Strategic Workforce Planning: A Rigorous Simulation Optimization Approach

Marco Better, Fred Glover, Dave Sutherland, and Manuel Laguna

INTRODUCTION

CEOS CONSIDER MANAGING TALENT A PRIORITY, as well as a top business challenge for the future—second only to managing business growth.¹ Business leaders recognize that their organization is only as good as its talent, and success depends on having the right people in the right place at the right time—a concept we call readiness. Achieving high workforce readiness requires the ability to anticipate changing workforce needs, as well as to allocate resources as effectively as possible in meeting those needs. While talent is a top priority, few organizations manage it as strategically as they do their financial and physical assets or their customer requirements.

In most organizations, workforce planning is in its infancy, if it is done at all, and the tools and analytics used to support talent management decisions are not nearly as advanced as they are in other disciplines (e.g., there’s no equivalent of a cash flow model or operations plan). Yet the complexity of the task is enormous! The pace of change within economies, industries, and organizations continues to accelerate; labor markets continue to become more competitive and more global; and the workforce continues to become more diverse in terms of its demographics, expectations, and goals.

Advanced workforce planning and talent management tools are needed to enable organizations to determine, among other things:

The effect of changes in business strategy and customer demand on workforce needs

Retirement, turnover, and undesired talent loss risks

Cost trade-offs between higher retention and recruitment to close gaps in critical segments of the workforce

The effects on cost and gaps of changing the skill mix of the workforce

The set of retention and recruitment programs that will maximize readiness at minimum cost

The bottom line is: If HR is to have a credible place at the strategic planning table, then human capital decisions must be made based on data and analytics instead of relying solely on anecdotes and assumptions. Experts and thought leaders in the field of Strategic Workforce Planning—among them practitioners, consultants, and academics—are increasingly advocating the use of advanced analytics and decision sciences in their field and in the talent management arena in general.²

Simulation and optimization technology can answer this call. Organizations use it to manage risky portfolios of projects and securities, optimize their business processes, and develop new products and services, among other uses. Organizations that are willing to apply this level of rigor to their human capital decisions will create a distinct competitive advantage.

Correctly applied to workforce planning, this technology enables organizations to optimize readiness (right people, right place, right time, and right costs) and representation (diversity) within defined constraints (budget dollars, scarce skills). Whereas traditional approaches limit their scope to projecting future workforce requirements based on static assumptions, the combination of simulation and optimization provides decision-making tools that support the development and implementation of strategies, programs, and policies to meet those requirements.

Think of budget dollars allocated to buckets representing specific practices (policies, programs, initiatives) used to attract and retain valued employees. Then, envision a dial beneath each bucket. You can turn the dial to increase or decrease the resources allocated to each bucket until you find the allocation of resources that is most likely to enable you to achieve specific goals (e.g., readiness, retention, cost), recognizing constraints and considering the demographics of your population.

Other applications include but are not limited to:

Identifying the most effective recruiting channels for the organization

Modeling the cost-effectiveness and risk of using contingent versus regular staff

Supporting the budgeting process by defining and communicating trade-offs between readiness and costs

Modeling the likely impact of compensation strategies

Identifying appropriate bench strength in key areas, given workforce mobility and associated costs

Ultimately, we believe simulation optimization can be applied to any strategic human capital decision-making process. A unique type of simulation modeling, known as “agent-based” simulation, is particularly well suited for this type of application. In this chapter, we describe the components of an agent-based simulation optimization application called OptForce, and how this approach is used to optimize workforce planning decisions.³ We begin by describing the agent-based simulation model; we then describe the optimization approach; next, we provide a use-case summary of the technology from one of our marquee customers, CH2M Hill; and we conclude with closing remarks about the benefits of our recommended approach.

AGENT-BASED SIMULATION

Agent-based simulation is a modeling technique where key entities (i.e., employees, passengers, etc.) are represented by computerized agents whose behavior is influenced by other agents and their environment. In SWP, agents represent employees whose behavior relates to the likelihood that they will stay in the organization, be promoted or transferred, become good performers, etc., based on environmental stimuli; such stimuli comes from within the organization, in the form of talent programs and practices designed to attract, retain, and promote target employees, or from the external environment, such as the state of the economy, the job market, and the competition for desired talent.

The agent-based model includes accurate forecasting methods for employee retention, not only in the aggregate, but at very granular levels, based on employees’ unique characteristics (such as age, gender, marital status) and career attributes (such as tenure, experience, role in the organization, and performance, to name a few).

