Onthemark: Integrated Stochastic Resource Planning of Human Capital Supply Chains

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/261965325
OnTheMark: Integrated Stochastic Resource Planning of Human Capital Supply

Chains
Article in Interfaces · October 2011

DOI: 10.2307/41318830
CITATIONS READS
16 212
12 authors, including:
H. Cao Jianying Hu
IBM IBM
16 PUBLICATIONS 76 CITATIONS 173 PUBLICATIONS 3,583 CITATIONS
SEE PROFILE SEE PROFILE
Chen Jiang Tarun Kumar

IBM Oxford Brookes University
4 PUBLICATIONS 19 CITATIONS 94 PUBLICATIONS 269 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
M.V.Sc. View project
Modeling using queueing theory View project
All content following this page was uploaded by Yingdong lu on 27 November 2014.
The user has requested enhancement of the downloaded file.

Vol. 41, No. 5, September–October 2011, pp. 414–435
issn 0092-2102 eissn 1526-551X 11 4105 0414 http://dx.doi.org/10.1287/inte.1110.0596
© 2011 INFORMS
OnTheMark: Integrated Stochastic Resource Planning

of Human Capital Supply Chains
Additional information, including rights and permission policies, is available at http://journals.informs.org/.
INFORMS holds copyright to this article and distributed this copy as a courtesy to the author(s).
Heng Cao, Jianying Hu, Chen Jiang, Tarun Kumar, Ta-Hsin Li, Yang Liu,
Yingdong Lu, Shilpa Mahatma, Aleksandra Mojsilović, Mayank Sharma,
Mark S. Squillante, Yichong Yu
Business Analytics and Mathematical Sciences Department, IBM Thomas J. Watson Research Center,
Yorktown Heights, New York 10598
{hengcao@us.ibm.com, jyhu@us.ibm.com, chjiang@us.ibm.com, ktarun@us.ibm.com, thl@us.ibm.com,
jyliu@us.ibm.com, yingdong@us.ibm.com, mahatma@us.ibm.com, aleksand@us.ibm.com,
mxsharma@us.ibm.com, mss@watson.ibm.com, yichong@us.ibm.com}
In this paper, we present a suite of innovative operations research models and methods called OnTheMark
(OTM). This suite supports the effective management of human capital supply chains by addressing distinct
features of human talent that cannot be handled via traditional supply chain management. OTM consists of
novel solutions for (1) statistical forecasting of demand and human capital requirements, (2) risk-based stochas-
tic human-talent capacity planning, (3) stochastic modeling and optimization (control) of human capital supply
evolutionary dynamics over time, (4) optimal multiskill supply-demand matching, and (5) stochastic optimiza-
tion of business decisions and investments to manage human capital shortages and overages. The OTM suite
was developed and deployed as an important part of the human capital management and planning process
within IBM, providing support for decision making to drive better business performance. This is achieved
through important contributions in the areas of stochastic models and optimization (control), and the innovative
application and integration of these models and methods in human capital management applications.
Key words: human capital supply chains; human capital management and planning; demand forecasting;
risk-based capacity planning; stochastic supply evolution; multiskill supply-demand matching; stochastic
models; stochastic optimization and control.
O ver the past 50 years, the services sector has

grown to dominate economic activity in most
advanced industrialized countries. Recent US Bureau
people with specific skills, in markets characterized by
highly volatile and uncertain client demands. Hence,
forward-thinking businesses are beginning to invest
of Labor Statistics data reveal that 75 percent of in human capital supply chain (HCSC) methods as a
the labor force is employed in the services industry major competitive differentiator.
and that services industry output represents nearly Over recent decades, organizations have achieved
70 percent of the total industry output (Woods 2009). significant gains in effectiveness and efficiency by
A critical driver of success for any service-delivery developing advanced models of traditional manufac-
organization is its ability to manage and deploy the turing and logistics systems to optimize their supply
skills, knowledge, and competencies of its human chain operations. Although such traditional supply
capital—“having the right people, with the right skills, chain models and methods have a rich history within
in the right place, at the right time.” Successful orga- the operations research (OR) community, they cannot
nizations realize that investments in their people are be directly applied to related problems in HCSCs.
key drivers behind growth, profitability, and client sat- People are quite distinct from and more complex to
isfaction. This is especially true for providers of infor- model than the inanimate machine parts in tradi-
mation technology (IT) services such as IBM, who tional supply chains; thus, new models and methods
offer a broad range of service products, each requiring are required to capture and represent these distinct
414
Cao et al.: Integrated Stochastic Resource Planning of Human Capital Supply Chains
Interfaces 41(5), pp. 414–435, © 2011 INFORMS 415
features and complexities. Concepts such as hiring, support risk-based capacity planning that determines
training, retaining, learning, and acquiring new skills human-talent levels to maximize business perfor-
fundamentally influence the optimization problems mance. We develop an approach to predict future tal-
relevant to human capital management (HCM), mak- ent and skill composition via stochastic modeling and
ing them more difficult to model and solve, but also optimization (control) of supply evolution. We per-
more realistic; therefore, these concepts are of crucial form an optimal matching of multiskill talent against
importance to the successful management and plan- demand targets; we then optimize investment deci-
ning of HCSCs in practice. sions to address talent shortages and overages via our
A prime example of the importance of effective combined modeling and optimization solutions.
HCM is Integrated Technology Services (ITS), an IBM Since 2008, OTM has been deployed as part of the
line of business. ITS is a premier integrator of IT in ITS HCM process, helping to improve the manage-
the services industry; its annual revenues are approxi- ment and planning of human talent. Based on our
mately $4 billion. Through its 10 service product lines, experience with quarterly reviews, the OTM solution
ITS delivers a range of integration and support service demonstrates relative revenue-cost benefits commen-
products, henceforth called service engagements (e.g., surate with 2–4 percent of ITS revenue targets over
the design and implementation of a data warehouse), previous approaches. Beyond this internal applica-
for all IBM hardware and software products and tion, we also develop prototype implementations of
beyond. The ITS business focuses on human talent; OTM models and methods within IBM platforms to
the majority of its engagement costs consist of human support future IBM solution offerings for a broad
services rather than costs related to hardware and collection of clients beyond IT services.
software. Unlike machine parts in traditional supply This paper presents an overview of the OTM solu-
chains, the people comprising the ITS supply chain tion, focusing primarily on mathematical modeling
are not consumed during the service-delivery process and optimization contributions. We first provide a
and represent a long-term investment. The productiv- high-level overview and then present a somewhat
ity and efficiency of people depend upon workload more detailed discussion of the OR models and
and utilization, whereas these notions for machine methods. Cao et al. (2010) provide a more detailed
parts are less sensitive to workload and utilization. technical description. We conclude with a summary
People can evolve in various complex ways, and are of our worldwide OTM deployment within IBM.
capable of deploying more than one skill at a time and Throughout the paper, we use the term skill to gener-
doing so across multiple engagements. The service- ally refer to the full extent of human capital abilities,
delivery process is further complicated by uncertainty competencies, knowledge, skills, and talents. Exam-
in resource staffing, simultaneous allocation of multi- ples include database architect, Java programmer, and
ple resources (because engagements require different network specialist.
skills), and resource sharing (because people time-
share their skills across different engagements).
The effective and efficient management and plan- Solution Overview
ning of human talent, while addressing the fun- To address the challenges and complexities of HCSCs,
damental differences between human capital and we develop OR solutions with the goal of deter-
traditional supply chains, is pivotal to the ITS deliv- mining the talent levels and investments over time
ery of service engagements. Seeking novel solu- that maximize business performance. Although var-
tions, IBM ITS partnered with IBM Research to ious investments are possible, we focus on hiring,
develop OnTheMark (OTM), an HCM framework training, promotions, and retention (via incentives) as
comprising innovative OR models and methods to representative decisions available to ITS. We imple-
address the challenges and complexities of HCSCs. ment our OR solutions as core capabilities of the OTM
We develop the capability to forecast engagement suite to support an overall HCSC process. OTM con-
demand and human-talent requirements. We derive sists of: (I) demand forecasting; (II) risk-based capac-
stochastic modeling and optimization methods to ity planning; (III) supply evolution and optimization;
416 Interfaces 41(5), pp. 414–435, © 2011 INFORMS
Skill shortage and overage management V

(Combination of II, III, and IV)
II
Risk-based capacity
planning
(Determine skill capacity target)

