This article was downloaded by: [132.174.254.

72] On: 28 June 2019, At: 20:00

Publisher: Institute for Operations Research and the Management Sciences (INFORMS)
INFORMS is located in Maryland, USA

INFORMS Journal on Applied Analytics

Publication details, including instructions for authors and subscription information:
http://pubsonline.informs.org

Optimizing New-Vehicle Inventory at General Motors

Robert R. Inman, Michael C. Frick, Thomas D. Hitchman, Robert A. Muiter, Jonathan H.
Owen, Gerald M. Takasaki

To cite this article:

Robert R. Inman, Michael C. Frick, Thomas D. Hitchman, Robert A. Muiter, Jonathan H. Owen, Gerald M. Takasaki (2017)
Optimizing New-Vehicle Inventory at General Motors. INFORMS Journal on Applied Analytics 47(5):396-410. https://
doi.org/10.1287/inte.2017.0902

Full terms and conditions of use: https://pubsonline.informs.org/page/terms-and-conditions

This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact permissions@informs.org.

The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.

Copyright © 2017, INFORMS

Please scroll down for article—it is on subsequent pages

INFORMS is the largest professional society in the world for professionals in the fields of operations research, management
science, and analytics.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
 INTERFACES
Vol. 47, No. 5, September–October 2017, pp. 396–410
http://pubsonline.informs.org/journal/inte/ ISSN 0092-2102 (print), ISSN 1526-551X (online)

Optimizing New-Vehicle Inventory at General Motors

Robert R. Inman,a Michael C. Frick,a Thomas D. Hitchman,a Robert A. Muiter,a Jonathan H. Owen,a Gerald M. Takasakia
a General Motors Company, Detroit, Michigan 48265
Contact: robert.inman@gm.com (RRI); mike.frick@gm.com (MCF); thomas.hitchman@gm.com (TDH); bob.muiter@gm.com (RAM);
jonathan.owen@gm.com (JHO); gtakasak@aol.com (GMT)

https://doi.org/10.1287/inte.2017.0902
Copyright: © 2017 INFORMS
Abstract. This paper describes how General Motors (GM) uses operations research to
optimize its new-vehicle inventory. The solution answers two complementary questions:
(1) what is the optimal number of vehicles to build? and (2) what are the optimal vehicle
configurations? To answer the first question, for each vehicle model, we find the inven-
tory that maximizes profit less inventory carrying costs, which differs from the standard
approach of finding the inventory needed to satisfy a given fill rate. To answer the second
question, we provide a decision support tool to help dealers order the best variations
for each vehicle model using a set-covering philosophy, which differs from the standard
approach of recommending the highest-selling variants. This paper describes the business
processes surrounding these decisions, how the need for implementation and ongoing
use guided our solution development, and the impact on the business.

Keywords: automotive • dealer • inventory • profit

Introduction the automaker to the dealer. Customers may request

The majority of new cars and trucks sold in the United
States are sold from dealers’ lots. Managing the num-
ber and variety of this dealer inventory is a challenge;
optimizing it is an opportunity. In the United States, as
in most of the world, most automakers distribute new
vehicles to franchised dealers who then sell them to
customers from inventory. Although some automakers
sell a limited number of vehicles to customers directly,
franchise laws generally prohibit direct-to-customer
sales for brands with existing franchise contracts. The
autonomous and independently owned dealer part-
ners complicate the inventory-management decisions.
Our project succeeded within this context by devel-
oping and implementing operations research solu-
tions that respect the business processes of both Gen-
eral Motors (GM) and its dealers; are understood and
trusted by GM and its dealers; and benefit GM, its deal-
ers, and customers.
GM and most other U.S. automakers typically re-
quire a dealer order before they produce a vehicle.
A dealer fully configures and orders a vehicle from the
automaker, who then manufactures the physical vehi-
cle to match that order. When the vehicle ships from the
assembly plant to the dealer, ownership transfers from
396
that a custom-configured vehicle be built to order; how-
ever, franchise laws typically require the customer to
engage a dealer to place the order on the customer’s
behalf. In 2015, only five percent of U.S. vehicle sales
were built to order, whereas 95 percent were sold from
dealer inventory (Vellequette 2016). Once delivered,
the dealer offers these physical vehicles in stock for
sale to customers. If a customer visits a dealer but can-
not find a suitable match in stock, the customer may
ask the dealer to order a custom-configured vehicle
or to search other dealers for a match. If the dealer
finds a match at a nearby dealer, the two dealers can
trade physical vehicles so that the customer receives
the desired vehicle. If dealer inventories grow too high,
dealers will order fewer, if any, additional vehicles.
These sales reductions from the automaker to the deal-
ers typically lead the automaker to offer incentives or
discounts to dealers and end customers.
Optimizing new-vehicle inventory requires the auto-
maker to make two decisions: (1) what is the optimal
number of vehicles to build? and (2) what are the opti-
mal vehicle configurations? Determining the optimal
number of vehicles on dealers’ lots helps GM make
better production decisions (e.g., overtime, assembly
Inman et al.: Optimizing New-Vehicle Inventory at General Motors
Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS
line rate, and number of shifts) and marketing deci-
sions (e.g., rebates and advertising). Determining the
optimal mix of vehicle configurations (i.e., the trim
levels, options, and colors) to have on dealers’ lots
makes the most efficient use of this inventory. Even
with ample aggregate inventory, dealers might have
stocked the wrong vehicle configurations and cus-
tomers might not be able to find their desired vehi-
cles. If inventory is too low, dealers and GM lose sales
because customers cannot find their desired vehicles.
If inventory is too high, customers might have a wide
selection (if the mix is correct); however, the cost bur-
den is substantial.
This operations research application improves both
the aggregate inventory level and inventory-mix deci-
sions. It improves the aggregate inventory decision
by finding the best number of vehicles to stock using
a bottom-up algorithm that considers each vehicle
variant at each dealer—as opposed to a top-down
aggregated approach that does not consider each indi-
vidual model’s variety and each dealer’s sales charac-
teristics. The aggregate vehicle inventory is the total
of the bottom-up inventory for each model variant
and for each dealer. The dealer inventory-mix decision
support tool improves the dealer inventory-mix choice
using a practical set-covering approach that shows
the dealer the best portfolio of vehicle configurations
to stock.
Instead of the standard approach of finding the
inventory needed to provide a given service level or
fill rate, we find the inventory that maximizes variable
profit less inventory carrying costs. The service level
is the number of stockouts per period, and the fill rate is
the fraction of desired sales fulfilled. Variable profit is
the revenue received less the variable cost for that vehi-
cle. We balance the variable profit from a sale with the
inventory carrying cost for the length of time needed
to sell. Some may argue that the automaker should be
unconcerned about the dealer’s inventory carrying cost
because it appears to be borne by the dealer. However,
taking a total supply chain viewpoint and consider-
ing it as a single aggregate entity, regardless of which
stakeholder within the supply chain pays the inven-
tory carrying cost (e.g., for floor space, insurance, and
cost of capital), the aggregate supply chain pays for it.
Since all the vehicles on dealers’ lots do sell eventually,
some managers pardon excessive vehicle inventory as a
397
legitimate inventory-pressure tool for increasing sales.
However, we argue that excess inventory can sub-
stantially hurt profitability. A useful insight is that a
vehicle’s well-known model-year designation causes
its value to customers to deteriorate over time. As a
stocked vehicle ages and its value deteriorates, larger
and larger profit-depleting discounts are necessary
to sell it. This observation provides a mechanism to
find the profit-maximizing inventory. Too little inven-
tory will result in lost sales because of stockouts. Too
much inventory will result in holding costs that erode
profitability.
Another innovation of our approach is that we take
a set-covering philosophy to guide individual dealer
ordering. Instead of ranking full configurations (i.e.,
vehicle variants determined by the combination of
model, body style, all options, and colors) by sales,
we identify gaps in each dealer’s inventory portfolio.
A full configuration is a vehicle configuration with
every option and color specified. Full configurations
are completely specified. If two vehicles differ in any
option or color, they have different full configurations.
Advising dealers to stock vehicles based on a sales
ranking of full configurations is easy, but flawed in
practice. The first major flaw is that most automakers’
models can be configured in many—usually millions—
of ways. No dealer can stock millions of a single model.
Instead a dealer might stock the highest-selling dozen
or so configurations—less than 0.01 percent of possi-
ble configurations. Moreover, two of these configura-
tions (e.g., Vehicle A and Vehicle B) might differ only
in minor features (e.g., wheel locks, all-weather floor
mats, or a cigarette lighter and ashtray). Most cus-
tomers would be reasonably satisfied by either Vehicle
A or Vehicle B; therefore, a dealer does not have to stock
both vehicles. Stocking only one of the two configu-
rations and using the other inventory spot for a very
different configuration would be advantageous to the
dealer. A dealer who stocks full configurations in pro-
portion to sales risks covering only a very small frac-
tion of demand. To address this, we developed a deci-
sion support tool that applies a set-covering approach
using partial configurations. The tool helps dealers
avoid concentrating their inventory on a few practi-
cally indistinguishable full configurations. To increase
dealer adoption, we made the tool transparent and
easy to use to ensure that dealers of all levels of

