An SQL-based K.

Katircioglu
T. M. Brown
cost-effective inventory M. Asghar

optimization solution
Successful implementation of an inventory optimization solution
requires significant effort and can pose certain risks to companies
implementing such solutions. Depending on the complexity of the
requirements, the solution may also involve a substantial IT
investment. In this paper, we present a cost-effective solution for
inventory optimization that can be useful for small and medium-
sized businesses with limited IT budgets. This solution can be
implemented on any application platform that is capable of
processing basic SQLe (Structured Query Language) commands.
The solution eliminates the need to purchase additional software
and has a framework in which sales data in an Enterprise Resource
Planning (ERP) system are accessed, demand statistics based on
this data are generated along with other key parameters, and
optimal inventory policies, such as those involving safety stocks
and lot sizes, are calculated and reported.

Introduction On the other hand, no matter how well the planning

One of the most tangible ways in which companies can processes are executed or how advanced they may be,
benefit from IT investments is through implementing some uncertainty almost always remains in a typical
solutions that are designed to provide inventory reduction supply chain on both the demand side and the supply
opportunities. Companies can reduce their inventory side. Companies continue to have a need for inventory
through more accurate and faster receipt of demand policies that are appropriate for their supply chain
information, more reliable and shorter supply lead times, environment and protect them from the serious
and better inventory policies. consequences of these uncertainties. For instance, a good
Faster receipt of demand information and shorter lead inventory policy takes into account demand variation,
times can be achieved by changing planning processes, supply lead times, and the service objective of a company
modifying supply chain infrastructure, and adopting and calculates the size of safety stocks that can provide
practices that enable rapid flow of information on both the targeted service level with the minimum possible
the demand and supply sides. Supply chain management inventory.
(SCM) solutions are designed to support these activities. Inventory management is a well-established field of
(The term lead time refers to the amount of time between research in both the academic and the industrial worlds,
the placing of an order and the receipt of the goods and numerous publications and popular books are
ordered.) Demand-planning solutions can also help available [1–8]. Analysis of Inventory Systems is an
increase the accuracy of demand forecasts. early classic and a valuable source [9].
Improved reliability of supply arrival times also has an Numerous academic papers have made significant
impact on inventory levels. Because companies must keep contributions to the field. While it is not possible to
a safety stock to protect against variations in either include all without providing an extensive survey, we can
supply quantity or supply arrival times, more reliable mention [10–16] as examples of classic publications in the
supply arrival times can reduce these safety stocks. (The field, and [17–19] as examples of later publications that
term safety stock refers to the extra units of inventory have made important contributions.
carried by a company as protection against possible stock Optimized inventory policies are the most important
outages.) Clearly, shorter supply lead times also reduce output provided by inventory optimization software.
the need for safety stock. Such policies are often characterized by a few

ÓCopyright 2007 by International Business Machines Corporation. Copying in printed form for private use is permitted without payment of royalty provided that (1) each
reproduction is done without alteration and (2) the Journal reference and IBM copyright notice are included on the first page. The title and abstract, but no other portions, of this
paper may be copied or distributed royalty free without further permission by computer-based and other information-service systems. Permission to republish any other portion of
this paper must be obtained from the Editor. 433
0018-8646/07/$5.00 ª 2007 IBM

IBM J. RES. & DEV. VOL. 51 NO. 3/4 MAY/JULY 2007 K. KATIRCIOGLU ET AL.
parameters such as safety stock and lot size. Optimal evolve and the applications are therefore required to
inventory policies can reduce inventory investment and support these changes, they may be inadequate or may
improve customer service significantly. Hence, most require substantial modifications to deliver the required
of the time, a very attractive business case exists for functionality. If the process changes are so extensive that
purchasing and implementing such inventory software. they affect a number of applications and the way they
Of course, companies that do not have the execution must integrate, it may be very expensive to perform
processes in place for proper implementation of these the necessary changes in these applications. Realizing
policies cannot benefit from them. Along with other that these costs can be very high, many companies
basics, execution typically requires the visibility of understandably resist additions to their systems portfolio
different kinds of inventories (such as materials on hand, because they do not know what changes will be required
on order, committed, or in transit) across the supply in the future.
chain, as well as the visibility of demand information such Cost of implementation: Standalone solutions must
as forecasts and sales orders. Most Enterprise Resource address issues such as Internet access infrastructure,
Planning (ERP) systems or SCM systems are capable of security, and system integration. Such concerns lead to an
providing this visibility, although fragmented systems— unavoidable infrastructure cost that has to be incurred
for example, in the form of IT systems with little in order to realize the benefit of the solution, which
integration or with different applications—can lead to facilitates inventory reduction and/or customer service
data-integrity problems, such as data with missing fields improvements. To the extent that a solution can take
or with different definitions. advantage of existing infrastructure in addressing these
As is the case for any other software solution, some issues, it will be more attractive from an implementation
concerns may exist with respect to using standalone and maintenance cost point of view.
software to optimize inventory. (Note that when we use Systems integration: Since inventory optimization has
the term standalone software, we refer to software that extensive data requirements, integration into existing
is designed to be self-sufficient in terms of performing a ERP and SCM systems can be time-consuming and
process or a part of the process, and that usually includes costly. However, robust and flexible data-extraction
such basic elements as a database, capability to store and templates can significantly reduce system integration
manipulate inputs and outputs, an engine that performs costs. Note that when major version updates occur,
the calculations, and a graphical user interface.) If many version compatibility can become problematic while
elements of an inventory management process are maintaining this integration. Complex forms of
missing, a standalone software package may be the right integration are clearly undesirable because business
choice for a company, because such a package can help changes may also require modifications in the way
establish and facilitate the entire process. However, if applications must integrate.
many of the elements already exist, and a company has to Purchase price, maintenance and user fees: Standalone
focus on optimizing safety stocks or lot sizes, a quick and software typically requires an up-front purchase cost,
inexpensive solution like ours can be a viable choice. Our annual fees based on the number of users, and
solution is designed as a part of the ERP system and maintenance contracts. Version upgrades add to these
hence does not require complex integration. Typical costs, as does training and/or hiring new staff who can
concerns associated with standalone software use are use the software effectively.
outlined in the following paragraphs.
Software proliferation: In order to gain full benefit of Cost-effective inventory optimization solution
a well-planned and well-executed supply chain process, Most of the up-front effort, and therefore cost, incurred
companies often find themselves in a situation in which with inventory optimization software comes from data
they must buy several layers of applications. Typically, and systems integration. Although the algorithmic part
their ERP system may be considered to be the bottom of a solution that performs the inventory optimization
layer. Systems or applications may be integrated into provides value, this is relatively inexpensive because many
ERP. These include applications for distribution algorithms with a range of features are readily available
requirements planning (DRP), materials requirements from various sources such as textbooks, handbooks, and
planning (MRP), demand planning, SCM, order academic publications. Some very specialized software
fulfillment, procurement planning, and inventory- companies or companies that have strong research
optimization systems. These applications, which may departments can provide proprietary techniques that
have complex integrations with one another, present a are not publicly available.
serious IT challenge to companies that are concerned The approach used by most standalone inventory
about cost-effective management and maintenance of optimization software packages may be described as
434 their IT portfolios. If major business process changes follows. First, the package obtains an extract of data