I Multiskill shortage and IV
Demand forecasting overage
(Determine offering (Matching target
demand) against supply)
Supply evolution and III
optimization
(Determine human
resource supply)
Figure 1: The flowchart illustrates the capabilities and interrelationships in the OTM suite of OR models and
methods.
(IV) multiskill shortage and overage analysis; (V) skill Given the distinct perspectives of different ITS users,
shortage and overage management (see Figure 1). OTM capabilities operate across a diverse set of plan-
Various business and organizational aspects of ITS ning horizons with time scales ranging from weeks
influenced our partitioning of the OTM solution and months to quarters and years. We next present
among its constituent elements, which are depicted an overview of the primary OTM elements and then
in Figure 1 and mapped to corresponding OR mod- describe each element in somewhat more detail in the
els and methods in Table 1. This highlights another sections that follow.
important contribution: the integration of such a col-
lection of OR methodologies. The OTM design also Demand Forecasting
supports interactive sessions that allow users to eval- The first step, demand forecasting (I), statistically
uate different service-delivery scenarios under vary- characterizes the service engagement demand over
ing assumptions and conditions. This requirement the planning horizon in terms of revenue to the
creates additional methodological challenges to com- provider, number of engagements to be delivered,
pute solutions of complex stochastic modeling and required time for starting each service delivery, and
optimization problems in an efficient manner. skills required for delivery. Service engagements are
Beyond combined usage via the OTM suite, each described in terms of revenue, duration, and type,
capability is used independently by different ITS often without linkages to staffing templates. To obtain
organizations to support various HCSC processes. a more accurate view of demand, we use machine
learning methods to estimate the skill staffing require-
ments of each engagement from historical data. Sta-
OTM capabilities OR models and methods
tistical forecasting techniques are used to estimate
Demand forecasting Logistic regression the demand for each engagement; together with our
Risk-based capacity planning Stochastic loss networks, stochastic staffing templates, they yield a complete character-
optimization ization of demand. An example output consists of
Supply evolution and optimization Stochastic temporal models,
stochastic optimization and control 10 and 5 engagements over the next two quarters,
Multiskill shortages and overages Mathematical programming respectively, to implement data warehouses (engage-
Skill shortage and overage Stochastic optimization ment demand), with start times uniformly distributed
management
over each quarter and each engagement requiring two
Table 1: The table illustrates a mapping of OTM capabilities to OR models database architects, four Java programmers, a net-
and methods. work specialist, and so on (staffing templates). Such
demand forecasting (I) outputs serve as input to risk- dynamics and trade-offs can be evaluated and opti-
based capacity planning (II) and supply evolution and mized from diverse perspectives by using risk-based
optimization (III). capacity planning in different ways. This includes
ideal skill capacity targets (independent of existing
Risk-Based Capacity Planning human capital) and optimal skill capacity levels and
The next OTM capability concerns capacity business investment decisions to achieve these levels
planning—determining the capacity levels for people
(subject to existing supply constraints), both to maxi-
skills that satisfy demand and maximize business
mize business performance.
performance. Given the distinct features of people
An example output includes expected profit (rev-
and HCSCs, OTM’s capacity planning requirements
enue) being maximized over the next quarter (under
create a need for new stochastic modeling and
a 12 percent gross profit margin constraint) with 200
optimization solutions that capture the business
(250) database architects and 500 (600) Java program-
risks, HCSC dynamics, and demand uncertainty. In
traditional supply chains, product demand is often mers. The outputs of risk-based capacity planning (II)
converted into machine part demand through bills over each time interval, which include corresponding
of materials and then provided as direct input into skill capacity levels, serve as input to each of supply
supply-demand matching. However, it is crucial in evolution and optimization (III) and multiskill short-
HCSCs to first analyze the financial implications age and overage analysis (IV).
and risks associated with distinct features of people
and human talents, including longer-term costs and Supply Evolution and Optimization
complex dynamics, as a function of engagement Another OTM capability concerns modeling the evo-
demand. lution of human capital and optimizing decisions to
To address these challenges, we develop a risk- control the evolutionary dynamics of human talent.
based capacity planning solution (II) that models The resource dynamics in traditional supply chains
the dynamics, trade-offs, and uncertainties associated are limited, with decisions typically restricted to
with HCSCs and solves a corresponding stochastic ordering. In contrast, HCSCs are often characterized
optimization problem based on this model. The risk- by complex time-varying supply-side dynamics, with
based capacity planning models consist of multiclass many people acquiring skills, gaining efficiencies, and
stochastic loss networks with simultaneous resource changing roles, some people attriting, and new peo-
allocation in which: losses model the risks of lost
ple being hired. This results in substantial changes
demand (and associated revenue) due to insufficient
in the talent levels and skill composition of human
human capital at the time of engagement delivery;
capital over a given planning horizon. The complex
teams of people are required to jointly deliver service
time-varying evolutionary dynamics of human cap-
engagements; and multiple classes model different
ital create a need for stochastic temporal models of
types of service engagements and human talents. Per-
supply-side dynamics and stochastic optimization of
formance is a function of the revenues for service
engagement delivery, discounted by probabilities for these human-talent dynamics over time through vari-
service engagements at risk of being lost, and the ous investment decisions, including traditional hiring
costs for maintaining skill capacity levels. We then (ordering) options.
solve the corresponding risk-based stochastic opti- To address these challenges, we develop a sup-
mization problem to determine the skill capacity tar- ply evolution and optimization solution (III) that
gets that maximize expected business performance. models the evolutionary dynamics of human tal-
OTM risk-based capacity planning addresses the ent and skill composition and solves a correspond-
revenue-cost dynamics and trade-offs between the ing stochastic optimization (control) problem based
risks of too few people with appropriate skills to on these models to maximize expected business per-
deliver an engagement when needed and the risks formance over time. The supply evolution models
of too many underutilized people. These revenue-cost consist of discrete-time multidimensional stochastic
processes whose states represent people with vari- remaining 4 (5) optimally matched to deploy Java
ous combinations of talents and skills. The multi- programmer skills in the next quarter (quarter after
period supply evolution optimization (control) prob- next). These multiskill shortage and overage anal-
lem incorporates these discrete-time stochastic pro- ysis (IV) outputs, including optimal matchings of
cesses together with decision variables and associated human capital to skill capacity targets and corre-
lead times for each action. sponding skill shortages and overages across every
An example output includes evolution without time interval comprising the planning horizon, serve
intervention, rendering shortages in database archi- as input to risk-based capacity planning (II) and sup-
tects and Java programmers that cause lost revenue ply evolution and optimization (III).
over the next few quarters, and maximization of
Skill Shortage and Overage Management
expected profit over the next few quarters by hiring
The final OTM capability concerns the optimization
two database architects and eight Java programmers,
of business decisions to manage skill shortages and
while incentivizing six Java programmers to delay
overages. We develop this solution (V) based on
retirement. Such investment decisions over planning
an iterative approach that combines the main OTM
horizons comprising stochastic behaviors that vary
optimization capabilities by exploiting their interac-
over time, together with the corresponding talent lev-
tions, dependencies, and time scales until conver-
els over the planning horizon, serve as input to both gence to a fixed-point equilibrium (steady state).
risk-based capacity planning (II) and multiskill short- Specifically, risk-based capacity planning optimiza-
age and overage analysis (IV). tion (II) determines the human-talent levels, supply
evolution and optimization (III) determines the evo-
Multiskill Shortages and Overages
lutionary dynamics of future skill composition, and
The next OTM capability concerns the analysis of skill
multiskill shortage and overage analysis (IV) deter-
shortages and overages. In traditional supply chains,
mines the matching of human capital against skill
the resource shortages and overages are determined
capacity targets, all with the goal of maximizing
directly from demand and supply quantities. In con-
business performance over time. This yields a set
trast, the ability of people to perform multiple func-
of investment decisions for optimally managing skill
tions, and do so at the same time within one or shortages and overages. An example output includes
more engagements, necessitates an advanced form of addressing an overage of one database architect and
matching the multiskill supply to the demand. a shortage of six Java programmers, among others,
To address these challenges, we develop a multi- over the next few quarters by retraining one database
skill shortage and overage analysis solution (IV) that: architect to acquire another skill, hiring four Java pro-
models the multiple skills of people over each inter- grammers, and incentivizing two Java programmers
val of the planning horizon involving time-invariant to delay their retirement.
stochastic behaviors; optimizes the matching of such The dynamics and uncertainties of HCM decisions
multiskill people against skill capacity targets to min- occur at various time scales, as reflected in our OR
imize a weighted sum of the skill shortages and over- models and methods. Risk-based capacity planning
ages; and computes shortages and overages for each operates over stationary intervals (typically on the
individual skill under this optimal matching. We for- order of a month or quarter) within the planning
mulate the optimization problem for each interval of horizon; supply evolution and optimization operates
the planning horizon as a linear program; we then across the entire nonstationary planning horizon (typ-
solve this optimization problem over the entire plan- ically on the order of a few quarters, a year, or longer);
ning horizon as a dynamic program. and multiskill shortage and overage analysis oper-
An example output includes, among a set of 10 peo- ates across both stationary and nonstationary inter-
ple with both database architect and Java program- vals. The notion of nonstationarity (stationarity) refers
mer skills in the next quarter and 14 such people to stochastic behaviors that vary in distribution (are
in the quarter after next, 6 (9) who are optimally distributionally invariant to time shifts) over suffi-
matched to deploy database architect skills, with the ciently long intervals. We consider the ITS HCSC
across relatively long planning horizons that include engagement delivery. Such clusters for each engage-
different forms of dynamics, uncertainties, and investment type are validated and refined through dis-
ment decisions. These time horizons are modeled as cussions with subject matter experts. Because these
nonstationary intervals that consist of stationary peri- staffing templates evolve and require some degree of
ods, where the different forms of complex dynamics customization, we develop an automated methodol-
and uncertainties change from one stationary period ogy for generating and adjusting staffing templates by
to the next period according to a general stochastic exploiting the semisupervised clustering and machine
modulation process. learning framework of Hu et al. (2008a, b). The result-
ing dynamic taxonomy of staffing templates provides
important input to ensure accurate forecasting of
Demand Forecasting Models
human-talent requirements from all three sources of
We forecast future demand for service engagements engagement demand.
from three sources: (1) ongoing engagements being
delivered; (2) opportunities of potential deals at dif-
ferent stages in the sales pipeline; and (3) deals antic- Risk-Based Capacity Planning
ipated based on market research and experience, but The service engagement demands and skill require-
not concrete enough for entry in the pipeline. Service ments from our forecasting models are provided
engagements require a collection of different skills, as input to the risk-based capacity planning solu-
where the staffing templates that link engagement tion. Our goal is to determine the skill capacity
types and expected revenue to skill requirements in levels that best satisfy demand while addressing
HCSCs are considerably more complex than tradi- the complex trade-offs among revenue, cost, and
tional bills of material and labor. associated risks. To highlight some of the difficul-
We apply statistical techniques to engagement ties involved, consider a scenario demand forecast
delivery data to compute probabilistic characteriza- for 10 service engagements over the next quarter.
tions for the completion times of ongoing engage- The best skill capacity levels to deliver this set of
ments and the corresponding availability of deployed engagements depend on the degree of overlap among
human capital. For pipeline opportunities, we the corresponding delivery-time processes. At one
develop statistical models based on logistic regres- extreme, if the delivery of all engagements over-
sion techniques to predict the probability of each laps, then 10 times the required staffing levels of one
pipeline deal being won (i.e., contract signed) based engagement is required; at the other extreme, if the
on attributes such as lapse time and movement in the delivery of these engagements is disjoint, then only
pipeline, client information, and deal size. These win the staffing levels of one engagement are required.
probabilities are then used to probabilistically charac- Because the reality is typically somewhere between
terize the number of engagements of each type from these extremes, a model of these complex dynam-
the pipeline and their start times over future periods. ics and stochastic behaviors is needed to appropri-