398
sophistication would understand and trust the recom-
mendations. This dealer inventory-mix tool provides a
dealer with a comprehensive view of demand, so that
he or she knows which colors and options have been
selling at nearby dealerships—not solely at his or her
store, displayed in a set-covering framework that iden-
tifies dealer inventory-portfolio gaps.
The paper proceeds as follows. First, we describe
the methodology for optimizing the aggregate new-
vehicle inventory for each vehicle model; this answers
the question “what is the optimal number of vehicles to
build?” Then we describe the dealer decision support
tool for optimizing the dealers’ inventory mix for each
vehicle model; this answers the question “what are
the optimal vehicle configurations?” Finally, we sum-
marize our solution’s usage and impact at GM. The
appendix provides mathematical documentation.
Optimizing Aggregate
New-Vehicle Inventory
An automaker’s cars and trucks on dealers’ lots com-
prise the automaker’s retail inventory. While standard
inventory analyses estimate the amount of inventory
Figure 1. (Color online) Rather Than Setting Inventory to Provide a Given Service Level, We Set It to Maximize Profit
99%
Service level or fill rate
Service level approach
Service level
$
X
Maximize profit approach
Max profit
Inman et al.: Optimizing New-Vehicle Inventory at General Motors
Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS
needed to provide a given service level from a top-
down perspective, we estimate the most profitable
inventory for GM using a bottom-up perspective that
explicitly models the details and complexities of the
product and the distribution network. Next, we high-
light our approach’s innovations.
Profit-Maximization Objective
Figure 1 contrasts our profit-maximization approach
with the standard service level approach. Because of
the difficulty of quantifying the inventory’s impact on
profit, most inventory analyses determine the inven-
tory needed to provide a given service level or fill rate.
In contrast, we maximize profit by balancing the vari-
able profit generated by an inventory level’s fill rate
with the costs of holding that inventory level. Taking
a portfolio perspective, our approach adjusts for diver-
sions that occur when a customer’s intended vehicle
configuration is not in stock, and the customer instead
diverts to a second-choice vehicle model from the same
automaker. The profit lost from not having the first-
choice vehicle in stock is mitigated because some cus-
tomers will not walk away; they will instead purchase
a different model from the same automaker.
Service level or fill rate
Inventory
Variable profit from sales
Carrying cost
Variable profit
Less carrying cost
Inventory
Inman et al.: Optimizing New-Vehicle Inventory at General Motors
Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS
Inventory Carrying Cost Borne by the Supply Chain
Recognizing that the supply chain incurs a substan-
tial cost for dealer inventory enables the profit-maxi-
mization approach. The typical carrying cost includes
the cost of capital, insurance, facilities, security, and
so forth. Since new vehicle inventory is on the deal-
ers’ books, some may argue that it does not cost the
automaker anything. However, Copeland (2009, p. 5)
notes “a striking feature of the automobile market is the
constant decline in prices throughout the model year.”
Copeland et al. (2011) document that for the typical
vehicle, prices fall over the model year at a 9.0 per-
cent annual rate. Copeland et al. (2005, p. 3) document
that “all other things being constant, higher inventories
are associated with lower retail prices.” Corrado et al.
(2006) recognized the offsetting impacts of rebates and
price increases, and report a significant price decline
over a model year. The authors estimate that the annu-
alized average transaction price that consumers pay
(considering customer rebates and interest subvention)
decreases by 6.1 percent over the model year. The price
before incentives generally declines from the start of
the model year, and the price after incentives declines
even more.
The model-year designation of new vehicles signif-
icantly influences their resale value. Used-car buyers
judge the age of a vehicle and the freshness of its
features by its model year. Vehicle inventory carried
into the following model year must be discounted to
attract buyers away from the new model-year vehicles.
So, although every vehicle produced does eventually
sell, excess model-year inventory from the previous
year will require a profit-depleting rebate. This can be
approximately modeled with a carrying-cost factor to
avoid excessive unsold inventory from the previous
model year.
Demand and Lead-Time Distributions
Instead of relying on standard symmetric probability-
density functions, we analyzed observed demand and
lead-time data to accurately represent these sources of
randomness. For example, we can see that a dealer’s
order-replenishment lead time will increase when
demand exceeds the finite plant capacity, thus leading
to skewed and positively auto-correlated lead times.
See Blumenfeld et al. (1999) for an analysis of the
impact of order-replenishment lead-time variability on
finished-goods inventory.
399
Geographic Granularity
Simpler analyses aggregate cars of all dealers into a sin-
gle inventory. However, recognizing that a car in inven-
tory in Alaska does little for a customer in Florida,
we analyze the inventory at each dealership; we map
these dealers in Figure 2. Each dot marks a dealer. Since
the dots overlap for dealers in close proximity to each
other, areas with a dense network of dealers appear as
dark shading.
Product Granularity
Although it is possible to use each of a vehicle’s mil-
lions of full configurations (i.e., the various combi-
nations of product content and colors) as the level
of granularity, we instead use the concept of prior-
itized partial configurations. A partial configuration
is an incomplete vehicle specification. A prioritized
partial configuration is a partial vehicle specification
where only the most important attributes—from the
customer’s perspective—are specified. We apply the
Pareto principle to identify a few principal vehicle
options to represent most customer preferences. For
example, a vehicle model’s prioritized partial config-
uration could be its body style and its powertrain
(e.g., two-wheel or all-wheel drive). By prioritizing the
attributes from the customer’s viewpoint, the complete
configurations within a prioritized partial configura-
tion would be similar in a customer’s mind. We assume
that the customer is relatively indifferent to the less
important options and could be satisfied if the sales-
person offers a portion of the dealer’s margin as an
inducement. For simplicity, we simply say partial con-
figuration when we mean prioritized partial configu-
ration. Since the vehicle color plays a substantial role
in customer choice, the bottom-up algorithm explicitly
includes external color as part of the partial configura-
tion characterization.
As an example, consider the model year 2013 Chevro-
let Equinox, which had millions of unique buildable
configurations—most of which were never built. How-
ever, by choosing the body style, powertrain, and color
as the most important attributes to the customer, a
more manageable 81 partial configurations were pos-
sible; this is less than 88 combinations (i.e., four body-
style choices times two powertrain choices times 11
exterior-color choices, as we show in Figure 3) because
not all combinations were orderable. Figure 3 displays
the breadth and depth of efficient inventory for a typical