K. KATIRCIOGLU ET AL. IBM J. RES. & DEV. VOL. 51 NO. 3/4 MAY/JULY 2007
from an ERP system (typically via a flat file, containing
Transaction data Web user access
records with no structured relationships, or a Microsoft
Excel** file). The data is imported to a software package ERP system
and converted to a form that an optimization engine can
use. The calculations are carried out using a language
such as C or Cþþ. The output is placed in a data
InfoCubes
repository that can be viewed by the user and sent back to Database reports
for data Order Part Order Create Requested Order

the ERP system or an Advanced Planning and Scheduling processing

No
1
2
Number
PN0001
PN0001
Date
1/2/02
12/31/01
Delivery Date
1/3/02
1/3/02
Quantity
20
20
3 PN0001 12/17/01 1/3/02 80

(APS) system. 4
5
6
7
PN0001
PN0001
PN0001
PN0001
12/31/01
12/20/01
12/31/01
1/2/02
1/3/02
1/3/02
1/3/02
1/3/02
20
20
20
20
8 PN0001 1/1/02 1/3/02 40
In many cases that are not complex, relatively simple 9 PN0001 1/2/02 1/3/02 20

algorithms may provide satisfactory solutions. In fact, the Library of inventory

solution-finding process of some common inventory optimization algorithms
problems can be simplified to the extent that these
solutions can be calculated using simple mathematical
functions available in Structured Query Language (SQL). Figure 1
SQL is a language that enables access to a database, Simple approach for inventory optimization using SQL. Arrows
performs queries with flexible conditions, and facilitates show the directions of information flows. Green double arrows
basic data processing operations using commands such represent the closely integrated InfoCubes and algorithms. All
as retrieve, insert, delete, and update. Although many calculations are performed using query commands, which are
closely tied to the data in the InfoCubes.
versions of SQL exist, basic commands in compliance
with American National Standards Institute (ANSI)
standards are supported by all providers. SQL is
commonly available in existing database software offered provider.] After the ETL step, we can calculate optimal
by various vendors and widely used by organizations for inventory policies using query commands in runtime.
managing large sets of data. An accessible introduction Database management systems (DBMSs) are designed
to SQL can be found in [20]. to maintain data and answer queries that are typically
Although the wide availability of SQL is an advantage, executed through SQL. Queries can be real-time
it does have limitations. For tasks such as inventory operations, and they allow real-time interaction with the
optimization, SQL is not the appropriate tool, since it is users. However, this interaction is mainly for the purpose
not designed for programming and executing advanced of reporting basic statistics about the data, such as total
algorithms during the runtime of an application. sales in a period or total inventory for a product group or
However, we can simplify some common inventory a geographical region. Complex operations, such as
optimization problems in order to perform the optimal algorithms used to optimize inventory policies, normally
safety stock and lot size (that is, economic order quantity) cannot be performed by queries. Although programming
calculations using basic mathematical functions available languages such as C, Cþþ, and Java** are certainly capable
of performing such operations, they cannot be parts of
in SQL. (The term economic order quantity refers to a
query commands that must process significant amounts of
quantity that minimizes the fixed inventory replenishment
input data in runtime. Java-based programs may be able to
costs and inventory carrying costs.) Coupled with readily
perform calculations with acceptable speed. However,
available data-extraction templates in many database
speed may be lost in open database connectivity (ODBC)
software packages, our approach makes it possible to
and Java database connectivity (JDBC) layers that are
create simple inventory optimization capabilities for
necessary for a Java inplementation. For runtime
databases. efficiency, these calculations must be performed outside
To take advantage of SQL, we attempted to make our the query commands in batch mode. Obtaining results this
approach work seamlessly for systems with databases way requires a few steps, as we described earlier, and the
that have strong capabilities for extracting and real-time nature of the interaction is lost. In addition, file
manipulating data from transaction systems. Figure 1 downloads from and uploads to databases can be time-
schematically illustrates how our approach works. Using consuming. The ability to perform optimization in the
available data extractors, we extract, transform, and load queries provides users with interaction capabilities and
(ETL) essential data such as sales transactions and lead enables them to do real-time ‘‘what-if ’’ analyses in
times into multidimensional summary data tables called addition to avoiding file downloads and uploads. Most
InfoCubes in the SAP** literature. [SAP, which stands for significantly, the optimization speed can be acceptable.
systems, applications, and products in data processing, is In a solution that we designed on a common ERP
a business application and an ERP solution software platform, we were able to run queries in order to calculate 435

IBM J. RES. & DEV. VOL. 51 NO. 3/4 MAY/JULY 2007 K. KATIRCIOGLU ET AL.
we needed no additional application development in
200,000 order to display the results of queries in the form of charts
Inventory position
and tables. A Java application typically requires the
150,000
creation of an HMTL document in order to communicate
the results over the Web.
100,000
We next present some classes of inventory problems
and show how their solutions can be simplified and
Units

50,000
implemented in SQL.
0
Some common inventory problems and their
⫺50,000 Inventory level solutions
Although many inventory optimization problems may
⫺100,000
Time (days) require relatively complex algorithms, some classes
of inventory problems can be simplified using
Safety Reorder Maximum
stock point on-hand inventory approximations. We focus on simple inventory policies
in which a safety stock and lot size characterize the
inventory policy. Figure 2 shows how a simple inventory
Figure 2 policy works. As discussed previously, safety stock is the
Simple inventory policy in action. When the inventory position stock that is held in order to protect against uncertainties
decreases to the reorder point, a replenishment order is placed. in the process, such as uncertainty in demand and
The replenishment order has a calculated lot size, and until the uncertainty in supply lead time. When the inventory
order arrives, inventory shortages are prevented through the use of
safety stocks. The inventory that is on hand can never exceed the
position (defined as the inventory on hand plus inventory
maximum on-hand inventory. (Inventory level: on-hand inventory on order minus the back orders) drops to the reorder
minus backlogs. Backlogs: demand that is waiting to be satisfied point, inventory is replenished at an amount equal to the
because of lack of inventory.) lot size. Most often, this replenishment is not instant (i.e.,
orders arrive after a lead time).
Next, we study different types of common inventory
optimal inventory policies and more than 30 related problems that have simple solutions. Before we proceed,
we need to mention that the calculations of optimal
parameters in 25 seconds for ten products with a total of
inventory policies that we present here are based on the
10,000 customer order transactions constituting demand
following assumptions.
data. We used an IBM iSeries* server with an OS/400*
Our model assumes that inventory is held at a single
operating system running on four processors. The data
location. This model can also be used to approximate
was stored in a DB2* database. Our solution design had
multiple-location models. A single-location model is
several advantages, as listed in the following paragraphs.
one in which inventory is held at a single location.
First, we note that our SQL-based approach was
Each SKU has independent and identically distributed
simple in that SQL commands are communicated to the
demand in each time period. (The term SKU stands for
database layer without implementing the ODBC and
stock-keeping unit; it is an identifier that helps merchants
JDBC connections and drivers required by a Java-based track products and services.) We assume a time period
solution. Data is efficiently retrieved from an already that corresponds to a day. Customer orders arrive
established data model. This architecture also efficiently according to a Poisson distribution. The quantity
and flexibly enables characteristics-based reporting on the requested in an order is random and follows a normal
data. distribution. The quantities of all orders are independent,
Second, we note that since the ERP system we used identically distributed random variables. Demand that
also had a flexible query-reporting capability based occurs during the replenishment lead time has (or is
on a data warehouse (i.e., a primary repository of an approximated by) a normal distribution. Inventory is
organization’s historical data), no additional vendor or replenished through manufacturing or a supplier with a
software products were needed to gather data from the positive lead time. In the case in which the inventory is
back-end databases and communicate the results to the reviewed periodically, the service targets are set for the
web applications. More generally, our approach allows worst period in the replenishment cycle. Therefore, the
us to calculate additional parameters by adding their actual inventory availability levels might be higher than
formulas to the query commands without having to the target. This is a conservative approach; if some of
modify underlying programs such as Java code. By using the assumptions made in the model do not hold, this
436 the web server that is native to the available ERP system, approach could compensate for a potential shortfall in