For expected deals, we use quarterly revenue targets ately determine the skill capacity levels that maximize
and typical deal sizes to probabilistically character- business performance subject to service engagement
ize the number of engagements of each type to be demand forecasts.
started to make up the difference between revenue We therefore model the capacity planning problem
targets and expected revenues from ongoing engage- as a stochastic loss network in which the risk of losing
ments and pipeline opportunities. service engagement demand (and associated revenue)
Given a lack of linkage between service engage- due to insufficient capacity in one or more required
ments and required skills, we develop models to skills at the time the service engagement must be
estimate staffing requirements for each engagement delivered is represented by the stationary loss prob-
type. Following the statistical clustering methodol- abilities of the stochastic network. It is important to
ogy of Hu et al. (2007), we generate groups of sim- note that for our purposes, we use the general notion
ilarly staffed engagements using historical data on of arrivals in the stochastic loss network to model
the epochs at which the delivery of an engagement of multiple skills and multiple service engagements.
is required. Because this differs somewhat from com- The delivery of an engagement requires units of
mon usage, we emphasize that the arrival process for capacity from people with specific skills. Instances of
each type of engagement captures the points in time each type of engagement demand must be delivered
when such engagements need to be delivered. A pri- at time epochs following an independent stochastic
mary reason for modeling these effects is because the arrival process. Such a delivery opportunity is at risk
loss of service engagement demand due to insuffi- of being lost if available skill capacity is less than that
cient capacity with required skills was an important required in one or more skills; otherwise, the engage-
issue at the start of the project. In particular, it is ment is delivered and the capacity is reserved for each
not uncommon in the services industry for clients to required skill throughout the duration of the engage-
switch to a competitor or otherwise withdraw a ser- ment. The delivery duration times are independent
vice engagement if delivery cannot be provided at and identically distributed random variables follow-
the required time. This, together with the revenue- ing a general distribution with unit mean. Engage-
cost trade-off between too many underutilized people ment delivery (arrival) epochs and duration times are
or too few people with required skills, greatly influ- mutually independent.
enced our decision to use a stochastic loss network to Our stochastic loss network models the following:
model and optimize the risk-based capacity planning for each skill, the stationary probability that service-
problem. delivery arrivals encounter insufficient capacity (per-
The planning horizon consists of a sequence of skill stationary blocking event probability); for each
intervals, each on the order of a month or quarter engagement, the stationary probability that service-
and each modeled as a stationary loss network whose delivery arrivals (at the times these engagements need
fixed parameters change from one interval to the next to be delivered) find insufficient capacity for one
according to a general stochastic modulation process. or more required skills (per-engagement stationary
The length of each interval is sufficiently long for loss-risk probability); and the vector n of the num-
the multidimensional stochastic process modeling the ber of active service-delivery engagements of each
loss network to reach stationarity. We then formulate type in equilibrium. Each column of the skill capacity
and solve a capacity planning optimization problem requirements matrix represents the skill staffing tem-
to determine the skill capacity levels for our stochas- plate for each type of service engagement from our
tic loss network that maximize performance, in terms demand forecasting models. Then, by definition, the
of expected revenue, cost, and profit, over the entire state space corresponding to n is constrained by the
planning horizon. skill capacity vector and given by the polytope (1) of
Our derivations and theoretical results for the risk- the appendix.
based capacity planning models are based partly on When, as part of satisfying demand, the times at
the slice methods developed by Jung et al. (2008) which engagements need to be delivered are mod-
when restricted to Poisson arrival processes. In addi- eled as Poisson processes, then the above multidi-
tion to the innovative application of these results, mensional stochastic process reduces to the famous
we extend our risk-based capacity planning mod- Erlang loss model, which has been studied since the
els and methods to handle general arrival processes. seminal work of Erlang (1917), as Brockmeyer et al.
This, together with our stochastic risk-based opti- (1948) discuss. In this model, it is well known that
mization results and methods, provides theoretical the stationary distribution, representing the proba-
support for our approach to model and optimize over bility of the system being in state n in equilibrium,
the complex dynamics and uncertainty that character- is unique and exhibits an explicit solution in terms
ize capacity planning trade-offs in HCSCs. of a multivariate Poisson distribution constrained by
the polytope (1) through a normalizing constant that
Stochastic Risk-Based Models quantifies the total probability mass of this polytope.
We model each stationary period of the capacity plan- The stationary loss-risk probability for each engage-
ning problem as a stochastic loss network consisting ment type is directly expressed as a function of the
stationary distribution and model parameters. This in a large-network limiting regime, as originally stud-
loss-risk probability vector plays an important role ied by Kelly (1991), under which the scaled vec-
in our stochastic model and optimization as a mea- tors of skill capacities and engagement arrival rates
sure of the service engagements at risk of being lost, tend toward infinity in fixed proportion with respect
together with associated lost revenue estimates. to a scaling parameter. More precisely, the loss-risk
The complexity of computing the above normaliz- probabilities from the slice methods are shown to
ing constant is known to be ]P -complete (a higher converge to the exact loss-risk probabilities as this
complexity class than NP-complete) in the size of the scaling parameter tends to infinity. We further prove
network (Louth et al. 1994). Hence, an Erlang fixed- strict dominance of the accuracy of the slice meth-
point approximation (EFPA) has been long used as ods over previous Erlang approximations by deriving
an efficient alternative to the exact Erlang loss for- and comparing large deviations results for the exact
mula. The EFPA is based on approximating the block- Erlang formula, the EFPA in the large-network limit-
ing event probabilities of individual skills by a set ing regime, and our slice-based methods.
of fixed-point equations, and then approximating the We also develop accurate solutions for stochas-
loss-risk probabilities for each service engagement tic loss networks under general renewal arrival pro-
type in terms of these skill blocking event probabili- cesses. Specifically, we derive a Gaussian fixed-point
ties. Whitt (1985) and Kelly (1991) provide additional approximation based on the underlying multidimen-
details on the Erlang loss formula and EFPA. sional Gaussian process in which the engagement
Although EFPA resolves the prohibitive computa- loss-risk probabilities are given by Equations (7)–(10).
tional costs of the exact formula, it is well known that We further establish the asymptotic exactness of our
EFPA can provide relatively poor estimates for loss- Gaussian fixed-point approximation under the above
risk probabilities in various model instances, which large-network scaling.
we found to be the case upon application within Stochastic Risk-Based Optimization
the ITS HCSC. To address these challenges, we first Although stochastic loss networks model the dynam-
observe that by definition the mode of the stationary ics and complexities of risk-based capacity planning,
distribution of n relates to a solution of the optimiza- our objective is to determine the skill capacity lev-
tion problem (2) in the appendix. Then, for each ser- els that maximize business performance over the
vice engagement type, we derive a convex relaxation planning horizon. Focusing on a stationary interval
of the original optimization problem (2) restricted to of the planning horizon to simplify the presenta-
each number of active engagements (slice) within the tion, our stochastic capacity planning optimization
polytope (1). We solve this optimization problem to is based on revenues gained for delivering engage-
obtain the corresponding mode of the distribution for ments at the time of their required delivery and
each slice. on costs incurred for deploying skill capacity levels.
This yields the approximation in Equations (3)– The former expected realized revenues represent the
(6) for the engagement loss-risk probabilities by difference between expected demand revenues and
exploiting the definition (3) and the derived relation- expected revenues at risk of being lost. More precisely,
ship (6). To reduce the computational complexity of the formulation is given by the optimization prob-
this approach to that of EFPA, we also develop a lem (11) with our loss-risk probability approximations
three-point slice method in which the above convex as constraints. Additional constraints can be included
relaxation is solved for the upper and lower limits to guarantee desired serviceability levels or market
of the polytope and for the mode of the overall dis- share. We also consider formulations that maximize
tribution; the mode is then estimated for all other revenue or minimize cost subject to constraints on
slices between successive pairs of the three computed gross profit margins or cost and revenue targets. In
modes, as Jung et al. (2008) and Anselmi et al. (2009) each formulation, the stochastic loss network captures
discuss. various sources of uncertainty and fundamental rela-
It has been shown by Jung et al. (2008) that these tionships among demand, supply, and engagement
slice-based approximations are asymptotically exact loss-risk probabilities.
These formulations render a nonlinear program, reflect the combination of existing and retained skill
and our solution exploits properties of the underlying capacity and new skill capacity from hiring and train-
stochastic loss network with either Poisson or general ing actions. The constraints can also include a budget
renewal arrival processes. This includes establishing for hiring, training, and retaining costs, either jointly
that, for a stochastic loss network in the above large- or separately, which is especially useful in practice
network limiting regime, the solution of the stochas- when various business investment decisions are man-
tic optimization problem converges asymptotically, in aged by different organizations, each with its own
the limit as the scaling parameter tends to infinity, to budget.
the optimal solution under the exact engagement loss Finally, Bhadra et al. (2007) consider the structural
probabilities. Our experience in practice indicates that properties of a multiperiod stochastic loss network
the scale of the risk-based capacity planning optimiza- under Poisson arrivals whose solution is obtained
tion is sufficiently large to yield accurate solutions. via the EFPA. However, the resulting properties do
We also determine a region of skill capacities that not generally apply to our multiperiod stochastic loss
ensures the arrivals will be served with loss-risk prob- network under renewal arrival processes, nor to our
abilities no more than a given set of bounds. This solution via the slice methods.
region characterizes fundamental properties between
the capacity vector and the loss probability vector Illustrative Examples
relative to the arrival rate vector, which then can To illustrate the complexities of risk-based capacity
be exploited to efficiently search the feasibility space planning and the benefits of our approach, consider a
of our risk-based capacity planning problem. Specif- simple scenario of an application development (AD)
ically, for a stochastic loss network in the large- engagement that requires one Java programmer and
network limiting regime, we establish a system of one engagement manager, and a data administration
polynomial equations and inequalities that define a (DA) engagement that requires one database architect
region within which the skill capacity vector must fall and one engagement manager. The arrival epochs for
to achieve a given bound on the engagement loss-risk when AD and DA engagements need to be delivered
probability vector. Upon incorporating these polyno- follow independent Poisson processes with rates 20
mial equations and inequalities in our optimization and 10, respectively.
problem to characterize its feasible region, adding Generally speaking, the higher the skill capaci-
corresponding constraints on the loss-risk probability ties, the lower the engagement loss-risk probabili-
vector, and exploiting theoretical results established ties. However, the sharing of managers among both
by Lasserre (2001), we obtain a near-optimal solution engagement types can create complex interactions
with polynomial computational complexity and prob- among the skill capacities. In two instances of this
abilistic accuracy guarantees. example, which shed light on the complex dynam-
The preceding formulations consider idealized opti- ics in general, we observe that simply incrementing
mal skill capacity levels under only costs for deploy- the engagement manager capacity by one can result
ing skill capacity levels without addressing the costs in dramatically different loss-risk probabilities for a
of adjusting these capacity levels. The general formu- given demand, depending on the various parameter
lation of risk-based capacity planning optimization regions in which the system may operate. Figure 2
incorporates all sources of human-talent capacity and displays the corresponding results for both cases.
their associated costs. In particular, our general for- In Case 1, fix the Java programmer and database
mulation is given by the optimization problem (11) architect capacity to 10, and vary the engagement
and (12), which includes the costs for hiring addi- manager capacity from 16 to 26. We observe that,
tional (retaining existing) skill capacity and training although AD and DA loss-risk probabilities both
existing capacity to acquire skills, where existing decrease monotonically, the rates of decrease for the
capacity takes into account expected attrition and two engagement types differ considerably. We also see
internal transitions from supply evolution and opti- that for the same type of engagement, the rate can
mization. The skill capacity vector is augmented to vary significantly for different manager capacities.
(a) Case 1 Java programmers are a bottleneck for AD engage-