400
Figure 2. (Color online) GM Has a Large, Geographically Disperse Dealer Network
Alaska
Hawaii
dealer. The symbol shape represents the combination
of body style and powertrain, and the hatching repre-
sents different exterior colors. If a dealer who sells 12
Equinox vehicles per month on average (or 0.4 per day)
stocked one of each of the 81 partial configurations,
this would represent 81 divided by 0.4 (i.e., 202.5 days-
supply). Days-supply is the number of vehicles in inven-
tory divided by the sales per day (i.e., the number of
days that a given number of units in inventory will stay
in inventory using the average sales rate). For a more
typical inventory level of 75 days-supply, Figure 3 dis-
plays a good mix. The inventory breadth is the number
of partial configurations the dealer stocks, and approxi-
mates the amount of customer selection. The inventory
depth is the number of vehicles within each partial con-
figuration to stock. For top-selling partial configura-
tions, the dealer may stock more than one unit to avoid
stockouts when a vehicle sells and the dealer must wait
for replenishment. To increase the selection within a
partial configuration, the additional unit will typically
differ in a less important option.
Inman et al.: Optimizing New-Vehicle Inventory at General Motors
Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS
The partial configurations on the horizontal axis in
Figure 3 are ranked by decreasing sales. We would plan
to hold two units of the two top-selling partial con-
figurations (e.g., a red two-wheel-drive Equinox with
trim level 1LT, and a black two-wheel-drive Equinox
with trim level 1LS) because of their high sales rates
and the order-replenishment lead time. As we move to
the right, the lower the sales rate, the lower the inven-
tory depth. Eventually, we reach a point at which the
expected sales rate is so low that we would expect
the vehicle to sit in inventory for an extended period.
In that case, the expected variable profit from a sale
does not compensate for the cost of carrying the unit
in inventory. In Figure 3, the sales rate of partial
configuration #29 is so low that we would expect it
to sit in this dealer’s inventory for 225 days, thus
incurring a substantial carrying cost. Hence, stock-
ing partial configuration #29 is not cost effective. We
account for the low-selling partial configuration inven-
tory, but a single dealer will rarely stock all partial
configurations.
Inman et al.: Optimizing New-Vehicle Inventory at General Motors
Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS 401

Figure 3. (Color online) Our Approach Determines Both the Depth and Breadth of Inventory for Each Model at Each Dealer

Equinox for a selected dealer

4 body styles, 2 powertrain choices, 11 exterior colors => 81 orderable combinations
This dealer sells 12 Equinox’s per month
Below are 30 vehicles (or 75 days supply) in 28 varieties that cover all major equipment variations
and over two-thirds of sales
2WD-1LS
Inventory
(vehicles)
2WD-1LT 8 major
equipment
2WD-2LT
combinations
Depth
Depends on 2WD-1LZ
sales, order

…
response
time, and
variability

1 2 3 4 5 6 7 8 9 10 15 20 25 26 27 28 29 30 … 81
Body style/Powertrain/Color combinations ranked by total sales volume

Breadth Configuration #29 represents 1.11% of sales

Depends on number of body styles, —Expect it to sit on the lot for 7.5 months
equipment options, colors, and sales —Not cost effective to stock it at this dealer

Dealer Trades Integrating Optimization into the Business Process

Although we consider each dealer individually, we
also consider the impact of the network of dealers via
dealer trades. When a customer cannot find a satis-
factory vehicle in a dealer’s inventory, that dealer can
trade vehicles with another dealer to satisfy that cus-
tomer. Since dealer trades expose the customer to a
wider selection of vehicles, they are effectively pool-
ing the inventory of nearby dealers, which reduces the
inventory needed by a specific dealer. However, as Iyer
et al. (2009) indicate, swapping physical vehicles costs
several hundred dollars.
Build-to-Order Sales
Vehicles that are built to satisfy specific customer
orders never sit in dealer inventory because the cus-
tomers are willing to wait for the new vehicles to be
built based on their exact specifications. If all sales were
built to order, a dealer would need little inventory; a
few demonstration vehicles would suffice. We account
for the impact of build-to-order vehicles on a dealer’s
inventory requirements.
The concept of lead-time demand—the demand that
occurs during the order-replenishment period—is cen-
tral to inventory calculations. Although the average
is reported regularly, the variability is not. Hence, for
lack of data, some inventory analyses assume that
demand follows a standard probability distribution,
such as the Poisson. Although the Poisson distribution
has theoretical appeal, actual sales data often exhibit
more variability than the Poisson distribution. Several
factors could cause sales data to exhibit greater vari-
ability than the Poisson distribution. First, automotive
sales exhibit seasonality. Second, many automakers
have dealer-incentive policies based on sales in a par-
ticular period, such as a month. At the end of the
period, a dealer who is within range of qualifying for
the incentive will use all available means to qualify.
Third, weather often makes car shopping difficult and
can delay demand until the following week. Fourth,
the competitive landscape changes with new model
entries, recalls, and competitor incentives and mar-
keting campaigns. Fifth, customer preferences change

