Electronic Markets for Truckload Transportation

Transcription

1 PRODUCTION AND OPERATIONS MANAGEMENT Vol. 16, No. 4, July-August 2007, pp issn $1.25 POMS 2007 Prodution and Operations Management Soiety Eletroni Marets for Truload Transportation Chris Caplie MIT E40 365, 1 Amherst Street, Cambridge, Massahusetts 02139, USA aplie@mit.edu A wide variety of eletroni maretplae formats are used in the Truload (TL) transportation industry, inluding ombinatorial autions, private and publi exhanges, and eletroni atalogs. Combinatorial multi-attribute autions are ommonly used strategially to populate eletroni atalogs, ommonly alled routing guides, with priing, assignments, and priority logi. Private and publi exhanges are used to omplement the eletroni atalogs in ases where the atalog fails. This paper disusses the TL transportation maret, plaes the prourement of servies in the ontext of eletroni maretplae formats, and illustrates how these are urrently used. Key words: freight transportation; autions; eletroni marets; eletroni atalogs; transportation management Submissions and Aeptane: Reeived Marh 2006; revision reeived June 2006; aepted July Introdution This paper provides an overview of the use of eletroni marets for the prourement of truload (TL) transportation servies by the buyers of these servies (shippers). Truload transportation firms (arriers) generally handle shipments that are pied up at a loation and driven diretly to a single destination with no intermediate stops. This is ontrasted with less-than-truload (LTL) or parel shipments where the individual shipment might be pied up and transported to an initial sorting hub on one vehile and reloaded onto a separate vehile for movement to another terminal before finally being loaded onto another vehile for final delivery. Currently, the use of a variety of different eletroni maret formats is widespread for TL servies. The names used to desribe these mehanisms, however, do not always math those used in the general eletroni ommere literature. The three most ommon forms of eletroni ommere trading models mentioned in the literature are autions, atalogs, and exhanges (see, for example, Kaplan and Sawhney 2000 or Sulley and Woods 1999). So, although reverse autions are the predominate mehanism used for strategi prourement of TL servies, they are ommonly referred to as either bids or request for proposals by shippers and arriers alie. Similarly, although all large shippers manage their TL operations through the use of eletroni atalogs, they all the systems routing guides or transportation management systems (TMS). Finally, although some form of publi or private exhange is used by most shippers for at least a portion of their TL business, the mehanism is referred to as the spot maret. In fat, the words aution and exhange have negative onnotations for both shippers and arriers alie in the TL industry. So, although the underlying mehanis of eletroni marets are ommonplae in TL prourement, the formal names are rarely mentioned. There are several reasons why it is worthwhile to study the use of eletroni marets for TL transportation servies. First, the U.S. TL maret is very large. In 2004, the domesti U.S. freight transportation maret was $770 billion and aounted for 6.6% of the GDP (Standard & Poor s 2006). Truing omprised $671 billion or 87% of the total maret with the TL setor alone aounting for $312 billion. Seond, the TL maret is very fragmented. As of 2002, there were over 45,000 TL arriers registered in the United States with 75% of them owning fewer than six power units (Amerian Truing Assoiations 2002). In 2003, there were only 500 TL firms with 423

2 Chris Caplie: Eletroni Marets for Truload Transportation 424 Prodution and Operations Management 16(4), pp , 2007 Prodution and Operations Management Soiety annual revenues in exess of $10 million and just 1,000 with revenues in exess of $2 million. The seven largest publily traded TL firms ombined represent less than 4% of the total TL maret (Standard & Poor s 2006). Third, in addition to being a large omponent of the overall eonomy, transportation is the largest logistis ost for most shippers, averaging over 60% of a firm s total logistis osts (Wilson 2005). Transportation has been estimated to average 3 to 4% of annual sales revenue for most firms (Ballou 2004; Enslow 2004). TL truing is typially the largest omponent of a shipper s transportation budget (Wilson 2005). With the shift to more international inbound shipments from ports to distribution enters, the importane of longer haul TL servies is only inreasing. Fourth, shippers use a variety of ontratual arrangements for TL servies, with the three most ommon being private fleets, for-hire ontrat arriers, and spot. Private fleets, valued at $294 billion in 2004, are omposed of truing operations run by firms whose primary business is not transportation. Contrat relationships onsist of a shipper and arrier who have entered into an annual (or longer) ontrat that stipulate rates for speifi lanes (origin destination pairings, e.g., Chiago, IL, to Atlanta, GA). This is the most ommon form of relationship for for-hire arriers. In a spot maret the arrier is seleted on a load-by-load basis and the prie is determined at the time of the physial shipment, rather than through a standing ontrat. It is diffiult to estimate the exat size of the ontrat and spot TL marets beause these statistis are not reported diretly to government agenies or analyst firms. The ommon approximation in pratie is that ontrat relationships over 90 to 95% of the $312 billion for-hire TL maret, whereas spot onstitutes the remainder. Shippers will typially use a portfolio of all three types of relationships, to some degree, aross their freight transportation networ. In fat, firms with large private fleets, suh as Wal-Mart Stores and Syso Foods, are typially also large buyers of ontrated and spot TL servies. Beause the TL maret is large, fragmented, important to buyers, and proured under many different relationships, it has been subjeted to an exeptionally large number of first-generation prourement mehanisms over the past 25 years sine it was deregulated in the United States. In 2000, for example, there were more than 50 Internet-based transportation exhanges in operation (Fontanella 2000). Very few of these were in existene just 5 years later. Optimization-based autions for TL servies were introdued in the late 1980s (Moore, Warme, and Gorban 1991), with more omplex ombinatorial autions in use as early as 1992 (Porter et al. 2002; Ledyard et al. 2000). The three maretplae forms (autions, atalogs, and exhanges) are intertwined. The objetive of a shipper s strategi aution, for example, is to populate his own private eletroni atalog for use in daily operations. When the atalog fails, that is, when the planned apaity and assignment is not suffiient, then the shipper relies on an exhange (publi, private, or some ombination) to find the needed apaity. Shippers utilize all three forms in the prourement and management of TL servies. The remainder of the paper is organized as follows. Setion 2 provides an overview of TL transportation operations and networs. Setion 3 disusses the use of autions for the strategi prourement of TL servies. Speifially, reverse ombinatorial autions will be disussed. Setion 4 presents the operational side of TL servies and shows how shippers use eletroni atalogs and exhanges. Additionally, the automation of these exhange forms is illustrated. Finally, Setion 6 summarizes the paper and reommends areas for future researh. 