K. KATIRCIOGLU ET AL. IBM J. RES. & DEV. VOL. 51 NO. 3/4 MAY/JULY 2007
actual service levels. Calculating optimal policies for OTDR: On-time delivery to customer request target,
average service criteria requires search algorithms. in percentages.
Although these algorithms would be simple to code in OTDC: On-time delivery to commit target, in
a language such as C, it is not possible to code loops percentages.
in SQL. Therefore, we restrict service targets to the OCOST: Fixed order cost (per order).
conservative scenario. In the case in which inventory
SCOST: Shortage cost (per unit per period).
is reviewed continuously, service targets are set to
HCOST: Inventory holding cost (per unit per period).
control the average performance during a cycle.
The objective is one of the following: 1) minimize COST: Cost of a unit (cost of manufacturing or
expected inventory costs; 2) maximize expected profit; 3) purchase).
achieve target service with minimum expected inventory; PRICE: Price of a unit.
or 4) calculate safety stock for a fixed ‘‘days-of-supply’’ SVALUE: Salvage value of a unit when season is over
policy. Days of supply is a measure of inventory level, (for seasonal items). This value is the price offered when
calculated by dividing inventory in units by the average trying to dispose of the unsold inventory, and this price
demand per day. is almost always a highly discounted one.
The service is defined by one of the following: 1) the PDECLINE: Price decline of a unit (during a season or
probability of no stock-out (i.e., the probability of period).
inventory being available to satisfy demand); 2) the QFIX: Fixed lot size given by the user.
probability of on-time shipment to meet the customer QUNIT: Unit lot size. (The output parameter LOT,
request date; 3) the probability of on-time shipment to which represents lot size, must be an integer multiple of
commit (i.e., the probability that shipment was made no this quantity.)
later than the date that was committed to the customer);
QMIN: Minimum lot size.
and 4) the fill rate (which relates to the fraction of
QMAX: Maximum lot size.
demand that can be met from inventory with no delay).
The lot-sizing policy is defined by one of the following: T: Number of days in the data time window. (Note: The
1) fixed lot size (provided by the user); 2) variable lot size length of the time window must be determined prior to
(the quantity that brings inventory position to a specified calculating T.)
level); or 3) economic order quantity (the optimal order N: Number of orders during T days (i.e., during the entire
quantity that minimizes the sum of fixed order cost and data time window).
inventory carrying cost). The inventory-replenishment NDAYS: Number of working days in a year.
process is defined by continuous review and I: Order number in the data window (I ¼ 1, 2, , N).
replenishment or periodic review and replenishment. J: Day number in the data window (J ¼ 1, 2, , T).
DOQ[J]: Daily order quantity [total order quantity in day
Notation J (J ¼ 1, 2, , T)].
In this section, we introduce the parameters that we need
ORQ[I]: Order quantity in order I (I ¼ 1, 2, , N).
in order to perform the calculations of optimal lot size
and optimal safety stock that make up an optimal TSERVICE: Target service level (percentage).
inventory policy. As we introduce these parameters, MLT: Manufacturing (or supply) cycle time or period
we indicate whether they are considered as input, length (in days) when inventory is reviewed and
intermediate calculations, or output. Some parameters replenished periodically. This parameter is also called
can be provided as inputs or can be calculated on the basis ‘‘manufacturing (or supply) frequency’’ and is the number
of their historical observations. We label such parameters of days during which supplies are regularly planned.
as input or intermediate calculations. In the lists, we PLT: Manufacturing (supply) lead time (in days). This is
follow an order of precedence; that is, inputs come first, the actual physical manufacturing time (or supplier’s
followed by intermediate calculations and outputs. turnaround time).
OPT: Order processing time (in days). This is the number
Inputs of days required to process the inventory replenishment
PNS: Inventory availability target in percentages orders from the stocking locations.
(probability of no stock-out, i.e., the event of running out TLT: Transportation time (in days). This is the number
of inventory). of days required to transport orders from the
FRT: Fill rate target in percentages, i.e., the fraction of manufacturing location to the stocking location.
demand that can be met from inventory without any IP: Inventory position. This is a key metric for the
delay. execution of any inventory policy. It must be monitored 437