402
over time (e.g., because of fluctuations in gas prices),
which shift customers from one segment to another. In
summary, for many reasons, observed sales vary more
than level (yet purely random) demand.
Sales variability and lead-time variability together
determine the variability in lead-time demand. In prac-
tice, actual lead times often exhibit a positive skew.
This is not surprising when one considers the delivery
process. If no problems occur, the vehicle should be
delivered in a specified number (x) of days; there is no
good reason for it to arrive in less than x days. So, the
lead-time distribution will have, at most, a short tail to
the left. If a delay occurs (e.g., traffic congestion, road
construction, an accident, inclement weather, a driver
arriving late to work or calling in sick, labor unrest, or
an inspection delay at a border), then the delivery will
be late. Many such delays can be lengthy and some can
cause additional delays. Hence, the distribution of lead
times has a short tail to the left and a long tail to the
right. In addition, sequential lead times may be posi-
tively correlated (because many delaying factors could
persist over time or create a backlog that will require
time to fulfill), which deviates further from the stan-
dard assumption of independent trials.
Recognizing the variability of sales and limited pro-
duction flexibility, the expectation of operating exactly
at the target (profit-maximization) inventory level is
unrealistic. Instead, we manage inventory by keeping
it within the normal operating bands as defined by the
minimum, target, and maximum inventory levels. To
find the lower limit of inventory, we estimate the inven-
tory needed to provide a high fill rate, ignoring the
product variety and financial impact. That is, at this
minimum inventory, the dealer will (almost) always
have at least one unit of the model in stock. It may
be the wrong style or color, but at least one is avail-
able. This is a lower limit on the most effective inven-
tory. Because it is based on an objective other than
profit maximization, it also serves as a safeguard for
the more sophisticated target inventory optimization.
The target is the bottom-up estimate of the most prof-
itable inventory. The appendix shows the mathematical
details. The upper limit is the maximum inventory-
control limit set to provide a buffer above the target.
Exceeding the upper inventory limit warrants correc-
tive action.
Figure 4 illustrates the profit-maximization frame-
work. The vertical axis represents cost, sales, or profit
Inman et al.: Optimizing New-Vehicle Inventory at General Motors
Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS
Figure 4. (Color online) In the Profit-Maximization
Framework, We Maximize Sales Contribution Less
Carrying Cost
$ Variable profit
Carrying cost
Variable profit
Less carrying cost
1 2 3 Inventory
measured in dollars. The horizontal axis is the amount
of retail inventory measured in the number of vehicles
in inventory. As inventory increases, its carrying cost
increases as the upward sloping dashed line shows. As
inventory increases, the sales contribution (i.e., vari-
able profit of vehicles sold) increases because fewer
customers are unable find an acceptable vehicle in
inventory, as the concave dashed line shows. The objec-
tive is to maximize the sales contribution less the inven-
tory carrying cost, as the unimodal solid curve shows.
In Figure 4, the numbers in the circles below the hor-
izontal axis identify the following key inventory lev-
els, which define the minimum, target, and maximum
inventory. The appendix provides details.
1. The lower limit on inventory is based solely on
the fraction of lost sales, regardless of financial impli-
cations. The key factors driving this lower limit include
the order-replenishment time (the longer the time, the
more inventory), order-replenishment time variability
(the more variable, the more inventory), demand vari-
ability (the more variable the demand, the more inven-
tory), and sales rate (the lower the sales rate, the more
days-supply).
2. The target inventory maximizes aggregate vari-
able profit less inventory carrying costs. The key factors
impacting the target include the variable profit (the
higher the variable profit, the more inventory), dealer
cost (the lower the dealer cost, the more inventory),
price elasticity, and diversions (the less likely to divert,
the more inventory).
3. The upper limit on inventory is set above the
target to accommodate forecast uncertainty, modeling
approximations, dealer stocking variances, and process
limitations.
Inman et al.: Optimizing New-Vehicle Inventory at General Motors
Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS
Figure 5. (Color online) Inventory-Control Bands Straddle the Profit-Maximizing Target to Trigger Production and
Sales-Promotion Adjustments

6EHICLESININVENTORY
Lower band
,EVELBELOWWHICH
INVENTORYRISKSLOSINGSALES
*AN &EB -AR !PR -AY *UN

The target inventory, and the lower and upper inven-
tory limits define control bands for the aggregate vehi-
cle inventory (Figure 5). The circled numbers 1, 2, and 3
in this figure correspond to the lower-limit, target, and
upper-limit inventories depicted in Figure 4. Since those
benchmark inventories are defined in terms of days-
supply, and the vertical axis in Figure 5 reflects vehicles,
the control bands oscillate over time as demand varies.
Figure 5 depicts a snapshot as of July 1, 2016. The solid
line from January 2016 through June 2016 tracks actual
inventory. The solid line from June through Decem-
ber 2016 tracks forecast inventory from the difference
between planned production and forecast sales. GM
uses these inventory-control charts when making pro-
duction, capacity, fleet sales, and rebate decisions. How-
ever, the dealers manage the vehicle-inventory mix (i.e.,
which dealers stock which configurations), as we dis-
cuss in the next section.
Dealer-Ordering Decision Support Tool
Optimizing GM’s retail inventory level is an important
factor in planning production. Yet, dealers are inde-
pendent franchisees; as such, they retain the authority
403
Upper band
,EVELABOVEWHICHINVENTORYRISKSCOMPROMISINGPROFIW
WITHHIGHCARRYINGCOSTANDORAGGRESSIVEINCENTIVES
Projected inventory Inventory target
!CTUALAND 0ROFITMAXIMIZINJ
FORECASTINVENTORY INVENTORYLEVEL
*UL !UG 3EP /CT .OV $EC *AN &EB -AR !PR -AY *UN *UL !UG 3EP /CT .OV $EC