2. Overview of Truload Transportation The physial proess for truload operations is straightforward and involves three parties: a shipper, a arrier, and a reeiver. The shipper is a retailer, manufaturer, distributor, or some other firm that needs to move goods. The arrier is a truing firm that owns transportation assets and whose primary business is the atual movement of freight. The reeiver is the firm to whih the shipment is being sent. The reeiver an be the same firm as the shipper or a different one (e.g., a shipment to a ustomer) Truload Operations The shipper s proess for managing TL operations is as follows. One it is determined that a load of freight is ready to move, the shipper selets a speifi TL arrier and ommuniates the partiulars of the shipment to the seleted arrier. The arrier will then either aept or rejet the shipment. If it is aepted, the arrier will oordinate with the shipper for sheduling the pi up. If the load is rejeted, the shipper will selet another TL arrier and repeat the proess. One a arrier has aepted the shipment, the arrier arrives at the shipper s loation, loads the trailer, and transports it diretly to the reeiver s loation, where it is unloaded. At that point, the arrier s tru is free to find its next load. In some ases, the shipper will pre-load a trailer and have it ready for the TL arrier to pi up. This is nown as drop and hoo and is used to redue the waiting and loading time for the arrier at the origin or destination. Three things are worth noting in the proess. First, after the arrier delivers its shipment, it must drive to its next ustomer s loation. This is referred to as an empty or deadhead move beause the arrier is not

3 Chris Caplie: Eletroni Marets for Truload Transportation Prodution and Operations Management 16(4), pp , 2007 Prodution and Operations Management Soiety 425 paid by any shipper for driving this distane. Large sophistiated TL arriers an average between 6 and 12% empty miles, whereas private fleets average over 24% 1 (see, for example, Werner Enterprises 2006; USA Tru 2006; Corsi and Stowers 1991). The arrier s searh and movement ost for finding and seuring the follow-on load an be a relatively large omponent of the TL arrier s osts. The need to reloate from the destination of one load to the origin of the subsequent load results in strong ost interdependenies or eonomies of sope for TL arriers. The ost of serving a lane is strongly affeted by the probability of finding a follow-on load out of that destination. The by-produt of one shipment (ending up at a ertain loation) an influene the ost of performing a subsequent shipment from some other loation. Seuring a balaned networ redues the unertainty in onnetion osts and an lower the arrier s overall osts. Thus, a arrier may offer a lower prie for hauling a given number of loads from A to B if it also hauls loads from B to A. Eonomies of sope explain why ombinatorial autions were first introdued in the TL industry. Combinatorial autions allow the bidder (the arrier in this ase) to submit a paage bid on a speified olletion or set of items (lanes). Paage bids, it is thought, allow a arrier to apture the eonomies of sope of multiple lanes within a TL transportation networ. Types of bids used in TL autions are disussed in greater detail in Setion 3.2. Eonomies of sope also explain why some arriers will provide a disount to a shipper if they an be offered a ontinuous move where the inbound delivery of a trailer to a faility is mathed in real time with an outbound shipment for the same onveyane. The seond point to note is that TL arriers will often rejet a load that is tendered to them. This is beause, in many ases, a arrier will not have the exat equipment available in the exat loation at the exat time that the shipper requests it. Cooe (2004) reported 5% of all loads offered to a TL arrier are rejeted by the arrier and almost 10% of the time the equipment is not ready when requested. Harding and Caplie (2005) analyzed a data set of 400,000 shipments valued at $300 million rossing multiple shippers and involving 284 arriers and determined arrier aeptane ratios averaged 74%, with a median of 79%. Although many arriers have stellar aept ratios (maximum was 99.9%), roughly half are below 80%. The high level of rejetion implies that shippers must have alternate arriers in plae, along with appropriate mehanisms to aess them. 1 This high empty mile perentage is usually for shorter haul trips where the vehile goes loaded in one diretion and returns ba to the warehouse empty. The third point to note is that the speifi method by whih a shipper selets and ommuniates to a arrier was not disussed in detail. In fat, this is the major topi of the paper. The proesses for arrier seletion and ommuniation during daily operations are intertwined. They an be as simple as a raned list of arriers on an index ard for seletion and a phone for ommuniation or they an be highly sophistiated software systems, featuring an eletroni atalog with exeption management logi and automati esalation to publi or private exhanges. The more sophistiated arrier seletion and ommuniation systems are quite ommon and are embedded in most top-tier TMS that shippers use to manage their daily TL operations TL Freight Networs Truload transportation networs are best haraterized as having many ars (lanes) and nodes (origins and destinations), with most of the total volume distributed on few lanes, while the majority of the lanes have little volume. For example, a typial onsumer paage goods manufaturer with just under $300 million in annual TL expenditures will have 1,000 2,000 speifi ship-to and ship-from points. The number of lanes within the TL networ depends on the geographi speifiity used to represent the origins and destinations. For example, using 5-digit postal ode regions to define eah loation results in over 5,000 distint lanes, using 3-digit postal ode regions results in 3,000 distint lanes, and using state level regions results in just over 500 lanes. Regardless of the level of loation speifiity, however, the distribution of volume on lanes within a TL networ follows a power law distribution where a very low perentage of traffi lanes arry a very high perentage of the volume. Typially, about 30% of the TL volume will flow on just 1% of the lanes and 80% of the volume will flow on 12 to 14% of the lanes. Also, most TL networs will have 15 to 30% of their lanes arrying only 1 load per year (Caplie 1996; Harding 2005). The ratio of the mean to the median volume of loads per year per lane ranges from 8 to Networ Representation Prior to running a strategi aution, the shipper must determine how to represent his networ to the arriers. A shipper will rarely inlude every 5-digit postal ode to 5-digit postal ode lane in an aution. Instead, he will segment his networ into different types of lanes by hanging the level of geographi speifiity of the origins and loations. Lanes are generally lassified as point to point, point to zone, zone to point, or zone to zone ( Caplie 1996). A point loation is usually a 5-digit postal ode region with anything larger being onsidered a zone. An annual threshold volume is typially used to determine how large a zone to use.