Loss risk probabilities and expected profit as a ments (i.e., an AD engagement is at risk of being lost
function of engagement manager capacity
0.7 most likely because of a lack of Java programmers).
Loss-risk probability
Eng. I
0.6 Eng. II
By increasing Java programmer capacity, we allow
more AD engagements to be accepted; however, this
0.5
0.4
causes greater utilization of the common (manager)
0.3
skill by AD engagements, and the DA loss-risk prob-
ability increases. Furthermore, these complexities sig-
0.2
16 17 18 19 20 21 22 23 24 25 26 nificantly impact the question of optimal skill capac-
14 ity levels, as we can observe from the expected profit
curves for both cases (see Figure 2).
Expected profit
12
More generally, in large real-world applications of
10 our risk-based capacity planning model and opti-
mization, it is quite difficult to distinguish the many
8
degrees of sharing among skills, given the com-
16 17 18 19 20 21 22 23 24 25 26
plex interactions and dependencies among the service
Capacity
engagements and their skill requirements. Hence, the
(b) Case 2 effects of changing skill capacity levels on engage-
Loss risk probabilities and expected profit as a ment loss-risk probabilities are often very compli-
function of Java programmer capacity
0.7 cated. These effects are further compounded by the
Loss-risk probability
0.6
Eng. I complex dynamics and interactions underlying the
Eng. II
0.5 optimization of skill capacity levels.
0.4 Finally, we consider a representative application
0.3 of risk-based capacity planning using ITS data. Fig-
0.2
ure 3 plots expected profits, revenues, and costs on
9 10 11 12 13 14 15 16 17 18 19 the y-axis as a function of risk tolerance constraints
14 along the x-axis. The leftmost set of results repre-
Expected profit
12
sents the solution that maximizes expected profit. The
risk tolerance constraints become more tight as we
10 move to the right along the x-axis, where the right-
8 most set of results represents when almost all of the
demand is being satisfied. We observe that although
9 10 11 12 13 14 15 16 17 18 19
Capacity expected revenues increase as more demand is sat-
isfied, expected costs also increase and tend to do
Figure 2: The graphs show an example of risk-based capacity planning.
so at a faster rate. This application illustrates how
executives can use risk-based capacity planning to
determine the best way to operate their business with
In Case 2, fix the engagement manager and respect to expected revenues, costs, and profits, rev-
database architect capacities to 20 and 10, and vary enue loss-risk tolerances, and other business and eco-
the Java programmer capacity from 9 to 19. This illus- nomic concerns.
trates that incrementing Java programmer capacity by
one can have a very different impact on the loss-risk
probabilities of both engagements types. Specifically, Supply Evolution and Optimization
incrementing one unit of Java programmer capac- The management of human capital supply over the
ity yields a decrease in the AD loss-risk probability, planning horizon is another fundamental aspect of
while increasing the DA loss-risk probability. Intu- HCSCs. Our goal is to understand the evolution of
itively, this is because under the parameter settings, human-talent levels over time and determine the best
45 Expected profit Expected revenue Expected labor cost