to order the vehicles they choose. To bridge the gap
between the inventory GM would like dealers to stock
and the inventory the dealers do stock, we developed
the inventory-balancing dealer-ordering tool. The term
inventory balancing reflects the advantages of carrying
a balanced portfolio of configurations, which aligns to
demand, instead of stocking only the top-selling full
configurations. It provides an organized view of sales,
inventory, turn rate, and price information to help the
dealers better understand the demand. By providing
dealers with a portfolio view of demand, which is con-
sistent with customer market segmentation, inventory
balancing can improve customer satisfaction (by reduc-
ing the gap between the customer’s ideal configura-
tion and the configurations found on the dealer’s lot),
dealer margins (by reducing dealer trades and increas-
ing the prices customers are willing to pay by stocking
vehicles in the most desired configurations), sales vol-
ume (by reducing the number of customers who can-
not find the desired vehicle configuration on the lot),
and ordering efficiency (by reducing the time needed
to choose and configure vehicles to order).
The philosophy underlying inventory balancing is
threefold. First, it provides an accurate demand-sensing

404
signal. Aggregating sales data with other dealers aver-
ages out some of the random noise that an individ-
ual dealer experiences, thus offering a clearer view
of demand. Dealers who rely solely on their own
sales data risk being fooled by randomness. We dis-
play the aggregate inventory and sales from deal-
ers in a geography that the dealer chooses for com-
parison. Second, it applies the Pareto principle for
market segmentation from a customer’s perspective.
Consistent with the bottom-up aggregate inventory
optimization, inventory balancing uses partial config-
urations based on the most important features that
are most important in defining a segment of demand.
Third, it promotes an inventory-portfolio approach to
cover all demand segments. It applies a practical set-
covering solution, which follows the hierarchy dis-
played on third-party vehicle-configuration websites
and which dealers understand immediately. Follow-
ing this industry-standard product definition avoids
the major challenge of weighting each option to define
nearness in a clustering approach. Finally, this high-
level categorization allows the dealers to use their judg-
ment and understanding of the local market to refine a
partial configuration into a fully configured vehicle.
Figure 6 is a conceptual map of sales of a vehi-
cle model’s various configurations projected onto two
dimensions. Each point represents a unique config-
uration. We can interpret the nearness between any
two points as the magnitude of the customer’s per-
ceived importance-weighted difference between those
configurations. The numbers indicate the number of
observed sales for each configuration.
Blank spaces represent configurations without sales.
Because most customers are willing to compromise on
their ideal configuration, there exists a region of rel-
ative indifference within which a customer would be
reasonably satisfied. In this example, if a dealer stocked
only the five top-selling complete configurations, as
some sales-ranking approaches advocate, that dealer
would cover the demand within the five circles shown
in Figure 6(a). The five circles are centered on the
five top-selling full configurations (i.e., the five largest
numbers in the figure). Based on the observed sales,
stocking these five configurations would provide over-
lapping coverage; the dealer would be covering many
demand points twice. That dealer could improve sales
and increase profits with a portfolio strategy that cov-
ers more demand.
Inman et al.: Optimizing New-Vehicle Inventory at General Motors
Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS
Instead of stocking the top-five full configurations,
Figure 6(b) displays a near-optimal set-covering strat-
egy that covers much more of the demand space.
Unfortunately, in practice we are not likely to clearly
understand all customers’ perceptions of the relative
importance and the relative nearness of all option com-
binations. A more sophisticated set-covering algorithm
would be based on approximate estimates of the near-
ness between two configurations in a customer’s mind,
and its solution accuracy would be limited by these
approximate underlying estimates.
A practical way to achieve a diversified portfolio is
to choose a configuration from each partial configura-
tion. Conceptually similar to stratification in sampling
or clinical trials, this segmentation subdivides demand
among mutually exclusive, collectively exhaustive sub-
sets. Figure 6(c) shows five full configurations, each
belonging to a different partial configuration, which
would cover demand. Using partial configurations
may not be theoretically optimal, but it is simpler and
more transparent, and provides reasonable set cover-
ing that is amenable to a real-world implementation.
Inventory balancing displays a vehicle’s demand and
inventory in mutually exclusive, collectively exhaus-
tive subsets designed to reflect the vehicle’s segmen-
tation in the customer’s mind. The combination of
selections of vehicle features that the customer con-
siders to be most important defines the partial con-
figuration subsets. Many dealers use a price-ladder
approach to stocking. Sometimes referred to as a stair
step, the price ladder is a series of progressively more
expensive configurations. In so doing, dealers cover
demand over one dimension—price. Although price
is a surrogate for content, it less accurately segments
demand because a price ladder could be constructed
that does not distinguish which options are important
to the customer. For example, many vehicles can be
configured with either two-wheel drive (2WD) or all-
wheel drive (AWD). Although AWD is more expen-
sive, it is possible to add other features to a 2WD
vehicle such that it is more expensive than the AWD
vehicle. Hence, using a simple price ladder to seg-
ment demand might allow either a 2WD or AWD
vehicle in each segment, thus missing important infor-
mation regarding customer demand for 2WD versus
AWD vehicles. Defining partial configurations based
on attributes important to our customers allows us to
build a portfolio based on content, not price alone.
Inman et al.: Optimizing New-Vehicle Inventory at General Motors
Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS 405

Figure 6. (Color online) A Portfolio Typically Covers More Demand Than the Top-Selling Full Configurations
(a) The top-selling configurations are often very similar and cover only a narrow slice of demand

1 4 1
9 2 1
7 5
5 2 6 5 1 3
4 1
1 1
1
1 1
2 3
1
2 8 1 21
2 1 7 2 5
1 2
3 4
2 3
1
3 1
2 1 3
1 1 1 1
1
1 3 1
1
1 3
1

(b) Choosing configurations with a set-covering approach provides wider choice

1 4 1
9 2 1
7 5
5 2 6 5 1 3
4 1
1 1
1
1 1
2 3
1
2 8 1 21
2 1 7 2 5
1 2
3 4
2 3
1
3 1
2 1 3
1 1 1 1
1
1 3 1
1
1 3
1

(c) The product-ordering hierarchy provides a stratification that approximates that of the set-covering approach
Body style
1LS 1LT 2LT 1LZ
Major option

1 4 1
9 2 1
7 5
5 2 6 5 1 3
2 wheel drive 4 1
1 1
1
1 1
2 3
1
2 8 1 21
2 1 7 2 5
1 2
3 4
2 3
1
3 1
2 1 3
All wheel drive 1 1 1 1
1
1 3 1
1
1 3
1