IBM J. RES. & DEV. VOL. 51 NO. 3/4 MAY/JULY 2007 K. KATIRCIOGLU ET AL.
during execution and not calculated in optimization. This Outputs
requires monitoring inventory on hand, inventory on SS: Safety stock.
order, and amount of demand back-ordered. The proper ROP: Reorder point.
implementation of ROP (reorder point) requires keeping LOT: Lot size.
track of the inventory position, because replenishment
MAX: Maximum inventory level.
orders should be triggered only when the inventory
position (not the inventory on hand) drops to ROP. IP is
Throughout the calculations below, we use sqrt(X) to
given by IP ¼ inventory on hand þ inventory on order denote the square root of X, max(X, Y) to denote the
back orders. maximum of X and Y, min(X, Y) to denote the minimum
of X and Y, and ln(X) to denote the natural logarithm
Inputs or intermediary calculations of X.
SPLT: Standard deviation of manufacturing (supply)
lead time PLT. Calculation of optimal inventory policies using SQL
SOPT: Standard deviation of order processing time OPT. functions
STLT: Standard deviation of transportation time TLT. All of the calculations presented in the following are
designed in such a way that they can be coded in SQL.
CLTR: Customer order lead time (requested by Since, as discussed, query languages are not able to use
customer). loops and they have limited mathematical functions
CLTC: Customer order lead time (committed to available, we provide approximate solutions for problems
customer). that normally require looping algorithms to solve.
SCLTR: Standard deviation of customer order lead time
CLTR. Calculation of key problem parameters
First, we introduce different cases that apply to practical
Intermediary calculations inventory problems. Then, in each case, we calculate three
RLT: Replenishment lead time (in days). important parameters: PVALUE, RLT, and SRLT. Very
briefly, PVALUE represents or summarizes the objective
SRLT: Standard deviation of replenishment lead time function of the optimization problem. The inverse of
RLT. the cumulative distribution of demand during the total
ADLT: Average demand during replenishment lead time. replenishment lead time at the PVALUE gives the
SDLT: Standard deviation of demand during optimal amount of safety stock. RLT and SRLT are
replenishment lead time. respectively the mean and standard deviation of the
demand during the total replenishment lead time. We
OAR: Order arrival rate (average number of orders per
refer to these three parameters as the key problem
day). parameters.
TOQ: Sum of all order quantities.
TOQS: Sum of the squares of all order quantities. Case 1. Minimize expected inventory costs
In Case 1, the objective is to minimize the expected
TOD: Total of all daily demand quantities.
inventory holding and backlogging costs. We assume
TODS: Total of the squares of all daily demand that the demand not met immediately in a period is
quantities. backlogged to the next period. Backlogging has costs
AOQ: Average order quantity (i.e., average quantity associated with expediting orders, paying a penalty to the
demanded in a single customer order). customer, buying inventory at a high price, or loss of
customer goodwill. We use a ‘‘myopic solution’’ to this
SOQ: Standard deviation of order quantity.
cost problem. This solution tries to minimize the expected
AOD: Average demand during a day. inventory-carrying and backlogging costs one period at a
SOD: Standard deviation of the amount of quantity time (hence our use of the term myopic). In general this
ordered during a day. solution is not optimal, and the optimal policies can be
PDURATION: Duration (in days) of a season (or a complex. However, the simplicity of the method makes
period) during which expected profit is to be maximized it practical. Reference [13] provides conditions under
which myopic policies are optimal. When significant
or expected cost is to be minimized.
obsolescence costs exist that are associated with unsold
k: Safety factor used for safety stock calculation. inventory, myopic policies may be unwise to use and
PVALUE: z value for the standard normal distribution. caution is warranted. One may also study [21] for such
438 This is used for the calculation of the safety factor k. cases.

K. KATIRCIOGLU ET AL. IBM J. RES. & DEV. VOL. 51 NO. 3/4 MAY/JULY 2007
In the following, we show the calculation of three key PVALUE ¼ ðPRICE COST Þ=ðPRICE SVALUE Þ;
intermediary parameters, PVALUE, RLT, and SRLT:
RLT ¼ PDURATION;

A. If lot size is flexible and inventory is reviewed SRLT ¼ SPDURATION:

periodically, then
Case 3. Achieve a target probability of no stock-out
SCOST
PVALUE ¼ In Case 3, the probability of no stock-out is defined by
SCOST þ HCOST :
PNS (see notations above). PVALUE, RLT, and SRLT
B. If a lot size LOT exists, then are calculated as follows. First, PVALUE ¼ PNS. Note
LOT 3 HCOST that if inventory is reviewed and replenished continuously
PVALUE ¼ : (i.e., manufacturing can be done at any time, or purchase
AOD 3 NDAYS 3 SCOST
can be done at any time), then RLT ¼ OPT þ TLT þ PLT.
A. If inventory is reviewed and replenished If inventory review and supply replenishment are done
continuously (i.e., manufacturing or purchasing can periodically with review period length MLT, then
be performed at any time), then
RLT ¼ OPT þ TLT þ PLT þ MLT;
RLT ¼ OPT þ TLT þ PLT: 2 2 2
SRLT ¼ sqrt ðSOPT þ STLT þ SPLT Þ:
B. If inventory review and supply replenishment are
performed periodically with review period length Case 4. Achieve a target probability of on-time
MLT, then shipment to customer request date
In Case 4, on-time delivery in response to a customer
RLT ¼ OPT þ TLT þ PLT þ MLT ;
request is defined by OTDR (see notations above).
2 2 2
SRLT ¼ sqrt ðSOPT þ STLT þ SPLT Þ : PVALUE, RLT, and SRLT are calculated as follows.
First, PVALUE ¼ OTDR. If inventory is reviewed and
Case 2. Maximize expected profit replenished continuously (i.e., manufacturing can be done
In Case 2, the objective is to maximize the expected profit at any time, or purchase can be done at any time), then
(revenue minus purchase cost minus inventory cost). This RLT ¼ OPT þ TLT þ PLT CLTR. If inventory review
maximization is typically used in cases in which a large- and supply replenishment are done periodically with
quantity purchase (or manufacturing build) exists prior to review period length MLT, then
a season, or a period such as a month or a quarter. The RLT ¼ OPT þ TLT þ PLT CLTR þ MLT;
objective is to purchase or build the appropriate quantity 2 2 2 2
in order to maximize the expected profit in that season or SRLT ¼ sqrt ðSOPT þ STLT þ SPLT þ SCLTR Þ:
period. Since the supply quantity must be decided before
observing the demand, a risk is associated with the supply Case 5. Achieve target probability of on-time
quantity. Excess supply is sold at a loss at the end of the shipment to commit date
season or period. Supply shortage causes revenue and In Case 5, on-time delivery in response to a customer
profit shortfall. Relevant parameters include COST request is defined by OTDC (see notations above).
(purchase price or manufacturing cost per unit), PRICE PVALUE, RLT, and SRLT are calculated as follows.
(price of the unit in the market), and SVALUE (value per First, PVALUE ¼ OTDC. If inventory is reviewed and
unit for any unsold units left at the end of the season or replenished continuously (i.e., manufacturing can be done
other period when salvaged). at any time, or purchase can be done at any time), then
When the problem has multiple periods, the salvage RLT ¼ OPT þ TLT þ PLT CLTC. If inventory review
value per unit must be calculated as SVALUE ¼ PRICE and supply replenishment are done periodically with
PDECLINE. Here, PDECLINE is the loss of value of a review period length MLT, then
unit (or price decline) during the season or other period. RLT ¼ OPT þ TLT þ PLT CLTC þ MLT;
For items that have a very long life, where there is no or
2 2 2
minimal price decline, this model is not appropriate. SRLT ¼ sqrt ðSOPT þ STLT þ SPLT Þ:
Because a perpetual demand exists for such items, the
focus is typically on cost minimization or service target Case 6. Achieve a target fill rate
achievement instead of profit maximization. In Case 6, fill rate is defined by FRT (see notations
The PVALUE that maximizes the expected profit above). PVALUE, RLT, and SRLT are calculated as
during the season or other period is given by the follows. First, PVALUE ¼ FRT. If inventory is reviewed
following: and replenished continuously (i.e., manufacturing can 439