Revenue curve
40
Labor cost curve
35
30
($)
25
20 $39.0 $40.8
$35.8 $37.1 $36.4
$35.4
15 $30.7
$26.5 $28.0
$26.0
10
5 $9.4 $9.4 $9.1 $8.3

$4.4 Gross profit curve
Revenue at $5.56 M $5.2 M $3.9 M $2.0 M $0.2 M
risk (optimal profit) (all risk < 20%) (all risk < 10%) (all risk < 5%) (all risk < 0.5%)
Capacity 247 251 266 292 346
Figure 3: The graph shows an example business application of risk-based capacity planning.
actions to adjust these evolutionary dynamics. To business and economic conditions, and input from
highlight some difficulties involved, consider a com- subject matter experts. The dynamic evolution of
mon scenario in the military in which a long period human talent also can be influenced in desired direc-
is required for one to reach the highest ranks (e.g., tions by various investments and policies in response
General). Steps must be taken at all levels of this to uncertain and changing business and economic
HCSC (e.g., recruiting, training, experience, and pro- conditions. Hence, we formulate and solve a mul-
motions) to ensure that enough talent is available at tiperiod stochastic optimization (control) problem
all levels, with lower-level ranks feeding higher-level based on our supply evolution model to determine
ranks over time. A model of these complex dynam- the investment decisions and resulting evolution of
ics and stochastic behaviors is needed to determine talent levels over time that maximize business perfor-
appropriate talent levels that maximize expected per- mance. Such decisions in each period include hiring,
formance over time, subject to supply, demand, and training, promotions, other internal talent transitions,
organizational constraints. This challenge involves and incentivizing to reduce attrition, together with the
stochastic modeling and analysis of the evolutionary lead times associated with each action and policy.
dynamics and flexibility of human capital, including To model and optimize the HCSC evolutionary
dynamics at temporal granularities dictated by busi-
future internal transitions (e.g., promotions, certifi-
ness operations, we use a multiperiod discrete-time
cation, and training) and future external transitions
stochastic process defined over a planning hori-
(e.g., hiring, attrition, and acquisitions). The time scale
zon comprising T + 1 periods. Although our OTM
of these dynamics is typically on the order of days,
approach is completely general, the time granularity
weeks, or months, whereas the overall planning hori-
most often used is monthly and quarterly periods.
zon is often on the order of months, quarters, or years.
Therefore, we model the temporal evolution of Stochastic Evolution Models
human talent as a discrete-time, multidimensional Our stochastic supply evolution models are based
stochastic process in which the dynamics and uncer- on aggregating people into human capital groups
tainties vary over time. For this purpose, we lever- (HCGs) according to attributes of interest. Examples
age historical data on human capital dynamics and include all talent and skills relevant to the business,
and other factors such as levels of competency, pro- example, assuming a current count of Java program-
ductivity, proficiency, and certification. Because each mers at time 0, the number of Java programmers will
person comprising the supply side is capable of increase or decrease to a specific number of Java pro-
attaining and employing a collection of attributes, our grammers at time 1 with a corresponding set of proba-
supply evolution models are more precisely based bilities for hires and internal transitions into this HCG
on aggregating people into HCGs according to var- and for attrition and internal transitions out from this
ious combinations of attributes. Within each period HCG. Upon considering this portion of the stochas-
of the planning horizon, individuals make dynamic tic process for all possible populations and expanding
transitions between these HCGs. Such dynamic tran- this view to the entire stochastic process across the
sitions include both internal and external transitions large number of HCGs, we clearly have a prohibitive
that lead to a complex stochastic network topology, curse of dimensionality problem as the state space
both within each period and from one period to the blows up.
next across the planning horizon. To address this curse of dimensionality, we con-
Consider a model comprising five HCGs: C pro- sider a decomposition of the discrete-time stochastic
grammers, Java programmers, SQL programmers, process that records the expected number of people
programmers proficient in C and Java, and program- in each HCG. This renders a single state for each
mers proficient in Java and SQL. Now consider the HCG and an analogous set of probabilities that gov-
evolution of individuals among these HCGs over erns transitions from one state to another within each
a single period. People are hired into each group period and across periods that comprise the plan-
or attritted out of each group with certain proba- ning horizon. More formally, we consider the fam-
bilities, such as the probability that some number ily of subsets of the set of skills that people possess,
of C programmers are hired (attritted) within the in which case the types of people who comprise the
current period. Similarly, probabilities are associated supply consist of elements of this family. The state
with internal transitions among HCGs, such as the vector y models the expected number of people of
probability with which some number of C program- each type (i.e., the expected number of people in each
mers become equally proficient in Java programming. state) over a period. We then express in expectation
The corresponding sets of probabilities depend upon the number of people in each state at time t + 1 equal
the period in which the dynamic transitions occur to the number of people in the state at time t plus
and the number of C programmers at the start of the number of people hired into the state plus the
this period. We exploit properties of this complex net- number of people transitioning into the state minus
work topology to obtain scalable solutions for our the number of people transitioning out of the state,
stochastic supply evolution models; one example is either to another state or through attrition, each over
sparse matrix methods, given that transitions tend to the period t. This expression for the net dynamics of
be localized. human capital evolution in each period can be written
Our stochastic supply evolution models consist of in matrix form with respect to a set of one-step tran-
a discrete-time, multidimensional stochastic process sition probability matrices. From these evolutionary
that records the number of people in each HCG. Now dynamics and direct calculations, we derive the skill
consider a portion of this stochastic process corre- composition at a given time in matrix form yielding
sponding to the population of a single HCG (e.g., Equation (13) in the appendix. By having the model
Java programmers). At the start of the planning hori- variables depend on the period, our stochastic models
zon, we will know with certainty that the organiza- support time-varying behaviors of various forms (e.g.,
tion has a specific number of Java programmers, thus seasonal effects) for the evolution of HCSC dynamics.
yielding a single state for each HCG at time 0. At We investigated various approaches to parameter-
the next period, the state space of the stochastic pro- ize our stochastic temporal models, obtaining accu-
cess then needs to allow for all possible populations rate predictions through the inference of base-model
within the HCG of Java programmers, governed by parameters from historical HCSC data. For example,
the corresponding set of transition probabilities. For the elements of the one-step transition probability
matrices can be calculated from statistical data on the include the corresponding salary and benefits (e.g.,
number of transitions from one state to another state medical, pension) as a function of the state popula-
over each period. Other model parameters can be cal- tion, with the analogous per-period rewards including
culated in an analogous manner. We needed to adjust revenues driven through service delivery as a func-
such historical data in a few circumstances primar- tion of the population and demand for this state. Cost
ily because of events that were not representative of and reward functions are also associated with transi-
actual business trends (e.g., one-time acquisitions) or tions between states, where the costs and rewards for
a lack of statistical confidence on the samples for some transitioning from one state to another is a function of
states. In the first case, we worked with subject matter the states involved and the number of people making
experts to filter out the impact of these events from the transition. For example, the per-period costs and
the data to ensure that the statistics properly represent rewards for a set of C programmers becoming equally
the number of transitions with respect to future busi- proficient in Java programming include training costs
ness trends. We took an analogous approach for other and related service-delivery revenues. Our stochastic
events that might bias the statistics to deviate from supply evolution optimization models associate lead
actual business trends, for example, the use of rules to times for any available actions taken with respect to
factor out errors resulting from data entry mistakes. In each state transition (e.g., hiring). These lead times
the second case, we leverage standard statistical tests capture the delays between the time the action is ini-
for each state to verify a sufficiently large sample size tiated (e.g., hiring starts) and the time the result of the
to guarantee a desired margin of error. Whenever the action is actually realized (e.g., new employee comes
statistical test fails, we aggregate similar skills into a on board). Note that adjusting transition probabili-
combined state to obtain statistics whose sample size ties between states within the corresponding stochas-
is within the desired margin of error; upon solving the tic decision process represents investments, policies,
resulting stochastic temporal model, we probabilisti-
and actions such as training and promotion.
cally distribute the predicted human capital for this
Then, in addition to skill composition trajectory, our
combined state into its constituent states.
stochastic optimization models characterize the evo-
Stochastic Evolution Optimization lution of expected cost, revenue, and related financial
Although discrete-time, multidimensional stochastic metrics of the HCSC over time as functionals of the
processes model the dynamics and complexities of discrete-time stochastic process. Specifically, under
human capital evolution, our goal is to determine appropriate independence assumptions, the expected
the investment decisions and policies that influence cumulative costs of human capital over the planning
HCSC dynamics in desired directions over the plan- horizon are given by the sum of dot products of the
ning horizon. The driving objective is to maximize expected cost vector and state vector over all peri-
expected profit across all periods, where revenues ods. The expected cumulative revenues are obtained
depend upon demand and costs include maintain- in an analogous manner where the expected revenue
ing and achieving the skill capacity levels, noting for each period t is a function of y4t5, subject to the
that maximizing expected revenue or minimizing componentwise demand constraints. Expected profit
expected cost subject to constraints on gross profit over the planning horizon is computed as a function
margin or on cost and revenue targets is included of these revenue and cost metrics, including factors
within our solution framework and methodology. such as productivity and efficiency levels that depend
To this end, we associate costs and rewards with upon human capital type, workload, and utilization.
each state of our stochastic supply evolution models. Investment decisions and policies can be exploited
Namely, the costs and rewards for each state are cap- to impact HCSC financial metrics; therefore, we
tured as functionals of the number of people in the solve a corresponding stochastic optimization (con-
state and the amount of time they spend in the state. trol) problem based on the above decision processes
For example, the per-period costs for C program- to determine the optimal human capital evolution
mers at a specific experience and proficiency level over time. To this end, we express HCSC dynamics
160
150
Staffing level
140
Demand
130 Shortage–based
Partial information
Optimal
120
Sept Oct Nov Dec Jan Feb Mar
Figure 4: The graph shows an example of supply evolution and optimization.
as the discrete-time linear dynamical system in Equa- (see the Demand curve in Figure 4). This suggests
tion (14) in the appendix. Our goal is to determine that there will be a significant shortage in this skill
the vectors of decision variables for interstate transi- relative to demand under these retirement predic-
tional actions and for hiring and retention actions that tions. To mitigate this problem, we investigate the
maximize expected profit over the planning horizon. possibility of incentivizing some of this attrition to
More precisely, we consider a formulation of the opti- postpone retirement until enough people with this
mization problem (15) (see the appendix) in terms of skill can be brought on board. We model the behav-
expected profits as a function of the system state and ioral responses of baby-boomer retirees as a concave
expected decision costs. Because expected revenue for function of increasing incentives where the higher
each period is linear in the state populations up to the incentives, the larger the fraction of people who
the forecasted demand for skills within the period, are willing to postpone retirement, with diminishing
the objective function is (piecewise) linear in the state returns. For this representative scenario, our supply
and decision vectors. Hence, we stack the summation evolution optimization solution determines the set of
in the objective function (15) (see the appendix) and investments to maximize profit relative to forecasted
combine this with the linear dynamical system con- demand (see the Optimal curve in Figure 4).
straints, expressed in a corresponding matrix form, In contrast, many organizations would never real-
into a (piecewise) linear program with respect to the ize the upcoming shortage because of future retire-
combined decision and cost vectors. We additionally ments under this scenario until it is too late; thus, they
have, in a corresponding matrix form, constraints on will lose considerable revenue while trying to catch
balancing total flow, limiting total outflow, and lim- up over the hiring lead time, as the Shortage-based
iting total attrition of human capital, and restricting curve in Figure 4 shows. Somewhat more enlightened
revenue from exceeding demand. The output of this organizations might realize this upcoming shortage
(piecewise) linear program provides an optimal deci- in advance. However, they will hire enough people
sion vector from which we cull the optimal decision to fill this shortage as they believe to be the case
and state vectors for each period. in September without understanding that transitions
into (out from) this skill will naturally occur over the
Illustrative Examples
next few months; thus, they will hire too many (too
To illustrate the complexities of supply evolution and
few) people and create a future skill overage (short-
optimization and the benefits of our approach, con-
age) problem, as the Partial information curve in Fig-
sider a simple application scenario starting with the
ure 4 shows.
evolution of a single skill. Using historical data, we
predict a significant increase in attrition in Novem-
ber and December because of baby-boomer retire- Multiskill Shortages and Overages
ments. Furthermore, the demand forecast for this Given skill capacity targets and multiskill human cap-
skill is relatively steady for the subsequent months ital predictions, we determine an optimal matching of
the latter to the former. The planning horizon (on the Skill Shortage and Overage
order of a year) consists of a sequence of stationary Management
intervals, each on the order of a month or a quar-
Once we have obtained skill shortages and over-
ter. We formulate and solve our skill shortage and ages by solving the multiperiod version of the above
overage analysis across these stationary periods to
linear program, we next determine the investment