Inventory-Balancing Tool split (i.e., divide the population based on whether that
Figure 7 displays the disaggregation tree view corre- option is present) on the few most important factors.
sponding to the set covering in Figure 6(c). For each These “splitters” segment the market. In this exam-
product, we rank the options by importance and only ple, the most important product attribute is body style
 Inman et al.: Optimizing New-Vehicle Inventory at General Motors
406 Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS

Figure 7. (Color online) The Inventory-Balancing Tool Compares Each Dealer to a Larger Geography on Inventory and Sales
Statistics by Partial-Configuration Subset

-ODEL %15)./8

3TYLE ,3 ,4 ,4 ,:

-AJOROPTION 7$ !7$ 7$ !7$ 7$ !7$ 7$ !7$

0ENETRATIONSINPARTCONFI 
/THEROPTIONS
•
•
•
#OLORSANDINTERIORS
•
•
•

#OMPARATIVEGEOGRAPHY
3ALESPENETRATIONS
)NVENTORYPENETRATIONS
%XPOSURE TO SALES

$EALER S
3ALESPENETRATIONS
)NVENTORYPENETRATIONS
%XPOSURE TO SALES

"ALANCEDINVENTORY

2ECOMMENDEDORDERS

Note. It identifies gaps in the dealer’s portfolio and recommends partial configurations to order.

(e.g., 1LS, 1LT, 2LT, 1LZ) and the second most impor-
tant attribute is the wheel drive, which distinguishes
between 2WD and AWD. The inverted tree’s branches
represent partial configurations. Each partial configu-
ration includes a variety of colors and less important
options, which the dealer specifies based on the pen-
etrations reported by partial configuration, together
with the dealer’s judgment, local-market experience,
and consideration of other models. The dealers’ prof-
itability depends on their choices; few would cede their
decision-making authority to an automaker’s algo-
rithm they do not understand.
We provide an online report, which follows the
template in Figure 7, for each vehicle model at each
dealer. This data can be downloaded to allow dealers
to further explore the statistics or link to other pro-
grams. The information is displayed compactly; the
most important data fit on a single sheet. Each col-
umn presents data for a partial configuration and,
taken together, they cover 100 percent of the model’s
sales. The columns are sorted in increasing price order
to align with the price-ladder framework with which
dealers are familiar. The top three rows of Figure 7
define the partial configuration, and the data beneath
each column refer to that partial configuration.
“Penetrations” in the partial-configuration (Part.
Config.) section displays the sales penetration of the
other options, colors, and interior trims within that
partial configuration. Each row in this section refers to
an option (e.g., sunroof) or a color (e.g., white exterior
color). The cells contain the percentage of that row’s
sales or inventory observed in that column’s partial
configuration. For example, the row that shows white
exterior color might report that 25 percent of 1LS-2WD
Inman et al.: Optimizing New-Vehicle Inventory at General Motors
Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS
sales are white, 15 percent of 1LS-AWD are white, and
so forth. The percentages across the row would gen-
erally not total one. Mutually exclusive options are
grouped together to allow the dealer to easily see the
mutual exclusivity. For categories whose elements are
mutually exclusive and collectively exhaustive, such as
exterior color, the percentages of the various elements
would sum to one vertically within the section, because
each vehicle must have one and only one exterior color.
The option penetrations of vehicles sold are reported
so that the dealer can see the option and color pop-
ularity for each partial configuration. When a dealer
decides to order a vehicle in a specific partial configura-
tion, he or she uses the option penetrations to configure
it based on the options that are selling. For example,
assume that white is the most popular color for the first
(1LS-2WD) partial configuration. If the dealer does not
have a white 1LS-2WD partial configuration in stock,
including a white vehicle in the next 1LS-2WD order
would be advisable. The fifth and sixth blocks report
inventory and sales penetrations and exposure to sales
for each partial configuration for the aggregate geog-
raphy and the dealer. The dealer compares his or her
inventory and sales penetrations with an aggregation
of other dealers. That dealer chooses the most relevant
aggregate geographic area (e.g., country, state, or local
district). The row titled exposure-to-sales reports the
number of days each vehicle sat on the lot before it sold
(plus the number of days the unsold vehicles sat on
the lot, summed over all vehicles, and divided by the
number of vehicles that have sold).
The balanced-inventory row displays the dealer’s in-
ventory rearranged so that the relative frequency of
inventory among partial configurations most closely
matches the relative frequency of the sales in the com-
parative geographic area. Whenever a dealer’s bal-
anced inventory exceeds his or her actual inventory
by a threshold amount, those inventory numbers are
highlighted to immediately suggest an opportunity for
the dealer to fill a gap in his or her inventory portfo-
lio. The inventory-balancing tool can also recommend
the partial configurations that the dealer should order.
The dealer enters the number of vehicles of a given
model to order, and the embedded optimization allo-
cates that number of orders among the partial config-
urations to align with the demand in the comparative
407
geographic area. The tool also allows the dealer to con-
strain or override the solution by forcing any number of
orders as input for selected partial configurations. This
allows the dealer to apply knowledge of impending
sales, customer inquiries, financial constraints, used-
vehicle inventory, and local-market trends. In addition
to new-vehicle ordering, many dealers also use the tool
when another dealer approaches them for a trade; the
tool allows the dealer to specify what he or she would
like in return. The inventory-balancing tool helps GM’s
dealer network stock the right mix of vehicles to make
the most efficient use of inventory.
Summary and Impact
GM developed two operations research applications
for optimizing the number and the mix of vehicles in
dealers’ inventories. Both applications rely on the con-
cept of prioritized partial configurations to represent
granular product variety. We apply the Pareto principle
to define prioritized partial configurations to segment
demand for a given model based on vehicle options
most meaningful to customers. Partial configurations
simplify the analysis by capturing the majority of cus-
tomer choices without resorting to a combinatorial
explosion of variety by recognizing that some options
matter much more than others. Partial configurations
are intuitive to dealers and easily communicated.
Instead of determining the inventory needed for a
given service level, we find each model’s profit-max-
imization inventory by trading off the variable profit
of sales with the carrying cost. While the automaker’s
inventory carrying cost is typically assumed to be zero,
our approach incorporates the broader supply chain’s
carrying cost and the observed increase in incen-
tives over the model year. The bottom-up optimization
includes rich details of price elasticity, diversions to
other products, lead-time distributions, demand dis-
tributions, and a model for vehicle trading between
dealers. The dealer inventory-balancing decision sup-
port tool marries a set-covering philosophy with the
dealers’ practical product-specification framework. It
promotes a portfolio viewpoint to counter the indus-
try’s one-dimensional price-ladder standard.
These innovations are in full use today and have
been extremely well received by General Motors and
its dealers, as the following dealer endorsements exem-
plify. For confidentiality, we do not disclose the names
of dealers who provided these endorsements.