4 Chris Caplie: Eletroni Marets for Truload Transportation 426 Prodution and Operations Management 16(4), pp , 2007 Prodution and Operations Management Soiety The shipper will aution off the higher volume point-to-point, point-to-zone, and zone-to-point lanes with a foreasted volume at an annual, monthly, or weely level. These are sometimes referred to as guaranteed traffi or primary lanes. The small amount of volume not inluded in the primary lanes is typially spread aross many lanes and will, therefore, not be expliitly autioned. Instead, shippers will usually as for ba-up rates from arriers for a mutually exlusive and olletively exhaustive set of zones (typially in a state-to-state matrix) without providing any volume projetions. The ba-up rates are rarely onsidered in any TL aution. Instead, the arriers that win any primary lanes in the aution have their state-tostate matrix rates uploaded into the TMS. When establishing the representation of the networ, the shipper must onsider both effetiveness and overage. The more speifi a lane is, the more effetive it will be in that the arrier s osts will be more aurately refleted. As the size of the origin or destination gets larger, the unertainty of the amount of required deadhead inreases. However, the shipper also must ollet rates on those regions where traffi might materialize over the ourse of a year. The net effet is that a shipper will reeive rates for overlapping lanes. For example, suppose the shipper has a plant in Atlanta, GA that delivers to a number of destinations in the state of Illinois that inlude an average of 10 loads a wee to a major distribution enter in Chiago, IL 60601, an average of 3 loads a wee to smaller ustomers loated around Chiago, all in the 606 postal ode region, and about 1 load a month to oasional sites elsewhere in the state. The shipper ould offer three distint lanes: Point to point: Atlanta, GA to Chiago, IL for 10 loads/wee; Point to zone: Atlanta, GA to IL 606 for 3 loads/wee; Zone to zone: Georgia to Illinois as a ba-up rate (no volumes provided). A arrier ould submit bids on all three of these lanes with very different rates. Interestingly, and ounterintuitively, the rates on the most speifially defined lanes are not always the lowest rates. Sometimes the ba-up rates an be lower than the rates a arrier provides as a primary lane. This is beause the primary lane assignment implies some level of aeptane ratio ommitment. The arrier is expeted to have apaity available for these loads. For ba-up rates, the arrier only must provide a tru if he has one that is onvenient thus, he an harge a lower rate and only honor it when it maes eonomi sense. The networ representation of a shipper s TL freight and the resulting multiple overlapping ontrat rates impat the way shippers utilize the three different forms of eletroni marets: autions, atalogs, and exhanges. The next two setions explore the use of these mehanisms at the strategi and operational levels. 3. Strategi Prourement: Autions 2 Shippers have been using reverse autions to establish long-term TL transportation ontrats sine the U.S. surfae transportation industry was deregulated in the 1980s. Initially, these were rather manual proesses with, perhaps, the use of spreadsheets to sort and ran eah arrier s bids by lane. The first reported use of optimization to determine the winner of a transportation aution (that is, solve the winner determination problem or WDP) an be traed to the Reynolds Metals Company in the late 1980s. Moore, Warme, and Gorban (1991) desribe how Reynolds entralized its transportation management system and how it bid out and assigned lanes of traffi to arriers. They developed a mixed integer program (MIP) model that minimized transportation osts by assigning arriers to speifi shipping loations and traffi, taing into onsideration individual arrier apaity onstraints, equipment ommitments, and other transportation-speifi onerns. Although it allowed for simple bids with volume onstraints, it did not permit paage or ombinatorial bids. Porter et al. (2002) desribed ombinatorial autions (whih they referred to as ombined value autions ) run in 1992 by Sears Logistis Servies, in what was probably the first appliation of paage bidding in the transportation ontext. They reported savings of 6 to 20%. Although the model allowed paage bids, it did not permit the use of any business-speifi side onstraints, as did the model developed by Moore et al. (1991). The use of ombinatorial autions for transportation servies (inorporating both paage bids and business side onstraints) inreased dramatially throughout the 1990s, as desribed by Caplie and Sheffi (2003, 2006) and Elmaghraby and Kesinoa (2003). The first ommerially available software speifially designed for ombinatorial autions for transportation servies, OptiBid, was released in Other software ompanies followed suit and by 2005 approximately half a dozen transportation prourement software paages that inorporate paage bids were available in the maret. Sine 1997, hundreds of ompanies have run optimization-based ombinatorial TL autions using these software tools, inluding The Proter & Gamble Co., Sears Roebu and Co., K-Mart Corp., Wal-Mart Stores, In., Best Buy Co., In., PepsiCo, International Paper, Crown Cor & Seal, The Home Depot In., Bridgestone Corp., Ford Motor Company, Compaq Computer Corp., Staples, In., Limited 2 Most of this setion is adapted from Caplie and Sheffi (2006).