determine an optimal supply-demand matching that decisions to manage these shortages and overages
minimizes a weighted sum of shortages and over- over the planning horizon. Our approach is based
ages and compute the corresponding skill shortages on an iterative combination of the OTM optimiza-
and overages. Any predetermined assignments can tion capabilities. Initially, a first-phase application of
be incorporated as constraints; our solution then opti- our overall solution process results in: (1) risk-based
mally matches the remaining multiskill human capital capacity planning optimization, which provides ideal
to the remaining demand. skill capacity targets based on business and economic
We capture the fundamental aspects of this problem considerations and the HCSC dynamics within each
for each stationary period as a linear program. Our period; (2) supply evolution and optimization, which
objective is to determine the matchings of human cap- provides human capital capacity levels and required
ital possessing multiple skills against the skill capacity investments across future periods to realize these
targets to minimize a weighted sum of the skill short- capacity levels; and (3) multiskill shortage and over-
ages and overages. The shortages and overages must age analysis, which provides optimal matchings of
satisfy a balance equation in which the sum of skill multiskill human capital to skill capacity targets.
matchings and expected shortage equate to the sum The intraperiod dynamics are captured by each
instance of risk-based capacity planning optimization,
of corresponding skill targets and expected overage—
and the interperiod dynamics are captured by sup-
constraint (16) in the appendix must be satisfied. Then
ply evolution and optimization, whereas intraperiod
our solution to the multiskill shortage and overage
and interperiod dynamics are captured by single- and
analysis across multiple stationary intervals under a
multiple-period instances of shortage and overage
general stochastic modulation process is obtained by
analysis, respectively. This partitioning of the over-
solving the corresponding dynamic program. all OTM solution is based on various business and
Our use of weights for each skill shortage and over- organizational aspects of the ITS HCSC. Given these
age in the objective function helps to facilitate skill dependencies and interrelationships, we combine all
priorities and importance factors when matching mul- three optimization capabilities through a subsequent-
tiskill human capital to the skill capacity targets. For phase iterative procedure as follows. First, the costs
example, it might be desirable to push the optimal of hiring, training, and retaining are included in risk-
matching toward hot skills (to reduce shortages of based capacity planning optimization. The resulting
such skills), all else being equal, and similarly to push skill capacity output and the skill composition out-
the optimal matching away from commodity skills (to put of supply evolution and optimization are then
reduce overages of such skills). In addition, it can be optimally matched via multiskill shortage and over-
desirable to include preferences among the subsets of age analysis. This skill-matching output is provided
skills in multiskill supply-demand matching. To this as input to both risk-based capacity planning opti-
end, we introduce an additional family of variables mization and supply evolution and optimization for
and exploit the big-M method for solving linear pro- updates on the recommended investment decisions.
grams (Bertsimas and Tsitsiklis 1997). More precisely, The entire process is then repeated until all solutions
our complete general formulation for each station- converge to their equilibria.
ary interval of the multiskill shortage and overage
analysis with preferences is given by the linear pro- OTM Business Benefits
gram with objective (17) and constraints (16) in the The OTM suite has been implemented and deployed
appendix. in several ITS organizations worldwide to support
the management and planning of major portions of of our OR models and methods over the approaches
the ITS HCSC. The business processes supported by used prior to the OTM deployment.
OTM include engagement delivery, resource capacity OTM risk-based capacity planning is applied in three
planning, engagement portfolio management, sales- ways to provide distinct insights: (1) fix revenue tar-
delivery interlock, business decision support, and gets and determine the optimal skill capacity lev-
strategic planning. In this section, we summarize els that minimize cost and achieve the targets (or
our experience with the OTM worldwide deployment gross profit margins); (2) fix cost targets and deter-
over the past couple of years. We first discuss vali- mine the optimal skill capacity levels that maximize
dation studies and then present a representative set revenue and achieve the targets (or gross profit mar-
of business applications based on recent ITS data. We gins); (3) determine the optimal skill capacity lev-
conclude with reviews of the OTM impact on the ITS els that maximize profit. Now consider a quarterly
business, and ongoing efforts to support future IBM capacity planning problem with 132 skill types and
solution offerings. 216 engagement types under both risk-based capac-
ity planning and a previous approach based on linear
Validation Studies
skill-demand projections.
Our experience includes an analysis of the accuracy
In a cost-minimizing application, the same rev-
of each OTM capability on a regular basis. From
enue targets and similar aggregate loss-risk probabil-
the start, we instituted a feedback loop to validate
ities are realized under both solutions. However, the
and calibrate OTM predictions and results against
skill capacity portfolios differ considerably, with risk-
actual business outcomes and ITS data. This involves
based optimization providing lower loss probabilities
quarterly reviews with IBM executives to track and
for engagements with relatively high profit margins
evaluate the accuracy of the OTM methodology.
and lower capacity levels for engagements with rel-
Such validation studies demonstrate strong agree-
atively low profit margins. The risk-based optimiza-
ment between OTM model and optimization outputs
tion solution yields more than a 15 percent reduction
and the corresponding business outcomes realized in
in the individual loss-risk probabilities for different
practice. For example, we observe a typical accuracy
engagement types and more than a 40 percent reduc-
of 85–90 percent for our pipeline revenue forecasts,
tion in expected capacity costs.
and an accuracy generally higher than 90 percent for
In a revenue-maximizing application, risk-based
overall human capital demand forecasts (relative to
optimization increases the capacity levels for those
quarterly revenue targets). Our experience demon-
skills that are critical to engagements with relatively
strates loss-risk probability estimates from OTM risk-
high profit margins or relatively high usage, across a
based capacity planning models to be within a few
wide range of engagement types. Hence, the aggre-
percentage points of ITS data on real-world engage-
gate loss-risk probability is 5 percent under risk-based
ment losses. We observe the accuracy of OTM supply
optimization, as compared to 23 percent under the
evolution models to be within a few percentage points
linear-projection approach, which in turn renders a
in predicting skill compositions of large organizations
(with many thousands of people) over planning hori- 35 percent increase in expected revenue.
zons of up to one year. The accuracy of such predic- In a profit-maximizing application, risk-based opti-
tions for smaller organizations, correspondingly, are mization balances the trade-offs among engagement
within 12 percent of the real-world evolution of skill revenues and capacity costs, yielding a 100 per-
composition. cent increase in expected profit over the previous
approach. Although the risk-based solution provides
Representative Business Applications lower expected revenue, it also renders signifi-
The OTM deployment supports ITS management and cantly lower capacity levels than the linear-projection
planning processes. To highlight such applications of approach, both for individual skill capacity levels and
OTM in IT services delivery, we briefly present sce- in the aggregate. These cost reductions in turn yield
narios based on recent ITS data. These include rep- a significant increase in expected profit, as Figure 3
resentative comparisons that demonstrate the benefits similarly illustrated.
Skill 1 Skill 2 Skill 3

SEM staffing levels
SEM staffing levels
SEM staffing levels

Q0 Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 Q0 Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 Q0 Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8
Quarter Quarter Quarter
Skill 1 Skill 2 Skill 3

Staffing from linear projections
Q0 Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 Q0 Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 Q0 Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8
Quarter Quarter Quarter
Figure 5: The graphs compare supply evolution models (SEM) and linear projections.
OTM supply evolution and optimization supports by preserving various business cycles and dynamics,
supply-side decision management. Consider a skills as evidenced by the moderate degrees of periodicity
capacity prediction problem, based on an ITS data and growth in the supply predictions. However, the
set consisting of nearly 500 skills, under both our linear-projection approach leads to overstaffing and
supply evolution models and a previous approach understaffing because it ignores important dynamics
of linear skill capacity projections. The supply evo- such as the probabilities associated with internal tran-
lution models are parameterized by historical quar- sitions between states of the HCSC; for example, refer
terly hiring, attrition, and internal transition data, to skill 3 in Q4 and skill 1 in Q4 in Figure 5.
whereas the linear projections are calculated from his- Next, consider the corresponding optimization
torical quarterly hiring and attrition trends. Figure 5 problem, given demand forecasts for the next four
compares the staffing-level predictions for three skills quarters. Our supply evolution optimization uses
from both approaches looking eight quarters into the human capital costs and service-delivery revenues to
future. We observe that our supply evolution mod- determine the investment decisions and human-talent
els provide better predictions of future staffing levels evolution that maximize profit, in comparison with a
(a) Profit results 40 percent, our stochastic optimization provides an

Comparision of profits of optimal and myopic policies expected profit increase of nearly 50 percent. These
Profit under myopic policy profit improvements are related to our observations
Optimal expected profit that unnecessary responses to temporary dips in
demand can be counterproductive. Given the costs
of staffing decisions, the time-varying system dynam-

ics over the planning horizon, and the statistical
Expected profit
nature of demand forecasts, it is often preferable to

retain human capital in excess of demand estimates.
For example, Figure 6(b) depicts the significant over-
staffing and understaffing obtained with the previ-
ous myopic policy; refer to skill B in Q1 and Q3
and skill A in Q1 and Q4, respectively. This in turn
entails unnecessary investment costs and reduced
profit, which is clearly an undesirable solution from
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
both a financial and personal perspective.
Percent shortage fulfilment OTM shortage and overage analysis determines the
optimal matching of multiskill human talent against
(b) Staffing levels
capacity targets to minimize a weighted sum of skill
Skill A Skill B Skill C
shortages and overages. Consider a supply-demand
matching problem based on an ITS data set compris-
SEO staffing level
SEO staffing level

SEO staffing level
ing 950 employees and 307 skills to compare OTM

optimization with results from a previous approach
in which multiskill human capital are partitioned into
individual skills using historical averages of their skill
Q0 Q1 Q2 Q3 Q4 Q0 Q1 Q2 Q3 Q4 Q0 Q1 Q2 Q3 Q4
deployment in the prior quarter. This business appli-
Quarter Quarter Quarter cation shows that our shortage and overage analysis
yields a 25 (42) percent reduction in skill shortages
Skill A Skill B Skill C
(overages) and a 32 percent reduction in the sum of
Staffing under myopic policy
skill shortages and overages.