408
“I wanted to thank you for the inventory balancing reports
and congratulate you on a[n] excellent execution. Those sheets
are the correct execution of Demand Sensing. I think that that
product could make dealers more capable of having the right
product on the ground.”
“Thank you for your excellent support on this, it has helped
us improve our inventory and our sales are more efficient due
to fewer dealer trades . . . . Again, thanks for your efforts on
this, it has a payoff.”
“I just wanted to say thank you for these overviews. As I get
deeper into my order cycle for inventory fill-ins, these reports
have really been useful.”
“It has helped me . . . . We appreciate that you are doing
this. It’s a very useful tool . . . . It’s making things better.”
These tools have significantly improved GM’s sales.
More than 800 dealer franchises piloted the inventory-
balancing tool for six months; these dealers averaged a
three to five percent lift in sales and revenue compared
to a control group of more than 7,000 franchises. The
tools have also helped to reduce GM’s retail inventory.
GM has historically held higher retail inventory than
its competitors. However, with the help of this appli-
cation, its 2015 year-end inventory was 61 days-supply,
which is substantially lower than Ford’s 79 days-supply
and Fiat Chrysler’s 81 days-supply. This innovation
enabled GM’s 2015 earnings report (General Motors
2016) to claim that the company had “actively managed
down U.S. dealer inventories” with “dealer inventories
down 100 K units or 14% at year-end 2015 versus year
end 2014” (p. 30). Finally, these tools have contributed
to GM’s significant financial gains. GM President Mark
Reuss points out that since 2008, GM increased its aver-
age transaction prices at an annual rate of 4.1 versus
2.85 percent for the industry (Reuss 2013).
Acknowledgments
The authors acknowledge the invaluable contributions of
David Vander Veen. His leadership was essential to the full
realization of the benefits of this work. They thank project
sponsors Jim Bunnell, Steve Carlisle, Brent Dewar, and Susan
Docherty. They also thank Betty Borgula and Krisztina M.
Gilezan for integrating the solution into the business process,
Craig Jackson for providing revenue-management expertise,
Kenn Griessel for his help in implementing both the bottom-
up and inventory-balancing tools, Clifford Hodges for his
leadership in implementing the inventory-balancing tool,
and Jason Brickl for explaining the viewpoint of the dealer
network. The authors also acknowledge the contributions
of Paul Beckett, Bob Glubzinski, Colleen Haesler, Bill Mis-
tele, Benoit Schlumberger, Todd Scott, Janet Shearer, and Dan
Smith. Finally, the authors are indebted to Dennis Blumen-
feld for many useful discussions.
Inman et al.: Optimizing New-Vehicle Inventory at General Motors
Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS
Appendix. Mathematical Formulations
U.S. automakers’ retail inventory and ordering are typi-
cally periodic review (S, TC ) inventory policies where dealers
order every TC days and order up to S to replenish their lots.
Refer to Sipper and Bulfin (1997), Blumenfeld et al. (1999),
and Blumenfeld (2010) for discussions of periodic-review
inventory systems. The objective function in Equation (A.1)
is the expected profit from sales (the first line of the objective
function) less the inventory carrying cost for model m. The
decision variable k is the number of lead-time demand stan-
dard deviations to hold in safety stock. Equation (A.2) rep-
resents the average inventory for a given k, and depends on
the standard deviation of lead-time demand, which is given
by Equation (A.3). Equations (A.4)–(A.7) describe how to cal-
culate the effective fill rate β 4 (i, j, k, m) used in the objec-
tive function. Equation (A.4) presents the fill-rate formula
for the typical periodic-review policy previously described.
Equation (A.8) documents the target inventory correspond-
ing to the optimal k ∗ . Equation (A.9) sums the inventories
(adjusted to comprehend skewed and auto-correlated lead-
time demand) of all the configurations at all the dealers to
arrive at the national target inventory. The solution proce-
dure searches over k.
X X
max D y S(i, j, m)[β 4 (i, j, k, m)MS (j, m)
k
i∈ψ(m) j∈η(m)
+ (1 − β4 (i, j, k, m))β 4 (i, j, k, m)MA (j, m)]
− Ī(i, j, k, m)c(j, m)h
subject to k ≥ 0 (A.1)
where the average inventory is
Ī(i, j, k, m)  TD + 12 TC S(i, j, m)


+ kσLTD (i, j, m), (A.2)
and the standard deviation of lead-time demand is
σLTD (i, j, m)
p
 (TT + TC )S(i, j, m)v + S(i, j, m)2 σLT (i, j, m)2 . (A.3)
If we assumed that demand (i.e., sales) is Poisson distributed,
then the exponent ν (a measure of demand variability) on
average sales in Equation (A.3) would be 1. Instead we esti-
mate ν empirically. The fill-rate expressions are
β 1 (i, j, k, m) standard fill rate for dealer i, configuration j
of model m, and decision variable k
σLTD (i, j, m)
 
1 −(1/2)k 2
1− √ e −k[1−Φ(k)] ;
S(i, j, m)(TT +TC ) 2π
(A.4)
β2 (i, j, k, m) fill rate with supply chain cost impacts
 β1 (i, j, k, m)(1−φ a (i, j, k, m)); (A.5)
β 3 (i, j, k, m) fill rate with impact of inventory cost and
dealer trades
β 2 (i, j, k, m)+(1−β2 (i, j, k, m))φ t (i, j, k, m); (A.6)
Inman et al.: Optimizing New-Vehicle Inventory at General Motors
Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS
β 4 (i, j, k, m) fill rate with impact of inventory cost, dealer
trades, and build-to-order sales
β 3 (i, j, k, m)+(1−β3 (i, j, k, m))φ b (i, j, k, m); (A.7)
and the fill-rate adjustments are
φ a (i, j, k, m)  fraction of sales lost to supply chain cost
impacts;
φ t (i, j, k, m)  likelihood of completing a trade when
looking for one;
φ b (i, j, k, m)  likelihood of completing a build-to-order sale
when a suitable vehicle is neither in inventory
nor available through a cost-effective trade.
The optimal average inventory is obtained by plugging the
optimal fill rate k ∗ into Equation (A.2), yielding
Ī(i, j, k ∗ , m)  TD + 12 TC S(i, j, m) + k ∗ σLTD (i, j, m).