5 Chris Caplie: Eletroni Marets for Truload Transportation Prodution and Operations Management 16(4), pp , 2007 Prodution and Operations Management Soiety 427 Brands, In., Ryder System, In., the Rite Aid Corp, and many others. The remainder of this setion disusses the standard aution proess, the bidding language used in TL autions, and different formulations of the WDP Strategi TL Aution Proess Strategi TL prourement events are usually onduted annually in a reverse aution. The items proured are hauling ommitments on primary lanes. In addition to the rates submitted by the arriers, many non-finanial fators an be onsidered, inluding level of performane, type of arrier, number of arriers, et. Beause of these additional onsiderations, the predominant mehanism used is a single-round, sealed bid ombinatorial aution utilizing optimization to solve the WDP. Although multiple-round ombinatorial autions are exeptionally rare, in any TL aution there is usually a last soft-negotiation round after the WDP is solved. In many ases, inumbent arriers are provided an opportunity to math the urrent winning rates on lanes they might lose. It is important to stress again that the outome of a strategi TL aution is not ompletely binding. As disussed previously, the arrier will not always have equipment available when the shipper requests it and the shipper does not guarantee a minimum volume or dollar amount to the winning arriers. Suppose, for example, a shipper runs an aution and awards the lane from Chiago to Atlanta to a arrier. If, over the ourse of the year, no truload traffi materializes on that lane, then the arrier will reeive no finanial ompensation. The arrier is, however, ontratually bound to honor the rate for a lane he won in an aution if he in fat does haul a load on that lane. Strategi TL autions all follow a standard threestep proess onsisting of pre-aution, aution, and post-aution ativities. During the pre-aution stage, the following tass are ompleted: The shipper foreasts the demand for the upoming period and reates a representative networ of expeted weely (or monthly) flows on eah primary lane. The issues of lane definition and geographi speifiity mentioned in Setion 2.3 apply here. The shipper determines whih arriers to invite to the aution the number of arriers invited to partiipate in TL autions ranges from a dozen to hundreds. An analysis of a sample of 50 TL autions run between 1997 and 2003 showed a orrelation of 0.34 for the number of arriers invited and the value of the business being autioned off. It is rare for inumbent arriers to not be invited to partiipate, so most of the bidders in a TL aution have at least some private information onerning the shipper s business. The shipper determines what information the arrier is required to submit ba. This usually inludes the form of the prie (per move, per mile, per weight, et.), servie details (days of transit, apaity availability, equipment type, et.), and the types of bid allowed (see Setion 3.2). During the aution stage, the following steps are performed: The lanes are ommuniated to the arriers. Although is the most ommon ommuniation tool used for transportation autions, larger shippers using speialty software typially use some sort of Web interfae (Caplie, Plummer. and Sheffi 2004). The arriers analyze the networ, determine the rates to offer, and submit their bid rates. Depending on whether the format of the aution has single (most ommon) or multiple rounds, the arriers may reeive feedba information and have to resubmit updated rates. During the post-bid stage, the following tass are performed: The shipper reeives the arriers bids and solves the WDP. One the shipper solves the WDP, the results are uploaded to the eletroni atalog (routing guide) used in operations. This is disussed in greater detail in Setion 4. The use of ombinatorial autions with side onstraints is widespread for TL servies prourement. Combinatorial autions permit arriers to use sophistiated bidding language to better apture their underlying eonomis and allow shippers to onsider nonfinanial aspets in the arrier seletion deision. The next two setions disuss eah of these points Bidding Language The ommuniation language used during the aution determines how the arriers an respond to the shipper s request for bids. Traditional pratie in transportation is for arriers to submit a per-shipment or permile rate for haulage on eah lane, regardless of the volume of business they might win on that lane or any other lane. These are usually alled simple bids. This form of bid language leads to the arriers hedging their bid pries to over those instanes where they do not win any supporting business. Combinatorial autions allow arriers to mae expliit their otherwise impliit priing assumptions. They an provide a lower bid prie, given ertain other onditions are met. In transportation, these are sometimes referred to as onditional bids. That is, the bid rates submitted are onditional on a pre-defined set of

6 Chris Caplie: Eletroni Marets for Truload Transportation 428 Prodution and Operations Management 16(4), pp , 2007 Prodution and Operations Management Soiety ations also taing plae. Lane-based paage bids are but one type of onditional bids. The different types of onditional bids that are urrently in use within transportation autions are desribed below Simple Lane Bid A bid rate applies to all shipments on that lane, regardless of the volume awarded. The number of shipments awarded to the arrier on that lane is determined by the shipper. Eah bid may inlude speifi servie apabilities (transit time, trailer size, weeend overage, additional safety fators, et.) that are only available if that bid at that rate is awarded. This is the most widely used type of bid. Oftentimes shippers do not even provide arriers with lane volume estimates or foreasts. Carriers an inlude different servie levels in multiple simple bids for the same business to de-ommoditize their offerings Simple Lane Bid with Volume Constraints A bid rate applies to all shipments on a lane but only if the arrier is awarded at or above the minimum ommitment onstraint and at or below the maximum apaity onstraint for that lane, region, set of lanes, or system, as speified. Capaity (upper bound) onstraints