OTM shortage and overage management identifies
important business decisions with respect to impact
on expected revenues through higher skill capac-
ity levels and on expected costs to realize these
Q0 Q1 Q2 Q3 Q4 Q0 Q1 Q2 Q3 Q4 Q0 Q1 Q2 Q3 Q4
capacity levels. Applying this OTM capability to
Quarter Quarter Quarter the above business application yields a 200 percent
improvement in expected profit over that of previ-
Figure 6: The graphs compare supply evolution optimization (SEO) and an ous approaches. This solution represents a 7 per-
-myopic policy, ∈ 801 0011 0 0 0 1 0091 19. cent decrease in expected revenues relative to the
first-phase application of risk-based capacity plan-
ning, under which the costs to realize the correspond-
previously used myopic approach that adjusts capac- ing ideal skill capacity targets are ignored. When
ity levels up to a fixed fraction of the skill short- such costs are factored into the analysis, the ideal-
ages and overages. Figure 6 compares the expected ized risk-based capacity planning solution yields only
profits and staffing levels for three skills obtained a 177 percent improvement in expected profit over
under both approaches. We observe that the myopic that of the previous approaches. The additional rela-
policy exhibits an interesting concave behavior as tive expected profit improvements of 200 percent are
a function of . Even under the best setting of the direct result of our iterative methodology.
Business Impact levels of sales can be driven with the current sales
OTM’s business impact has been evaluated through force? How many financial analysts should be hired
quarterly reviews with IBM executives based on next year to deliver on a set of growth objectives?
comparisons with previous approaches used by ITS: As a result of these and related questions, the issue
(1) linear projections to forecast ongoing demand and of HCM is becoming one of the most important fac-
pipeline demand; (2) skill shortages and overages tors on CEO agendas. The ability to manage and
obtained by partitioning multiskill human capital plan human talent more effectively and efficiently is
among their individual skills using historical averages a critical driver of success for most services organiza-
of skill deployment in the previous quarter; (3) invest- tions, especially those with a large number of employ-
ment decisions to hire and train human talent to ees and a diverse portfolio of products and services.
reduce a certain fraction () of skill shortage and over- Analyst research indicates that, despite the majority
age estimates. In addition, the efficiency and efficacy of organizations making significant financial invest-
of our automated solution enables execution of OTM ments in the recruitment, training, and development
capabilities in response to unexpected events to adjust of their people, they spend up to an additional 8 per-
and fine-tune HCM investments and strategies. cent of total wages and salaries (on average) to man-
The quarterly executive reviews demonstrate age human-talent issues; many of these issues could
potential for reductions in skill shortages and over- be avoided or turned into bottom-line contributions
ages by 10–80 and 30–150 percent, respectively. The by exploiting advanced HCM solutions. The OTM
business benefits of OTM solutions over previous suite is an important and innovative step in this direc-
approaches are then computed based on: (1) increased tion. We continue to enrich our solution methodology
costs for human capital investments and decreased in collaboration with various IBM product groups to
profits for lower utilization under skill overages; create a general HCM platform for a broad client base.
and (2) increased costs for hiring and training and
decreased revenues for insufficient human capital to Appendix
fulfill the demand under skill shortages. The initial Risk-Based Capacity Planning
benefit comparisons reviewed with IBM executives Let I 2= set of skills, indexed by i, and K 2= set of ser-
demonstrated the financial impact in the first year vice engagements, indexed by k. Define Ei 2= station-
of OTM deployment to be over $11 million within a ary skill i blocking event probability, Lk 2= stationary
single quarter in the United States in terms of cost engagement k loss-risk probability, nk 2= number of
savings and increased revenue. Similar results have active engagements k in equilibrium, n 2= 4nk 5. By def-
been repeated within other ITS organizations in later inition, we then have n as an element of the polytope
stages of worldwide deployment, translating to rela- K
tive revenue-cost benefits of OTM solutions over pre- S4C5 2= n ∈ + 2 An ≤ C 1 (1)
vious approaches commensurate with 2–4 percent of
where A 2= 6Ai1 k 7, C 2= 4Ci 5, Ai1 k 2= amount of skill
ITS quarterly revenue targets.
i capacity required to deliver engagements k, Ci 2=
Although our focus has been on the ITS deploy-
amount of skill i capacity overall, ∀ i ∈ I1 ∀ k ∈ K.
ment, the possibilities for exploiting the OTM frame-
Assuming Poisson arrival processes with rate vec-
work and methodology in a broader application
tor Í = 4k 5, we address the challenges of EFPA by
setting are numerous. Concepts of labor and peo-
observing that the mode n∗ of the stationary distribu-
ple are central to all services organizations; compa-
tion Ï4n5 corresponds to a solution of the optimiza-
nies in health care, finance, insurance, retail, and the
tion problem
public sector are asking many of the same questions
addressed by the OTM solution. What is the outlook K
X
for a nursing workforce three years from now if cur- max nk log k − log nk ! s.t. n ∈ S4C51 (2)
n
k=1
rent hiring, training, and attrition patterns continue?
Will enough insurance agents with a particular set and by defining a continuous relaxation S̄4C5 2= 8x ∈
K
of skills be available to meet product demand? What + 2 Ax ≤ C9 of the polytope S4C5 and subsets of
this relaxation S̄`1 k 4C5 2= S̄4C5 ∩ 8x2 xk = `9, with x 2= possibly subject to constraints of the form Lk ≤ k ∈
4xk 5 the corresponding continuous relaxation of n. For 401 17, where uk is the base revenue rate for engage-
engagements k, we derive a convex relaxation of opti- ment k; vi is the base cost rate for skill i; viH , viT , viR are
mization problem (2) by exploiting Stirling’s approx- the cost rates for hiring, training, and retaining skill i
imation, ignoring O4log xk 5 terms, and restricting to capacity; CiE , CiH , CiT , CiR are the amounts of existing,
the slice nk = ` ∈ 8nk 2 n ∈ S4C59. We solve this opti- hired, trained, and retained capacity for skill i; and
mization problem to obtain the mode x∗ 4`1 k5 of the Ci = CiE + CiH + CiT + CiR ; for all i ∈ I and k ∈ K.
corresponding distribution, which we use to obtain
our approximation of 6nk = `7 as Supply Evolution and Optimization
X Let J 2= family of subsets of the set of skills I that
Ɛ6nk 7 = ` 6nk = `71 (3) are possessed by people, indexed by j. For all j ∈
`∈8nk 2 n∈S4C59
J1 t ∈ 801 0 0 0 1 T 9, define yj 4t5 2= expected number of
exp4q4x∗ 4`1 k555
P
`∈8nk 2 n∈S4C59 ` people in state j at time t, hj 4t5 2= expected amount
≈ P ∗ 4`1 k555
1 (4)
`∈8nk 2 n∈S4C59 exp4q4x of hiring into state j over 6t1 t + 15, aj 4t5 2= expected
K amount of attrition from state j over 6t1 t + 15, y4t5 2=
X
q4x5 = xk log k + xk − xk log xk 0 (5) 4yj 4t55, h4t5 2= 4hj 4t55, and a4t5 2= 4aj 4t55. In addi-
k=1 tion, define pj1 j 0 4t5 2= stationary probability of tran-
This yields an approximation for L 2= 4Lk 5 in terms of sitions from state j to state j 0 over 6t1 t + 15, where
P
Ɛ6nk 7 by exploiting the derived relationship j 0 ∈J pj1 j 0 4t5 ≤ 1. When state j attrition is strictly pos-
P
itive, this inequality is strict and 1 − j 0 ∈J pj1 j 0 4t5 rep-
Lk = 1 − Ɛ6nk 7/k 0 (6) resents the stationary probability that an individ-
Under general renewal arrival processes 4k 1 k2 5, ual leaves state j through attrition, such that aj 4t5 =
PJ
we derive our Gaussian fixed-point approximation yj 4t541 − j 0 =1 pj1 j 0 4t55. The dynamics of our supply
in which engagement loss-risk probabilities Lk are evolution models are derived in matrix form (using
expressed as row vector notation) to render
I t t t
41 − Ei 5Ai1 k 1
X Y Y
P4s 0 5 + y405
Y
Lk ≈ 1 − (7) y4t + 15 = h4s5 P4s51 (13)
i=1 s=0 s 0 =s+1 s=0
Ei ≈ G4i 1 i 1 Ci 51 (8) where P4t5 2= 6pj1 j 0 4t57, ∀ j1 j 0 ∈ J and ∀ t ∈ 801 0 0 0 1 T 9.