IBM J. RES. & DEV. VOL. 51 NO. 3/4 MAY/JULY 2007 K. KATIRCIOGLU ET AL.
be done at any time, or purchase can be done at any Consider a case with K transportation lead-time
time), then RLT ¼ OPT þ TLT þ PLT. If inventory observations in our sample given by L[1], L[2], , L[K ].
review and supply replenishment are done periodically We then calculate TLT as follows:
with review period length MLT, then
TLT ¼ ðL½1 þ L½2 þ þ L½K Þ=K
RLT ¼ OPT þ TLT þ PLT þ MLT; (average of all transportation times in the sample)
2 2 2
SRLT ¼ sqrt ðSOPT þ STLT þ SPLT Þ: To calculate the standard deviation, we need the sum of
the squares of transportation lead times, which we denote
Explanation of steps in calculations it as TLTS. In other words,
We now give a step-by-step description of the method for 2 2 2
calculating the optimal inventory policies by using the key TLTS ¼ L½1 þ L½2 þ þ L½K
problem parameters that we have calculated above. (sum of squares of all transportation times in the sample).
We then calculate the standard deviation as follows:
Step 1: Calculate daily demand statistics
2
TOQ ¼ ORQ[1] þ ORQ[2] þ þ ORQ[N] STLT ¼ sqrt½ðTLTS ¼ K 3 TLT Þ=ðK 1Þ
(sum of quantities of all orders in the data time window). (standard deviation of all transportation times in the
AOQ ¼ TOQ/N (average order quantity). sample).
TOQS ¼ ORQ[1]2 þ ORQ[2]2 þ þ ORQ[N]2
(sum of squares of all orders in the data time window). Step 3: Calculate key problem parameters
For various different types of problems, earlier in this
SOQ ¼ sqrt [(TOQS N 3 AOQ2)/(N 1)]
paper we have provided ways to calculate the key
(standard deviation of order quantity).
intermediary parameters PVALUE, RLT, and SRLT.
OAR ¼ N/T (average number of orders per day).
These are used to calculate the optimal inventory policy
AOD ¼ OAR 3 AOQ (average demand during a day). parameters.
SOD ¼ sqrt (AOD2 þ OAR 3 SOQ2)
(standard deviation of demand during a day). Step 4: Calculate mean and standard deviation of demand
during lead time
Step 2: Calculate lead-time statistics Mean demand during the replenishment lead time
Up to this point, in the above calculations we have is ADLT ¼ AOD 3 RLT. If the lead time is
assumed that total replenishment time RLT is a known random with standard deviation SRLT, standard
and fixed quantity. If the total replenishment time is not deviation of demand during lead time is
fixed, and every time it can be a different number, one SDLT ¼ sqrt (SRLT 2 3 AOD2 þ SOD2 3 RLT ).
must keep track of this lead time for each item-location If lead time is fixed, standard deviation of demand
combination through time. Then, by using this data during lead time is SDLT ¼ SOD 3 sqrt (RLT ).
one can estimate the mean and standard deviation of
this lead time. Step 5: Calculate fixed lot size
We assume that all lead times (i.e., OPT, TLT, PLT, All formulas provided so far are appropriate for one-for-
and CLTR) are random, except for the manufacturing one replenishment where no restriction exists on the lot
cycle time MLT and the customer order lead-time commit size or the frequency of orders. When the lot size is fixed
CLTC. or a minimum or maximum restriction exists on it,
In the following, we demonstrate how to calculate the because of replenishment process limitations, supply
standard deviation of the transportation lead time TLT requirements, or cost considerations, these restrictions
if one has a sample of transportation lead times. These must be taken into account.
calculations can be used for other lead times for which a The user provides a fixed lot size as an input, and this
sample of observations is available. First we introduce the size is used directly in the calculations. In this case, the
following notation: inventory policy is still to bring ‘‘inventory position’’
to ROP level whenever it falls below ROP. However,
TLT: Mean (or average) transportation lead time (as because of the restriction on the lot size, the order
previously defined). quantity must reflect these restrictions. We have five
TLTS: Sum of the squares of transportation lead-time different types of lot size restrictions, discussed in the
observations in the sample. following sections.
STLT: Standard deviation of transportation lead
time (this must be calculated from sample data of Case 1: Fixed lot size
transportation lead times for that item, from the plant In this case, there is a predetermined lot size QFIX.
440 to the stocking location for the item). Whenever the inventory position drops to ROP,

K. KATIRCIOGLU ET AL. IBM J. RES. & DEV. VOL. 51 NO. 3/4 MAY/JULY 2007
this quantity is ordered. Here, LOT ¼ QFIX. If no where MULTIPLIER ¼ 1 if PVALUE , 0.5,
predetermined lot size exists and economic order quantity MULTIPLIER ¼ 1 if PVALUE 0, A1 ¼ 1.363471, and
cannot be calculated because of the lack of fixed order A2 ¼ 0.266705. The formula we provide here for the k
cost data (or if economic order quantity is not value is an approximation for the inverse of the standard
desired), the following lot size can be recommended: normal distribution function. Since some versions of SQL
LOT ¼ AOD 3 RLT. Here RLT is the total do not have the inverse normal as a built-in function, this
replenishment lead and AOD is the average daily demand. formula is useful, since it uses mathematical function
available in SQL. The formula is very accurate for
Case 2: Minimum increments practical purposes.
In this case, the lot size has to be an integer multiple of We tested the values of k in the interval [3, 3]
a minimum number (QUNIT). The lot size is given by
and found that the absolute maximum and absolute
LOT ¼ M 3 QUNIT. Here M is the smallest integer for
minimum errors are 0.035711 and 0.035711,
which M 3 QUNIT is greater than ROP IP, and IP is
respectively.
the inventory position.
Case 6: Achieve service target (fill rate)
Case 3: Min lot size
Safety factor k is given by the following:
Lot size can be any number, but it must be above a
minimum. That is, LOT ¼ max (QMIN, ROP IP). Here, 2 3
A0 þ A1 Z þ A2 Z þ A3 Z
QMIN is the minimum lot size and IP is the inventory k¼ 2 3 4
;
B0 þ B1 Z þ B2 Z þ B3 Z þ B4 Z
position.
where
Case 4: Max lot size
Lot size can be any number, but it must be below a GVALUE ¼ ð1 PVALUEÞ 3 LOT=SDLT
maximum. That is, LOT ¼ min (QMAX, ROP IP). if there is a fixed lot size, LOT,
Here QMAX is the maximum lot size and IP is the GVALUE ¼ ð1 PVALUEÞ 3 ADLT=ðRLT 3 SDLTÞ
inventory position. if lot size is flexible,

Case 5: Economic order quantity Z ¼ sqrt f2 ln ½maxð1; 5=GVALUEÞg ;

