuristic approaches with mem-
Acknowledgments ory and evolution for a multi-product production/distribution
This research was supported by Frito-Lay and PepsiCo system design problem. Eur. J. Oper. Res. 182(2) 663682.
under contract C05-00582 and Texas Engineering Exper- Kleywegt, A. J., V. S. Nori, M. W. P. Savelsbergh. 2002. The stochas-
iment Station Proposal 05-1073, during August 2005 to tic inventory routing problem with direct deliveries. Transporta-
February 2006. We particularly thank Greg Bellon, Robert tion Sci. 36(1) 94118.
Lee, C.-Y., S. etinkaya, W. Jaruphongsa. 2003. A dynamic model
W. Mackley, and Larry Sweet of Frito-Lay for their support for inventory lot-sizing and outbound shipment scheduling at
and input throughout the project. a third party warehouse. Oper. Res. 51(5) 735747.
Lei, L., S. Liu, A. Ruszczynski, S. Park. 2006. On the integrated
production, inventory, and distribution routing problem. IIE
References Trans. 38(11) 955970.
Alp, O., N. K. Erkip, R. Gll. 2003. Optimal lot sizing/vehicle Lippman, S. A. 1969. Optimal inventory policy with multiple set-up
dispatching policies under stochastic lead times and stepwise costs. Management Sci. 16(1) 118138.
xed costs. Oper. Res. 51(1) 160171. Sarmiento, A. M., R. Nagi. 1999. A review of integrated analysis of
Anily, S., A. Federgruen. 1993. Two-echelon distribution systems production-distribution systems. IIE Trans. 31(11) 10611074.
with vehicle routing costs and central inventories. Oper. Res. Thomas, D. J., P. M. Grifn. 1996. Coordinated supply chain man-
41(1) 3747. agement. Eur. J. Oper. Res. 94(1) 115.
Archetti, C., L. Bertazzi, G. Laporte, M. Speranza. 2007. A branch- Toptal, A., S. etinkaya. 2006. Contractual agreements and vendor-
and-cut algorithm for a vendor-managed inventory-routing managed delivery under explicit transportation considerations.
problem. Transportation Sci. 41(3) 382391. Naval Res. Logist. 53(5) 397417.
Baita, F., W. Ukovich, R. Pesenti, D. Favaretto. 1998. Dynamic ster, H., B. B. Keskin, S. etinkaya. 2008. Integrated warehouse
routing-and-inventory problems: A review. Transportation Res. location and inventory decisions in a three-tier distribution sys-
Part A 32(8) 585598. tem. IIE Trans. 40(8) 718732.
Bertazzi, L. 2008. Analysis of direct shipping policies in an Van Eijs, M. J. G. 1994. Multi-item inventory systems with joint
inventory-routing problem with discrete shipping times. Man- ordering and transportation decisions. Internat. J. Production
agement Sci. 54(4) 748762. Econom. 35(13) 285292.
Bertazzi, L., G. Paletta, M. G. Speranza. 2005. Minimizing the Viswanathan, S., K. Mathur. 1997. Integrating routing and inventory
total cost in an integrated vendor-managed inventory system. decisions in one-warehouse multiretailer multiproduct distri-
J. Heuristics 11(56) 393419. bution systems. Management Sci. 43(3) 294312.