Define Cj 4t5 2= expected costs of state j over

C −
G41 1 C5 = 1 − ê 1 (9) 6t1 t + 15, Rj 4t5 2= expected revenues of state j over

K I
6t1 t + 15, Ã4t5 2= 4Cj 4t55, and Ò4t5 2= 4Rj 4t55. Then,
41 − Ei0 5Ai0 1 k 0 the expected cumulative costs of people over 601 T 5
X Y
i = 41 − Ei 5−1 Ai1 k k (10)
is given by Tt=1 6Ã4t5 · y4t57, under appropriate inde-
P
k=1 i0 =1
The one-dimensional Gaussian process in Equation pendence assumptions. The corresponding expected
(9) is for an isolated skill with capacity C, interarrival cumulative revenues and profits are obtained in an
mean (variance) −1 ( 2 ), and standard normal distri- analogous manner as functionals of both Ò4t5 and
bution function ê. y4t51 subject to the componentwise demand con-
The objective of our general stochastic risk-based straints. We rewrite Equation (13) into the follow-
capacity planning optimization problem is then ing discrete-time linear dynamical system recursion
given by (using column vector notation)
K I
X X y4t + 15 = y4t5 + B4t5u4t51 (14)
max uk 41 − Lk 5 k − vi Ci (11)
C
k=1 i=1
where B4t5 captures the sparsity patterns of P4t5, and
I I I
u4t5 2= vector of decision variables for transitions
viH CiH − viT CiT − viR CiR 1
X X X
− (12)
i=1 i=1 i=1
between states and for hiring and retention actions.
Our goal is to solve the stochastic optimization and programming (Bertsimas and Tsitsiklis 1997) and add
control problem MZi1 j + Zi1 j 0 in the objective function. More precisely,
our general multiskill shortage and overage optimiza-
TX
−1
min −Ð4t + 11 y4t + 155 + d4t5 · u4t51 (15) tion with preferences for each stationary interval of
u40510001u4T −15
t=0 the planning horizon consists of solving the linear
program
where Ð4t + 11 y4t + 155 2= expected profits at time
I
t + 1 as a function of the system state y4t + 15 sub-
4wiS Si + wiO Oi 5 +
X X
min 4Mi1 j1 j 0 Zi1 j + Zi1 j 0 51 (17)
ject to demand, and d4t5 2= expected decision costs at Mi1 j
i=1 i1 j1 j 0
time t, for all t ∈ 801 0 0 0 1 T − 19. Finally, we stack the
summation in the objective function (15) and formu- subject to the foregoing constraints, where wiS and wiO
late this optimization problem as a (piecewise) linear are the weights associated with skill i shortages and
program in terms of a decision variable vector z 2= overages. The solution to our overall multiskill short-
6y4151 0 0 0 1 y4T 51 u4051 0 0 0 1 u4T − 157 and a weight vec- age and overage analysis across multiple stationary
tor w 2= 6−Ð411 ·51 0 0 0 1 −Ð4T 1 ·51 d4051 0 0 0 1 d4T − 157, intervals under a general stochastic modulation pro-
subject to the constraints (14) expressed in a corre- cess is then obtained by solving the corresponding
sponding matrix form together with a related set of dynamic program.
constraints also in matrix form. The resulting opti-
mal decision vector z∗ yields the optimal state vec- Acknowledgments
We thank the following people for helpful comments
tors y∗ 4151 0 0 0 1 y∗ 4T 5 and optimal decision vectors on earlier drafts of the paper: B. Denton, J. Milne,
u∗ 4051 0 0 0 1 u∗ 4T − 15. D. Yao, S. Bollapragada, M. Campbell, A. Degbotse,
T. Ervolina, M. Ettl, and C. Hoffmann. We thank IBM exec-
Multiskill Shortages and Overages utives B. Dietrich, A. Greenland, S. Johnson, K. Kitagawa,
Define Ti 2= capacity target for skill i, Aj 2= expected S. Kumar, A. Manuel, and E. Tsukada for their support
number of available people with skill subset j, Si 2= throughout the Wagner Prize application process. Finally,
we thank the following people for their collaborations
expected shortage for skill i, Oi 2= expected overage and support throughout the development and deployment
for skill i, Mi1 j 2= expected number of people capable of OTM: R. Brown, M. Callahan, M. Collins, M. Eaton,
of employing skill subset j matched to employ skill A. Gill, S. Koh, W. Ng, A. Radovanović, B. Ramachandran,
i, for all i ∈ I, j ∈ J. Our objective is to determine L. Sproull, A. Waechter, and M. Yan. The comments of the
the matchings Mi1 j of Aj against Ti to minimize a guest editor and editor in chief helped turn very different
earlier versions of the paper into its more readable, so it is
weighted sum of Si and Oi . People cannot be matched hoped, present form.
to a skill they do not possess; thus, Mi1 j = 0 for all
i ∈ I that are not an element of j ∈ J, with Mi1 j ≥ 0
otherwise. All people need to be matched, implying References
PI Anselmi, J., Y. Lu, M. Sharma, M. S. Squillante. 2009. Improved
i=1 Mi1 j = Aj . The shortages and overages must sat-
approximations for the Erlang loss model. Queueing Systems
isfy the balance equations Theory Appl. 63(1–4) 217–239.
Bertsimas, D., J. N. Tsitsiklis. 1997. Introduction to Linear Optimiza-
J tion. Athena Scientific, Belmont, MA.
X
Mi1 j + Si = Ti + Oi 1 ∀ i ∈ I0 (16) Bhadra, S., Y. Lu, M. S. Squillante. 2007. Optimal capacity plan-
j=1 ning in stochastic loss networks with time-varying workloads.
Proc. ACM SIGMETRICS ’07, San Diego. Association for Com-
To support preferences among subsets of skills, we puting Machinery, New York, 227–238.
Brockmeyer, E., H. L. Halstrom, A. Jensen, eds. 1948. The Life and
define Zi1 j 2= remaining number of people capable of Works of A. K. Erlang. Academy of Technical Sciences, Copen-
employing skill subset j ∈ J not matched to employ hagen.
Cao, H., J. Hu, C. Jiang, T. Kumar, T.-H. Li, Y. Liu, Y. Lu, et al. 2010.
skill i ∈ I. This expands the above matching con-
PI Integrated stochastic resource planning of human capital sup-
straints to i=1 Mi1 j + Zi1 j = Aj , where Zi1 j = 0 for all ply chains. Technical report, IBM Research Division, Yorktown
i ∈ I that are not elements of the subset j ∈ J and Heights, NY.
Erlang, A. K. 1917. Solution of some problems in the theory of
Zi1 j ≥ 0 otherwise. Then, when j is preferred over j 0 probabilities of significance in automatic telephone exchanges.
for skill i, we exploit the big-M method of linear Elektroteknikeren 13 5–22.
Hu, J., B. K. Ray, M. Singh. 2007. Statistical methods for auto- Proc. ACM SIGMETRICS ’08, Annapolis, MD. Association for
mated generation of service engagement staffing plans. IBM Computing Machinery, New York, 407–418.
J. Res. Development 51(3) 281–293. Kelly, F. P. 1991. Loss networks. Ann. Appl. Probab. 1(3) 319–378.
Hu, J., M. Singh, A. Mojsilovic. 2008a. Categorization using semi- Lasserre, J. B. 2001. Global optimization with polynomials and the
supervised clustering. Proc. ICPR 2008. Institute of Electrical problems of moments. SIAM J. Optim. 11(3) 796–817.
and Electronics Engineers, New York, 1–4. Louth, G., M. Mitzenmacher, F. P. Kelly. 1994. Computational com-
Hu, J., M. Singh, A. Mojsilovic. 2008b. Using data mining for plexity of loss networks. Theoret. Comput. Sci. 125(1) 45–59.
accurate resource and skill demand forecasting in services Whitt, W. 1985. Blocking when service is required from several
engagements. Proc. KDD Workshop on Data Mining for Business facilities simultaneously. AT&T Bell Laboratories Tech. J. 64(8)
Applications. Association for Computing Machinery, New York, 1807–1856.
12–17. Woods, R. A. 2009. Industry output and employment projections to
Jung, K., Y. Lu, D. Shah, M. Sharma, M. S. Squillante. 2008. 2018. Monthly Labor Rev. 132(11) 53–81. Bureau of Labor Statis-
Revisiting stochastic loss networks: Structures and algorithms. tics, U.S. Department of Labor, Washington, DC.
View publication stats

Onthemark: Integrated Stochastic Resource Planning of Human Capital Supply Chains

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Onthemark: Integrated Stochastic Resource Planning of Human Capital Supply Chains

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

OnTheMark: Integrated Stochastic Resource Planning of Human Capital Supply

Article in Interfaces · October 2011

SEE PROFILE SEE PROFILE

Chen Jiang Tarun Kumar

SEE PROFILE SEE PROFILE

M.V.Sc. View project

Modeling using queueing theory View project

The user has requested enhancement of the downloaded file.

OnTheMark: Integrated Stochastic Resource Planning

O ver the past 50 years, the services sector has

Skill shortage and overage management V

(Determine skill capacity target)

(a) Case 1 Java programmers are a bottleneck for AD engage-

45 Expected profit Expected revenue Expected labor cost

5 $9.4 $9.4 $9.1 $8.3

Figure 4: The graph shows an example of supply evolution and optimization.

linear program, we next determine the investment

Skill 1 Skill 2 Skill 3

SEM staffing levels

SEM staffing levels

Skill 1 Skill 2 Skill 3

Staffing from linear projections

Staffing from linear projections

(a) Profit results 40 percent, our stochastic optimization provides an

of staffing decisions, the time-varying system dynam-

nature of demand forecasts, it is often preferable to

SEO staffing level

ing 950 employees and 307 skills to compare OTM

Staffing under myopic policy

Staffing under myopic policy

skill shortages and overages.

Ei ≈ G4i 1 i 1 Ci 51 (8) where P4t5 2= 6pj1 j 0 4t57, ∀ j1 j 0 ∈ J and ∀ t ∈ 801 0 0 0 1 T 9.

View publication stats

You might also like

Ei ≈ G4i 1 i 1 Ci 51 (8) where P4t5 2= 6pj1 j 0 4t57, ∀ j1 j 0 ∈ J and ∀ t ∈ 801 0 0 0 1 T 9.