When costs are available, an economic order quantity and A0 ¼ 5.3925569, A1 ¼ 5.6211054, A2 ¼ 3.883683,
(EOQ) can be calculated. The cost inputs required for this A3 ¼ 1.0897299, B0 ¼ 1.0000000, B1 ¼ 0.72496485, B2 ¼
calculation are OCOST (fixed order cost paid per supply 0.507326622, B3 ¼ 0.066913687, and B4 ¼ 0.003291291.
order) and HCOST (inventory holding cost, dollars The derivation of these formulas can be found in [6].
per unit held per day). The lot size is given by These formulas are used for continuous review problems
LOT ¼ sqrt (2AOD 3 OCOST/HCOST).
in which demand is also continuous. In this case, a simple
(s, Q) inventory policy can be used if no more than a
Step 6: Calculate safety factor
single replenishment order is outstanding in the supply
When we previously described six different cases, we
pipeline. Here, Q is the lot size (LOT), and s is the reorder
showed how to calculate key problem parameters in each
point (ROP). According to this policy, one places a
case. Now we show how to calculate safety factors. In the
replenishment order of Q when inventory position IP
inventory literature, these safety factors are calculated
drops to s.
using functions such as the inverse of the standard normal
The above approximation is a very accurate estimation
distribution and the inverse of what is called g-function.
These functions do not exist in standard query languages. of safety factor k for normal demand. We tested the
Here we show the use of approximations to these values of k in the interval [3, 3] and found that the
functions; these approximations can easily be coded maximum and minimum errors are 0.000295
in query languages. and 0.000205, respectively.

Cases 1 through 5 Step 7: Calculate optimal inventory policy

Safety factor k is given by the following: Inventory policy parameters include safety stock (SS),
reorder point (ROP), and maximum inventory level
k ¼ MULTIPLIER 3 ð0:5=A2 Þ (MAX). These parameters were explained earlier; their
2
3 A1 sqrt maxf0; A1 formulas, based on the safety factor and lot size
(LOT), are SS ¼ k 3 SDLT, ROP ¼ ADLT þ SS, and
þ 4A2 ln ½max ðPVALUE; 1 PVALUEÞ 1g ; MAX ¼ ROP þ LOT. 441

IBM J. RES. & DEV. VOL. 51 NO. 3/4 MAY/JULY 2007 K. KATIRCIOGLU ET AL.
acceptable statistical tolerance levels in our test cases.
Activity 1 2 3 4 5 6 7 8 9 10…… However, caution must be used in cases in which the
Data analysis and
assumptions we made here are arbitrary. In using the
algorithm selection formulas, we have shown a few issues that must be kept
Performance review and in mind; we reiterate some of these issues below.
algorithm refinement Our basic assumption on which the inventory policy
Solution design and calculations are based is that historical demand is a good
configuration representation of future demand. If trends in the demand
Implementation data exist, statistical forecasting models can be used to
planning estimate demand during lead times. Such forecasting
Pilot implementation models can generate the mean daily demand (i.e., the
daily demand forecast) and the standard deviation of
Full implementation daily demand. This forecast can replace the demand
statistics calculated in Step 1. It is possible to program
in SQL simple but effective forecasting techniques such
as exponential smoothing and moving average, since
Figure 3
these have simple forms.
Typical plan for the SQL-based inventory optimization solution The formulas normally work better with products that
implementation. The numbers at the top represent weeks, from
have a high demand volume (e.g., products that receive
week 1 to week 10 and beyond.
frequent orders). For low-demand products, inventory
policies are very sensitive to the amount of individual
orders; therefore, the actual service levels may vary
Implementation issues significantly for such products. For some low-demand
When implementing our approach, one must test the products, if it is critical to achieve desired inventory
calculations with respect to actual data, since many of availability levels, it would help to increase the lot sizes
the assumptions may not be appropriate for the data. to cover several months of demand.
Simulation tests can be performed using actual sales and Updating the inventory policy parameter (i.e., safety
lead time data to observe how the inventory policies stock, reorder point, and lot size) calculations may or
perform. If the performance is poor because of the may not be desirable. If the demand pattern changes very
normality of demand assumption, other distributions little, the inventory policy should also not be expected to
(such as gamma or log-normal) can be used. change much. In steady demand situations, monthly or
Approximations similar to what we presented for the quarterly parameter updates are reasonable. Otherwise,
normal distribution can be developed for other demand more frequent updates are needed.
distributions. Alternatively, empirical distributions can be Inventory replenishment orders should be triggered by
derived on the basis of sales data, and their percentiles observing the relative value of inventory position IP to
can be used to calculate the safety stocks. the reorder point ROP. That is, when IP drops below
After achieving satisfactory performance results, a ROP, an inventory replenishment order should be
solution can be designed with a configuration that triggered for the predetermined lot size. If the reviews are
addresses the customer’s requirements. A quick pilot periodic and the lot size is flexible, an order is placed in
implementation can follow and policies can be every period so as to bring the inventory position to ROP.
monitored in action for a subset of products before This is also known as a basestock policy.
full implementation. A project plan for a simple We note that IBM Global Business Services, IBM
implementation is given in Figure 3. The durations of Research, and NIBCO (Northern Indiana Brass
activities are our estimates for a standard implementation. Company) have jointly implemented a version of the
Actual implementation times can vary depending on the approach described in this paper to provide NIBCO a
complexity of a customer’s data systems and requirements. cost-effective inventory management solution. NIBCO
Some of the algorithms we have presented are tested has observed significant improvements in both inventory
using actual historical data from clients. The simulation levels and on-time shipment performance since 2005.
tests based on actual transaction data showed that for
a variety of demand patterns, the formulas used in this Conclusion
paper are able to deliver target service levels on the
average, although for some SKUs actual service can be Advantages
below the target and for some it can be above. Most of We have briefly explained our approach to create
442 the deviations from target service levels were within a virtually seamless inventory optimization solution