are more ommonly submitted by arriers than minimum ommitment onstraints. They are equivalent to budget onstraints in that they allow arriers to bid beyond their available apaity on different lanes. Shippers rarely allow minimum ommitment onstraints as a result of the issues of feasibility in solving the WDP Stati Paage Bids (AND) Stati paage bids are a set of individual lane bid rates that apply to eah lane within that set, onditional on the shipper awarding the arrier all lanes within the set at the exat volume levels speified by the arrier. Most ommerially available software tools handle stati paage bids Stati Either/Or Paage Bids (XOR) Stati either/or paage bids have two or more paage bids with rates that apply onditional on the shipper (1) only awarding the arrier one of the bids and (2) awarding that arrier all lanes within that paage bid. This ommuniates the message, give me this set of lanes, or this set of lanes, but not both. The message, Give me this set of lanes or that set of lanes or both is referred to as an OR bid. It an be ahieved through the use of non-overlapping AND bids Flexible Paage Bids A set of individual lane bid rates applies to eah lane within that set, onditional on the shipper awarding the arrier all lanes within the set within volume ranges speified by the arrier for eah lane within the set. Note that with stati paage bids the shipper does not determine the speifi volume level awarded on eah lane within that paage. The arrier determines the lane volume as part of the submission of the stati paage bid. With flexible paage bids, on the other hand, the shipper selets the speifi volume level awarded on all lanes within the awarded paage bid, as long as it adheres to the arrier s ranges. This means that although a arrier nows the total value of a stati paage bid at the time of bid submission, he only nows the potential range of values for a flexible paage bid at that same time. Only after the WDP is solved will the arrier now the atual number of shipments and total dollar value of a flexible paage bid. The arrier speifies for eah lane within the paage both the rate per load and the minimum and maximum volume per wee, month, or year. Additionally, the arrier an provide paage level apaity ranges. If the shipper is awarding only one arrier per lane, then these bids are equivalent to stati paage bids Simple Reload Bids A arrier speifies that the total number of awarded inbound loads to a faility is equal to, or within some parameter of, the number of awarded outbound loads from the same faility. The WDP model determines the atual volume awarded, so that the onditional bid only speifies the ratio of the awards. This is done to improve the balane at a speifi site and inrease the potential for ontinuous moves at that site. It differs from flexible paage bids in that it inludes the ondition that balane between two sets of lanes must be met Tier Bids A shedule of bid rates applies to a lane for a predetermined set of volume ranges on that lane. The relevant rate is applied to eah shipment depending on the volume of loads proessed that wee or month. This aptures the eonomies of sale of volume on a lane. Beause the atual rate harged is determined during exeution, it more aurately maps the arrier s osts. Regardless of the type of onditional bid used, the end result is a rate per load for eah lane used in exeution. Although the total value of eah bid is used for analysis, it is always divisible and easily alloated to eah speifi lane. In fat, the final upload to the downstream transportation management system is typially just a set of individual lane rates for eah winning arrier Winner Determination Problem The WDP is at the heart of optimization-based autions. For all TL autions, the WDP is formulated as a MIP. Although most software systems use a third-

7 Chris Caplie: Eletroni Marets for Truload Transportation Prodution and Operations Management 16(4), pp , 2007 Prodution and Operations Management Soiety 429 party solver, suh as CPLEX, to solve the WDP, there has been some reent wor around developing other more speialized solution approahes (Sandholm et al. 2005). The WDP assigns volume to arriers by lane (a single arrier is responsible for hauling on eah lane) or by load (eah arrier is assigned a number of loads to haul on eah lane awarded). In pratie, most software appliations use models that assign by load beause this permits other networ and business-speifi aspets to be onsidered. The most straightforward arrier assignment model permitting both simple bids and stati paage bids results in the formulation min subjet to i, j i, j y i, j i, j x i, j, (1a) x i, j i, j y x i, j i, j (1b) x i, j 0 i, j,, y 0, 1,, (1) (1d) where the notations are as follows. Indies: i: Shipping origin region j: Shipping destination region : Bid paage identifiation : Carrier identifiation. Deision variables: x i,j : Number of loads per time unit (wee, month), on lane i to j, assigned to arrier, under paage (whih in this ase is a simple bid) y : 1 if arrier is assigned stati paage bid ; 0 otherwise. Data: x i,j Volume of loads from shipper s, on lane i to j, that are being bid out i,j : Bid prie per load on lane i to j, for arrier, as part of onditional bid i,j : Volume of loads on lane i to j that arrier is bidding on as part of paage bid The objetive funtion [1a] minimizes the ost of assigning arriers to haul loads over the shipper s networ. The paage bid ost oeffiient is the total ost per planning time period for all volume on all of the lanes inluded in the paage bid submitted by arrier. Constraints [1b] ensure that the planned volume on eah lane is overed either by simple or stati paage bids. Note that the arrier must speify the exat number of loads requested for eah lane within eah stati paage bid, i,j. Stati paage bids are the most ommon form of paage bids used in transportation autions the arrier speifies the lanes and the exat level of flow per eah lane. Most of the ommerial software programs use similar formulations. More reently, flexible paage bids are being disussed, both with and without apaity limits. This is essentially just the introdution of paage level apaity onstraints. By introduing flexible paage bids, the model beomes min subjet to i, j i, j x i, j, x