(A.8)
This formulation assumes that lead times are normally dis-
tributed and independent. However, we adjust the above
model’s optimal inventory because lead times are positively
skewed and auto correlated. The multiplicative adjustment
factor is found via simulation. The optimal national retail
inventory for model m is the sum over all configurations and
dealers. X X
Î ∗ (m)  γs (Ī(i, j, k ∗ , m)). (A.9)
i∈ψ(m) j∈η(m)
Notation
i dealer index;
j prioritized partial configuration (of model m);
k decision variable representing the number of
lead-time demand standard deviations to hold
in safety stock;
m model index;
h holding-cost factor;
c(j, m) cost of partial configuration j of model m;
D y days per year;
MS (j, m) standard variable profit for configuration j of
model m;
MA (j, m) variable profit for configuration j of model m
adjusted for same-automaker diversions;
P(j, m) average transaction price of configuration j of
model m;
S(i, j, m) sales per day of prioritized partial
configuration j of model m at dealer i;
σLT (i, j, m) standard deviation of the total lead time for
partial config. j of model m at dealer i;
TC cycle time between ordering opportunities;
TT total time from order to delivery;
TD delivery time from plant to dealer (in-transit
vehicles are counted as dealer inventory);
v factor representing empirical demand
variability as a function of mean demand;
409
γs factor that adjusts for the impact of lead time’s
skewness and autocorrelation;
η(m) set of partial configurations for model m;
ψ(m) set of dealers selling model m;
Φ( ) standard Normal distribution.
References
Blumenfeld DE (2010) Operations Research Calculations Handbook (CRC
Press, Boca Raton, FL).
Blumenfeld DE, Daganzo CF, Frick MC, Gonsalvez DJA (1999)
Impact of manufacturing response time on retailer inventory.
Internat. J. Oper. Production Management 19(8):797–811.
Copeland A (2009) The dynamics of automobile expenditures.
Federal Reserve Bank of New York Staff Report 394, New
York. Accessed January 16, 2017, https://www.newyorkfed.org/
medialibrary/media/research/staff_ reports/sr394.pdf.
Copeland A, Dunn W, Hall G (2005) Prices, production and inven-
tories over the automotive model year. Working paper, National
Bureau of Economic Research, Cambridge, MA. Accessed Jan-
uary 16, 2017, http://www.nber.org/papers/w11257.pdf.
Copeland A, Dunn W, Hall G (2011) Inventories and automobile
market. RAND J. Econom. 42(1):121–149.
Corrado C, Dunn W, Otoo M (2006) Incentives and prices for motor
vehicles: What has been happening in recent years? Finance and
Economics Discussion Series, Divisions of Research & Statistics
and Monetary Affairs, Federal Reserve Board, Washington, DC.
Accessed January 16, 2017, https://www.federalreserve.gov/
pubs/feds/2006/200609/200609pap.pdf.
General Motors (2016) CY 2015 results. Accessed February 22, 2017,
https://www.gm.com/content/dam/gm/mol/docs/GM-2015
-Q4-Chart-Set.pdf.
Iyer A, Seshadir S, Vasher R (2009) Toyota Supply Chain Management
(McGraw-Hill, New York).
Reuss M (2013) Mark Reuss remarks to Automotive News World
Congress. Accessed December 11, 2016, http://media.gm.com/
media/us/en/gm/news.detail.html/content/Pages/news/us/
en/2013/Jan/0116_anwc-reuss.html.
Sipper D, Bulfin RL (1997) Production: Planning, Control, and Integra-
tion (McGraw-Hill, New York).
Vellequette LP (2016) Why Americans reject build-to-order cars.
Automotive News (June 6), http://www.autonews.com/article/
20160606/RETAIL/306069951/why-americans-reject-build-to
-order-cars.
Robert R. Inman is a technical fellow in General Motors’
Operations Research group. He was named an INFORMS
Fellow in 2014 for advancing the application of operations
research at GM through broad and high-impact contribu-
tions to applied supply chain, manufacturing, vehicle sales,
service, and dealer inventory management. He earned his
PhD in industrial engineering and management sciences at
Northwestern University and joined General Motors in 1989.
Michael C. Frick retired from General Motors in 2009 after
working the majority of his career as a staff research scientist
in GM Research. His primary research areas included traffic
safety, production control, and logistics. Since his retirement,
he has worked as a consultant for the Vehicle Order Manage-
ment group at GM.
Thomas D. Hitchman is the Director of Forecasting and
Market Analysis for General Motors. He joined GM in 1984
and has held leadership roles in sales, marketing, brand

410
development, global product research, and new business
development. In his current role, Tom leads the forecasting
group and the process for establishing new vehicle produc-
tion to align supply and demand. Tom earned a BS degree
in industrial and operations engineering and an MBA, both
from the University of Michigan, and has held Professional
Researcher Certification (PRC) since 2007.
Robert A. Muiter is GM’s North America Director of
Order Fulfillment. His responsibilities include overseeing a
series of time sensitive (annual, monthly, and weekly) end-to-
end business processes and decisions related to volume plan-
ning, retail and fleet allocation, material forecasting, order
management, Vehicle Order Management (VOM), Produc-
tion Order Management (POM), vehicle specifications (mar-
ket and engineering intent), order processing, order schedul-
ing and sequencing for North America and Joint Venture
assembly centers. Bob has a bachelor’s degree from Eastern
Michigan University and completed an executive education
program from the Columbia School of Business.
Jonathan H. Owen is Chief Scientist of Operations Re-
search and Advanced Analytics at General Motors and the
Inman et al.: Optimizing New-Vehicle Inventory at General Motors
Interfaces, 2017, vol. 47, no. 5, pp. 396–410, © 2017 INFORMS
Director of the GM R&D Operations Research (OR) Lab. An
18-year GM veteran, Jon leads strategic innovation activi-
ties for applied OR across diverse areas of the business that
include pricing and revenue management, portfolio plan-
ning, vehicle technology selection and content optimization,
supply chain and logistics, market demand modeling, and
dealer effectiveness. Jon earned his PhD in IE/MS at North-
western University and graduated from the General Man-
agement Program at Harvard Business School. Since 2014,
Jon has served on the Analytics Certification Board and as
the INFORMS VP of Practice. He also represents GM on the
INFORMS Roundtable and is a member of the IE/MS Advi-
sory Board at Northwestern University.
Gerald M. Takasaki worked as a staff research engineer
within the GM Research and Development Center and within
GM’s supply-chain organization. He worked on developing
systems and processes for demand sensing, production con-
trol and scheduling, and inventory management. He joined
General Motors after earning a PhD degree in electrical engi-
neering at the Ohio State University in 1978, and retired from
GM in 2007.