K. KATIRCIOGLU ET AL. IBM J. RES. & DEV. VOL. 51 NO. 3/4 MAY/JULY 2007
for database systems that use SQL. This approach has a Since sales data is accessed and other historical demand
number of potential advantages from both technical and statistics are calculated during data extraction from the
business process perspectives. Here, we briefly explain transaction system, if advanced forecasting algorithms
some of these advantages. are desired in order to perform more accurate forecasts,
Our approach simplifies optimal inventory policy these algorithms can be called during the data extraction
calculations in such a way that existing databases and process. The ability to use C or Java at the data
basic SQL commands can be used to rapidly code these extraction level gives enough flexibility to code advanced
calculations. Data-extraction templates and reporting algorithms.
templates of databases can be used to provide a fully Figure 4 shows details of how this flexible design
functional solution. Hence, for simple inventory may work. As the figure indicates, the optimal policy
problems, no need exists for additional software. calculations are performed during data extraction from
Because this is a solution provided on an existing the ERP system. The optimal inventory policies are
database system, there is no need to create new security stored, along with input data, and are made ready for
protocols, network integration infrastructure, and Web reporting purposes. In addition, some metrics such as
access protocols. All of these elements are typically projected inventory levels, projected backlogs, and their
provided by the existing database system infrastructure confidence bands can also be calculated and stored.
for the users. Queries are used for reporting only, and no optimization
By using query capabilities of the existing database calculation is performed at the query level. This
application, various reports of optimal policies, key architecture provides the flexibility to place complex
performance indicator (KPI) projections and ‘‘what-if ’’ algorithms during the data extraction. Programming
analyses for decision support can be created. Many of languages such as Java, C, or Cþþ can be used to code the
these basic analyses can be done using mathematical algorithms.
functions available in SQL. This solution design has a drawback in that it may not
Because companies that use database applications be possible to obtain quick real-time results because of
already have trained staff, much less technical training is architectural and algorithmic complexity. However,
required in order to use this solution. As for the business potential scenarios can be analyzed in anticipation of user
process, the solution does not necessarily require needs, and results can be stored prior to user requests.
changing existing inventory planning processes. It Thus, real-time response can be provided to anticipate
requires that the execution of recommended inventory queries.
policies be done properly by monitoring key metrics Numerous sources of complications exist in inventory
such as inventory position and sales orders. problems, and this is why the inventory-control literature
Because this solution requires no new software, is so rich. The multi-echelon nature of supply chain
typical purchase costs and user fees do not apply. The systems, substitutability of products, nonstationary
maintenance costs can be controlled because this solution
nature of demand, products with bills of materials,
can be implemented as a part of the existing database/
and distribution systems are examples of the elements
data-warehouse maintenance program.
that make inventory problems complex and variable.
Because the calculations are done at runtime through
(Distribution systems are systems in which inventory flows
query commands, inventory managers and analysts can
from manufacturing to central distribution centers, then
perform real-time what-if analysis using query reports.
to warehouses, and then to local stores.)
For instance, analysts can observe how inventory levels
Since the literature is extensive, we can mention only
are affected if they change some key factors such as
a few key relevant publications. Readers who wish to
customer service objectives, operating objectives, and
become more familiar with these complexities may
inventory policy types.
consult [10, 11, 19, 22–30] for multi-echelon systems.
More specifically, for one-warehouse, multi-retailer
More complex inventory problems
systems, see [5, 14, 15, 31–36]. For assembly
Although the approach we have presented in this paper
manufacturing systems, see [4, 30–35, 37].
can solve some common inventory problems, many other
cases exist for which more complex algorithms are
*Trademark, service mark, or registered trademark of
needed. These can be coded following the standard International Business Machines Corporation in the United States,
approach and using a common programming language other countries, or both.
such as C or Cþþ or Java. Integration of these algorithms
**Trademark, service mark, or registered trademark of Sun
into the solution is typically done at the data extraction Microsystems, Inc., Microsoft Corporation, or SAP
level. The results are then stored in database tables. Aktiengesellschaft in the United States, other countries, or both. 443

IBM J. RES. & DEV. VOL. 51 NO. 3/4 MAY/JULY 2007 K. KATIRCIOGLU ET AL.
Optimal Cubes
Processed Demand Processed
policy and
data forecasts and data
performance
(demand) scenarios (other)
metrics
Data warehouse

Data Demand Inventory

Data Query
processing forecasting optimization
extractor engine
engine engine engine

Optimal Policy parameters

recommended
Transaction data policies
LIBRARY 1 LIBRARY 2 Business
Financial data Forecasting Optimization performance targets
Metrics projections
algorithms and simulation
Operational data algorithms
What-if
“What-if ” questions
analysis reports

ERP system Web

Figure 4
Flexible architecture that can support use of complex optimization algorithms. Within the “Web” box, yellow boxes represent user inputs,
and light blue boxes represent outputs to the user.

References 13. E. Ignall and A. Veinott, ‘‘Optimality of Myopic Inventory

Policies for Several Substitute Products,’’ Manage. Sci. 15,
1. S. Graves, A. R. Kan, and P. Zipkin, Editors, Logistics of
No. 5, 284–304 (1969).
Production and Inventory, Vol. 4, Elsevier (North-Holland),
14. J. A. Muckstadt, ‘‘A Model for a Multi-Item, Multi-Echelon,
Amsterdam, The Netherlands, 1993.
Multi-Indenture Inventory System,’’ Manage. Sci. 20, No. 4,
2. A. C. Hax and D. Candea, Production and Inventory
472–48l (1973).
Management, Prentice-Hall, Upper Saddle River, NJ, 1984.
15. H. Scarf, ‘‘The Optimality of (s, S) Policies in the Dynamic
3. F. S. Hillier and G. J. Lieberman, Introduction to Operations Inventory Problem,’’ Mathematical Methods in The Social
Research, Holden-Day, San Francisco, CA, 1980. Sciences, Stanford University Press, Stanford, CA, 1959,
4. S. Nahmias, Production and Operations Analysis, Fourth pp. 196–202.
edition, McGraw-Hill, New York, 2001. 16. D. Topkis, ‘‘Optimal Ordering and Rationing Policies in a
5. L. B. Schwarz, Editor, Multi-Level Production/Inventory Non-Stationary, Dynamic Inventory Model with n Demand
Control Systems: Theory and Practice, North-Holland, Classes,’’ Manage. Sci. 15, No. 3, 160–176 (1968).
Amsterdam, The Netherlands, 1981. 17. R. Ehrhardt, ‘‘Power Approximation for Computing (s, S)
6. E. A. Silver, D. F. Pyke, and R. Peterson, Inventory Inventory Policies,’’ Manage. Sci. 25, No. 8, 777–786 (1979).
Management and Production Planning and Scheduling, John 18. J. R. Freeland and E. L. Porteus, ‘‘Evaluating the Effectiveness
Wiley & Sons, Third edition, Hoboken, NJ, 1998. of a New Method for Computing Approximately Optimal
7. R. J. Tersine, Production and Operations Management, North- (s, S) Inventory Policies,’’ Oper. Res. 28, No. 2, 353–366 (1980).
Holland, New York, l980. 19. Y. S. Zheng and A. Federgruen, ‘‘Finding Optimal (s, S)
8. P. Zipkin, Foundations of Inventory Management, McGraw- Policies Is About as Simple as Evaluating a Single Policy,’’
Hill, New York, 2000. Oper. Res. 39, No. 4, 654–665 (1991).
9. G. Hadley and T. M. Whitin, Analysis of Inventory Systems, 20. J. R. Groff and P. N. Weinberg, SQL: The Complete
Prentice-Hall, New York, 1963. Reference, Second edition, McGraw-Hill, New York, 2002.
10. K. J. Arrow, S. Karlin, and H. Scarf, Mathematical Theory of 21. J. S. Song and P. H. Zipkin, ‘‘Managing Inventory with the
Inventory and Production, Stanford University Press, Stanford, Prospect of Obsolescence,’’ Oper. Res. 44, No. 1, 215–222
CA, 1958. (1990).
11. A. J. Clark and H. Scarf, ‘‘Optimal Policies for a Multi- 22. D. Atkins and D. Sun, ‘‘98% Effective Lot Sizing for Series
Echelon Inventory Problem,’’ Manage. Sci. 6, No. 4, 475–490 Inventory Systems with Backlogging,’’ Oper. Res. 43, No. 2,
(1960). 335–345 (1995).
12. D. L. Iglehart, ‘‘The Dynamic Inventory Problem with 23. A. J. Clark and H. Scarf, ‘‘Approximate Solutions to a Simple
Unknown Distribution of Demand,’’ Manage. Sci. 10, 429– Multi-Echelon Inventory Problem,’’ Studies in Applied
444 440 (1964). Probability and Management Science, K. J. Arrow, S. Karlin,