i, j x i, j i, j M i, j y x i, j 0,, i, j LB i, j y x i, j 0,, i, j UB i, j y x i, j 0,, i, j PL y x i, j 0,, i, j ij 0 i, j,, s, y 0, 1,, x i, j (2a) (2b) (2) (2d) (2e) (2f) (2g) (2h) where the additional variables and data are as follows: LB i,j: Lower bound in loads on lane i to j that arrier is bidding on as part of flexible paage bid ; UB i,j: Upper bound in loads on lane i to j that arrier is bidding on as part of flexible paage bid ; PL : Lower bound in loads aross all lanes that arrier is bidding on as part of flexible paage bid. The objetive funtion [2a] sums the produt of the individual lane bid pries and the awarded lane volume on eah lane within eah onditional bid. Constraints [2b] ensure that the volume in eah lane is overed by some arrier; onstraints[2] enfore the ondition that any arrier assigned any volume on a lane within a flexible paage bid is awarded the entire paage bid; onstraints [2d] and [2e] enfore the onditions that if any volume is assigned to a lane within a flexible paage bid, it satisfies the arrier s speified minimum and maximum lane volume requirements for that bid; and onstraints [2f] enfore the ondition that if any volume is assigned to any lanes within a flexible paage bid, the total paage volume awarded to that arrier under that bid paage satisfies the arrier s minimum volume requirement for the entire paage. Note that [2] is a more general formulation than [1] in that it handles simple, stati paage and flexible paage bids, all within the same deision variables. Simple bids are modeled as flexible paage bids onsisting of just one lane. Stati paage bids are modeled as flexible paage bids but with the upper and lower lane volume restritions set equal to the same value. Thus, the same deision

8 Chris Caplie: Eletroni Marets for Truload Transportation 430 Prodution and Operations Management 16(4), pp , 2007 Prodution and Operations Management Soiety variable, x i,j, an be used for all three of the primary onditional bid types. Simple reload bids also an be inorporated into [2] by adding onstraints [3a] and [3b] for eah faility, j, that is subjet to reload simple bid,, for arrier. j i j i x i, j x i, j i / i x j,i j j,, (3a) x j,i j j,, (3b) The terms j, j, j, j are absolute and relative onstants, respetively, that apture the possible relationships between the outbound and inbound volumes. Shippers typially use either of these two sets of parameters, but rarely both. A simple reload bid would typially also ontain minimum and maximum volume onstraints at the lane and paage levels. Within the WDP, the shipper an apply different business rules or priorities through the use of relative and absolute onditions. Relative onditions imply that some sort of monetary trade-off is made, whereas absolute onditions are hard onstraints Relative Conditions Relative onditions are applied in the WDP by modifying the oeffiients in the objetive funtion the ij in [1a] and [2a]. This is typially done by alloating penalties and rewards to the bids based on various servie attributes. For example, a shipper might want to favor inumbent arriers to minimize the hurn involved with introduing new vendors to a system. Bids from inumbent arriers would be disounted by, say, 5% within the WDP. If the inumbent arrier wins, the original rate would be fed into the TMS. This allows the shipper to test out the ost and importane of an otherwise arbitrary business rule. Many shippers onsider the utility exploration proess of agreeing on these relative fators aross the organization one of the most important benefits of a strutured aution proess. One of the primary benefits of relative onditions is that they an be applied without any modifiation to the underlying formulation used to solve the WDP; only the ost oeffiients must be adjusted Absolute Conditions Absolute onditions applied in the WDP tae the form of side onstraints. The three most ommon are business guarantee, arrier size, and minimum volume onstraints. Business Guarantee Constraints. A shipper often wants to ensure the amount of traffi that a arrier, or set of arriers, wins is within a ertain bound. The shipper might not want to rely too heavily on a single arrier, thus setting a maximum overage. Conversely, the shipper might want to give enough business to a arrier to remain a signifiant ustomer, thus setting a minimum. Coverage an be measured in terms of loads won or in total estimated dollar value. The onstraints below ensure that all arriers within some set of arriers C are awarded business within some preset volume (dollar value) bounds. CMin Value K N i, j x i, j CMinVolume K N C K ijn C K ijn C MaxValue K N, x i, j K C MaxVolume N (4a) (4b) Note that these onstraints an apply to a speified set of arriers (C), bid paages (K), or geographies (N). Some ommon onstraints inlude guaranteeing that the ore arrier group is awarded, say, at least 100 loads a wee out of a faility; ensuring that at least half of the loads overed in the northeast are awarded to arriers providing 53 trailers; setting a maximum of 20% of the total volume in the networ to be awarded to inter-modal servies, et. The onstraints are easy to explain and shippers tend to thin of their business in these terms. Care must be taen when MinVolume or MinValue onstraints are used to ensure feasibility is maintained. There is a tendeny for some shippers to over-speify or over-engineer a final award using these types of onstraints. Carrier Base Size Constraints. Another typial business onstraint is the restrition of the total number of arriers winning at the system, region, or lane levels. The number of arriers in the system or at a loation an be restrited through the use of relative or absolute onditions. The absolute ondition approah adds the following onstraints to limit the number of arriers assigned at the system and faility levels: M i, j w i x i, j 0,, i, j (5a) w i L i i (5b) M i, j z x i, j 0,, i, j (5) z S (5d) w i 0, 1, i (5e) z 0, 1, (5f) where the additional variables and data are as follows: w i : 1 if arrier is assigned to faility i, 0 otherwise; z: 1 if arrier is assigned to the networ, 0 otherwise; L i : Loation limit of arriers desired to serve faility i; S: System limit of arriers desired to serve networ as a whole; M i,j : Large onstant.

9 Chris Caplie: Eletroni Marets for Truload Transportation Prodution and Operations Management 16(4), pp , 2007 Prodution and Operations Management Soiety 431 The traditional approah of using a single large M variable, while reating more ompat formulations, an result in extremely frational LP solutions, maing it very wea in solving the IP. Barnhart et al. (1993) demonstrated that, in most ases, disaggregating the model leads to tighter bounds when solving the IP, as will minimizing the onstant, M. Setting M i,j to the minimum of ( x i,j ) for eah arrier, bid identifier, and lane ombination aomplishes this. Although the absolute onstraints mae sense at the faility or system levels, when applied to individual lanes it often results in one arrier winning the lion s share of the volume and the others winning the bare minimum to satisfy the onstraint. This is less desirable in pratie beause many shippers want a more balaned distribution. One way to reate more balane is to simply add in a maximum volume onstraint for eah arrier for the loation or lane in question equal to the perentage of the business the largest arrier is desired to haul using the business guarantees onstraints shown earlier. A relative ondition an also be used to disourage additional arriers from being awarded business by modifying the objetive funtion as follows: min i, j x i, j i, j i F i w i F z, (6) where all variables are the same as previous models with the addition of F : Cost of inluding arrier into the system; F i : Cost of inluding arrier to serve loation i. The fixed osts an be both arrier and loation speifi, as shown above, or the same for all arriers and all loations. Essentially, the fixed osts at as penalties for adding additional arriers to the winning set. The two most ommon uses of these onstraints are to limit the total number of arriers awarded any business and to limit the number of arriers serving a faility on both the inbound and the outbound sides so as to minimize the size of the required trailer pool. The latter onsideration also enourages the use of ontinuous moves at that faility, beause speifi arriers will tend to win both inbound and outbound business. Minimum Volume Constraints. Shippers will often wish to guarantee that if a arrier is awarded any business, then it has to be of a ertain minimum level. Constraints [7] below ensure that if a arrier is awarded any business aross the networ, then it must be at least SV loads. SV z x i, j 0 (7a) i, j M i, j z x i, j 0,, i, j (7b) The use of ombinatorial autions for the prourement of TL servies revolutionized the industry. What was one a straightforward rate olletion exerise is now a strategi planning event. Through the use of a wide variety of onditional bids, the arriers an better express their underlying eonomis and hopefully reate better servie networs. The use of both relative and absolute onditions within the WDP allows the shipper to apply business rules and priorities to the arrier seletion proess. Regardless of the types of bids allowed or the onditions applied within the aution, however, the end result is a set of arrier assignments (whih arrier is awarded whih lanes or what perentage of whih lanes) and pries (how muh is eah arrier to be paid). 4 Operational Prourement: Catalogs and Exhanges Although onsiderable attention has been paid in the literature to the details of ombinatorial autions, very little has been written on what happens when the aution ends. Operationalizing the results of a strategi prourement an be quite ompliated. For TL servies, most shippers use an eletroni atalog for the majority of their business with exhanges for emergeny, distraught, or distressed freight. This setion disusses the issues surrounding the use of eletroni atalogs, exhanges, and the automation of these tehnologies Eletroni Catalogs The output from a strategi TL aution is a set of rates and assignments of arriers to lanes. The rates and assignments an be thought of as a representation of the shipper s transportation strategy and business priorities beause they were shaped by all of the side onstraints and business rules that were applied by the shipper during the solving of the WDP. Carriers that were awarded volume on a lane in the WDP of the aution are usually referred to as primary arriers on that lane. Lanes may have more than one primary arrier. The rates and assignments are uploaded into a database ommonly referred to as a routing guide. Additionally, a number of the non-winning rates for eah lane are often loaded into the routing guide and used as ba-up or alternate arriers. 3 As mentioned earlier, most shippers routing guide is an eletroni relational database that interfaes with an exeution software system, a TMS. The TMS is a deision support and transation proessing system that handles the day-to-day management of transportation operations. It is usually modular and an inlude suh sub-systems as order management, load 3 Carriers are expeted to honor the rates submitted in an aution for a lane they did not win, but they are not held to the same arrier aeptane performane standards as the primary arrier.

10 Chris Caplie: Eletroni Marets for Truload Transportation 432 Prodution and Operations Management 16(4), pp , 2007 Prodution and Operations Management Soiety aggregation, shipment onsolidation, tendering, arrier seletion, payment, and reoniliation. The arrier seletion and ommuniation funtionality of a TMS is essentially an eletroni atalog. It is omposed of three main omponents: a rating database that ontains all valid and upto-date rates for the different lanes; a routing guide that desribes the priorities, arrier apaities, and other seletion logi; a ommuniation module that allows the shipper to send tenders to arriers and reeive aeptanes or rejetions. In some simple ases, the routing guide and the rating database are one and the same. But, as the rules surrounding the strategi seletion of arriers get more omplex, they are usually broen out separately to speed up the response time. The routing guide provides rules for hierarhy and order preferenes. The rate database is then only a depository of rates with no imbedded logi. Through the TMS, load planners at any loation an determine whih arriers are authorized for hauling on any lane. This apability allows a shipper to entralize prourement (through a firm wide aution) while deentralizing operations. The TMS ensures that eah load planner will adhere to the strategi prourement plan. The user enters the origin, destination, and equipment type and the TMS returns the primary arrier or, perhaps, a list of the preferred and alternate arriers authorized for that lane. Only those arriers that have been approved by the shipper are listed in the atalog. Essentially, this is an aggregated eletroni atalog for buyers. If a load planner deviates from the routing guide, this maveri buying behavior an be flagged, traed, and measured. In sophistiated systems, the entire arrier seletion and ommuniation proess is automated. The shipment is automatially mathed to the orret rate and that arrier is ommuniated with using eletroni data interhange (EDI) standards. Exeption management an also sometimes be handled in an automated fashion. The most ommon form of automation is nown as a waterfall or sequential proess. The shipment is first mathed and then tendered to the primary arrier in the relevant lane. The arrier is given a set amount of time to respond. The allowed response time is usually based on the available lead time for the shipment and an range from minutes to days. If the arrier rejets the load, the TMS will move to the alternate arrier list for that lane and tender it to the next arrier in the list. The sequene of the alternate arriers on a lane is typially from the lowest to highest ost. The proess ontinues sequentially yling through arriers until either time or the number of alternates runs out. When this happens, the atalog has failed and the shipper typially esalates the searh to an exhange, as desribed in Setion 4.2. Generally, the mathing of a shipment to the appropriate arrier rate is straightforward. However, the more omplex bid strutures and onstraints that are used within the strategi TL aution, the more sophistiated the eletroni atalog must be. For example, Various geographi speifiity: As disussed in Setion 2.3, the lanes represented in the strategi aution might onsist of zones or regions. This requires the routing guide to selet whih rate orretly applies. Using the same example as in Setion 2.3, the shipper might have olleted rates from the same arrier for the following three lanes: Atlanta, GA to Chiago, IL for 10 loads/wee at $1.10/mile; Atlanta, GA to IL 606 for 3 loads/wee at $1.25/mile; Georgia to Illinois for no volume guarantees at $1.15/mile. The routing guide would need to now whih of these rates to apply to a speifi lane. This is handled either expliitly, by exploding and dupliating rates for zones into the finest grained representation possible, or through imbedded logi. For the expliit method, the shipper reates dupliate detailed rates at the finest detailed points within eah of the zone loations. So, for example, 99 additional rates of $1.25 per mile would be added to the rating database for the lanes from to 60600, 60602, 60603, 60604, et., that is, for every 5-digit postal ode within the 606 region with the exeption of The same would be done for the state-to-state lane rate. This approah would obviously signifiantly expand the size of the routing guide but it would not require any hange to the simple mathing logi. The seond approah requires the routing guide to have some sort of searh and prioritization logi. Typially, the rule most TMSs follow is to searh from the most to the least geographially speifi lane definition. Multiple Primary Carriers on a Lane: It is ommon for a shipper to assign two or more primary arriers to a high-volume lane. 4 The thought is that having more arriers on an important lane an provide additional apaity muh more easily than a single arrier. In the WDP, eah winning arrier is awarded a portion of that lane s total volume. 5 The routing guide, then, musto be able to determine whih arrier to selet on a load-by- 4 An opposing shool of thought is to single soure eah lane to one arrier and give them broerage rights up to a pre-determined level, say, 15%. 5 In the WDP arrier apaity is treated as an absolute value, whereas in the eletroni atalog it is treated as a relative value.