K. KATIRCIOGLU ET AL. IBM J. RES. & DEV. VOL. 51 NO. 3/4 MAY/JULY 2007
and H. Scarf, Editors, Stanford University Press, Stanford, Kaan Katircioglu IBM Research Division, Thomas J. Watson
CA, 1962, pp. 88–110. Research Center, P.O. Box 218, Yorktown Heights, New York
24. M. A. De Bodt and S. C. Grave, ‘‘Continuous Review Policies 10598 ([email protected]). Dr. Katircioglu is a leading researcher
for a Multi-Echelon Inventory Problem with Stochastic and a consultant in supply chain management at the IBM Thomas
Demand,’’ Manage. Sci. 31, 1286–1295 (1985). J. Watson Research Center. He has more than ten years of
25. A. Federgruen and P. Zipkin, ‘‘Computational Issues in an experience in the field of operations research, management science,
Infinite-Horizon Multiechelon Inventory Model,’’ Oper. Res. and logistics. His expertise covers the areas of inventory
32, No. 4, 818–836 (1984). optimization, distribution and manufacturing operations
26. S. C. Graves and L. Schwarz, ‘‘Single Cycle Continuous management and planning, and e-business and supply chain
Review Policies for Arborescent Production/Inventory management. After completing his Ph.D. studies in management
Systems,’’ Manage. Sci. 23, No. 5, 529–540 (1977). science, in 1996 he joined the IBM Thomas J. Watson Research
27. P. Jackson, W. Maxwell, and J. Muckstadt, ‘‘Determining Center. Since then, Dr. Katircioglu has worked on several projects
Optimal Reorder Intervals in Capacitated Production– for various divisions of IBM and its customers. He is the inventor
Distribution Systems,’’ Manage. Sci. 34, No. 8, 938–958 of the Express Services for Inventory Management offering
(1988). announced by IBM in July 2005. Dr. Katircioglu has published
28. L. B. Schwarz and L. Schrage, ‘‘Optimal and System Myopic numerous papers, made several conference presentations, and
Policies for Multi-Echelon Production/Inventory Assembly received patents for his work. He is a member of INFORMS
Systems,’’ Manage. Sci. 21, No. 11, 1285–1294 (1975). and IEEE.
29. C. C. Sherbrooke, Optimal Inventory Modeling of Systems,
Multi-Echelon Techniques, John Wiley & Sons, Hackensack,
NJ, 1992.
30. W. Zangwill, ‘‘Eliminating Inventory in a Series Facility,’’ Timothy M. (Tim) Brown IBM Global Business Services,
Manage. Sci. 33, No. 9, 1150–1164 (1987). 4111 Northside Parkway NW, Atlanta, Georgia 30327
31. S. Axsater, ‘‘Simple Solution Procedures for a Class of Two- ([email protected]). Mr. Brown is an Associate Partner
Echelon Inventory Problems,’’ Oper. Res. 38, No. 1, 64–69 with the Global Business Services (GBS) division of IBM Global
(1990). Services. Within GBS, Mr. Brown currently focuses his efforts
32. S. Axsater, ‘‘Evaluation of Lot-Sizing Techniques,’’ Intl. J. on defining and implementing supply chain transformation
Production Res. 24, No. 1, 51–57 (1986). approaches for midsized companies. He is a certified business
33. J. A. Muckstadt and R. O. Roundy, ‘‘Multi-Item, One- transformation consultant in the IBM Supply Chain Management
Warehouse, Multi-Retailer Distribution Systems,’’ Manage. competency, with 25 years of experience in logistics management,
Sci. 33, No. 12, 1613–1621 (1987). supply chain strategy, and supply chain operations improvement.
34. K. Rosling, ‘‘Optimal Inventory Policies for Assembly Systems He is also a certified Project Management Professional and is
Under Random Demands,’’ Oper. Res. 37, No. 4, 565–579 certified in production and inventory management. Mr. Brown has
(1989). worked with clients of all sizes and numerous industry sectors. He
35. R. Roundy, ‘‘98%-Effective Integer-Ratio Lot-Sizing for One- received an M.B.A. degree from Georgia State University and a
Warehouse Multi-Retailer Systems,’’ Manage. Sci. 31, No. 11, B.S. degree in management science from the Georgia Institute of
1416–1430 (1985). Technology.
36. L. B. Schwarz, B. Deuermeyer, and R. Badinelli, ‘‘Fill-Rate
Optimization in a One-Warehouse N-Identical Retailer
Distribution System,’’ Manage. Sci. 31, No. 4, 488–498 (1985).
37. C. C. Sherbrooke, ‘‘VARI-METRIC: Improved Mateen Asghar IBM Global Business Services, One Rogers
Approximations for Multi-Indenture, Multi-Echelon Street, Lotus Development, Cambridge, Massachusetts 02142
Availability Models,’’ Oper. Res. 34, No. 2, 311–319 (1986). ([email protected]). Mr. Asghar is an Industry
Solution Manager at IBM Global Business Services (GBS). He is
Received September 25, 2006; accepted for publication responsible for the SAP Business Intelligence Express solution and
October 16, 2006; Internet publication May 25, 2007 is a subject matter expert on the cross-solution integration for
small and medium-sized companies in various industries. He has
more than ten years of experience with SAP R/3 systems as a
technical development leader, and he is a Business Intelligence
subject matter expert, an integration subject matter expert, and a
leader for various large-scale and complex projects. Mr. Asghar
has participated in several projects for Fortune 500 companies,
primarily in the area of consumer products, aerospace, and defense
industries. His expertise is in managed full-scale SAP systems
development projects with large teams utilizing onsite/offshore
models. He was the lead architect of the Express Services for
Inventory Management offering announced by IBM in July 2005,
and he designed the complete solution utilizing various
components of SAP, ERP, and BI.

445

IBM J. RES. & DEV. VOL. 51 NO. 3/4 MAY/JULY 2007 K. KATIRCIOGLU ET AL.