11 Chris Caplie: Eletroni Marets for Truload Transportation Prodution and Operations Management 16(4), pp , 2007 Prodution and Operations Management Soiety 433 load basis. Suppose two arriers are assigned primary status on a lane, one with 60% and the other with 40%. Some alloation logi must exist within the routing guide to ensure the volume is tendered equitably to eah of the primary arriers. This requires that the routing guide store the awarded or planned apaity (usually as an alloation perentage) for eah arrier for eah lane, as well as eep tra of the urrent ount of all loads tendered to eah arrier on eah lane. 6 An even more sophistiated approah is to ount not only loads hauled by eah arrier, but also the turn-downs by a arrier as well. This serves to penalize poor performane. The alloation sheme usually tenders the next shipment to the arrier with the largest differential between tendered and awarded perentage. Tier Bids: These bids allow a arrier to speify a sequene of rates for a lane that are dependent on the weely volume of traffi already awarded to that arrier on that lane. To apply these bids, the eletroni atalog must tra the weely volume by arrier by lane, as well as retain multiple rates for the same lane that differ only in the volume range in whih they apply. Very few TMSs an handle these types of rates. In pratie, the eletroni atalog simply annot handle the impliit assumptions or rules applied during the strategi aution. For example, Simple Lane Bids with Volume Constraints: Although the amount of traffi awarded to eah arrier in the WDP is prediated on some foreasted flow on eah lane, in pratie, the amount of volume on a lane will vary dramatially from wee to wee. Indeed, the oeffiient of variation of loads per wee on a traffi lane is usually above 200% (Harding 2005). So, although the WDP might apply a system-wide or faility-speifi maximum (or minimum) volume per wee for a arrier, this is rarely enfored or even traed within a TMS. Minimum Volume Constraints: In the WDP, the shipper an apply a onstraint that guarantees that if a arrier wins any business (on a lane, from a faility, or aross the system), then they must win at least some threshold volume. This is meant to ensure that a arrier has enough volume to justify his serving the lane in the first plae. This also applies to the ases where a set of arriers are awarded some minimum level of business in the strategi aution. This an be minority arriers, regional arriers, or some other ategory. In the eletroni atalog, however, the atual amount of 6 The time period used to alulate the perentage of traffi awarded is typially weely, but an be longer for some shippers. business awarded to a single or set of arriers is not traed or managed. In the best ase, the atual awarded traffi is olleted and reported on a monthly or quarterly basis during a arrier performane meeting. The only exeption to this is where some shippers will measure arrier turndowns differently if the total volume on the lane is above the total planned volume. Paage Bids: One of the justifiations for using optimization for solving the WDP is to allow for paage bids. The impliit assumption that many arriers mae when submitting paage bids is that the sequene of shipments within a paage bid will be suh that a ontinuous move or tour will be possible. In pratie, there is not only no guarantee of this, but also usually no mehanism to even he. In most TMSs a rate olleted as a paage bid is onverted into a primary arrier rate for the relevant lanes and annot be distinguished from a rate olleted as a simple bid in the routing guide. It is interesting that although there has been tremendous ativity both in the literature and in pratie to improve on and add to the sophistiation of the strategi aution proess, the real limiting fator for widespread adoption is the la of sophistiation in the eletroni atalog or routing guide. Even some nowstandard aution praties annot be applied or enfored aurately, even in state-of-the-art systems Exhanges TL transportation exhanges are used to aess the spot maret primarily when ontrated rates fail. Although there were preditions during the Internet bubble that publi freight exhanges would handle all of a shipper s TL needs through a entralized learing house where mathing would our automatially, these have not ome to fruition. We estimate the spot maret omprises less than 10% of the total TL maret and there does not appear to be any automati mathing ourring in pratie. In all ases of a TL exhange, some human (from either the shipper or the arrier side) is pushing the requirement or onfirming a reommended math. The phrase freight doesn t just move or the need to play dial-a-tru imply that to math a load to a tru in stressed times someone must get personally involved. There are very few publi exhanges in the U.S. that handle spot traffi. The two most ommon publi exhanges today for TL servies are TransCore s DAT Connet 7 and Internet Trustop. They are essentially automated eletroni load boards. Shippers an post available loads while arriers an use sophistiated filters (geographi region, maximum distane from a 7 DAT originally stood for Dial-A-Tru.