The forecasting technique must be dynamic and adapt to changes in the employee’s attributes, as the employee’s age advances, tenure increases, or role changes after a promotion or transfer. The model must also adapt to changes in assumptions about the economy, the job market, and any programs and policies the organization plans to put in place. Thus, the model can be used for scenario planning and “what-if” analysis to assess the sensitivity of the workforce to different sets of assumptions.

In the agent-based model, employees interact with the environment and periodically make decisions about their career in the organization. These decisions are based on their perceptions of the degree to which the organization is meeting their needs, and the likelihood it will in the future. Ultimately, the objective of the model is to determine future workforce needs, how well the current and projected workforce meets those needs, and how to close any resulting gaps.

Simulation in Action

Figure 1 provides a graphical representation of the workforce simulation process. The simulation runs for a number of measurement periods that may be expressed in months, quarters, years, etc. During each period, the following events occur:

1. Each employee makes a decision about whether to stay in or leave the organization. This decision depends on the employee’s retention probability, which is obtained from the retention forecasting model.

2. Employees who remain in the organization are assigned to available jobs, contingent upon a match between their attributes and stated job requirements and qualifications.

3. Remaining jobs are filled by employees promoted according to eligibility criteria or probability of mobility/promotion.

4. New employees are recruited from the appropriate sourcing channels to fill any remaining available jobs, as allowed by budget limits.

Consider the process in Figure 1. Circles represent employees and rectangles represent jobs. The top section represents executive-level employees, the center section represents middle management, and the bottom section represents non-managerial-level employees. Finally, circles represent external hires. In this example, the employee decision and job assignment processes occur annually for three years.

FIGURE 1. REPRESENTATION OF THE OPTFORCE SIMULATION PROCESS.

In Figure 1, the initial workforce is composed of two executives, three middle managers, and four non-managerial employees. However, during Year 1, one executive leaves the organization, as depicted by the dashed arrow leading out of the executive section. Remaining employees are assigned to available jobs. In addition, one middle manager is promoted into an executive-level job, and one non-manager is promoted into a middle management job, as depicted by the dashed arrows going from the middle management to the executive section. Finally, a new employee is hired to fill an available non-managerial position.

During Year 2, one middle manager and one non-manager leave; one non-manager is promoted into a middle manager job; and three new employees are hired. During this year, an additional non-managerial job is created, but it remains unfilled due to budget limits.

During Year 3, one executive leaves; there are no promotions; and a new middle management job and a new non-management job are created, requiring three new employees to be hired.

The basic steps in building an agent-based model for SWP are:

1. Defining future workforce requirements (demand planning)

2. Defining key attributes relevant in forecasting employee behavior (retention forecasting)

3. Identifying current and proposed talent policies, programs, and initiatives designed to influence employee retention and mobility (internal supply planning)

4. Determining the impact of each policy, program, and initiative as well as the effect of different economic conditions on retention of employees with different attributes (sensitivity analysis—predictive analytics)

5. Defining current and potential sourcing channels (external supply planning)

6. Defining eligibility and constraints with respect to promotion and mobility within the organization (career development)

These steps are designed to be able to (1) define future workforce requirements based on the organization’s strategic plan, (2) determine how the forecasted workforce meets those requirements, and (3) decide how best to close the gap between (1) and (2).

Defining Future Workforce Requirements. Defining future workforce requirements serves as the foundation for effective Strategic Workforce Planning. This estimation of workforce demand is used to perform so-called gap analysis that identifies the difference between what the organization needs and what it actually has or is projected to have in future periods. Workforce demand planning involves translating business plans into a forecast of specific workforce requirements—number of positions, headcount, skills, timing, location, etc.—and identifying factors that could change the required profile so that contingency plans can be developed.

The most common methods to forecast future workforce needs include aggregation of line manager “guesses,” past trends in workforce growth to forecast future growth, and correlation of corporate metrics to historical workforce needs. Such demand metrics may be tied directly to financial projections such as sales, revenue, and gross margin. Alternately, the metrics may relate to operational goals, such as volume, throughput, and cycle time. Metrics may drive the growth in one or more roles in the organization. For example, let’s assume that the need for doctors and nurses in a hospital ward is tied to the number of beds. The hospital projects a certain growth rate in the number of beds. The growth rate may be constant over time, or it may vary. Let’s imagine that there are two expansions planned, respectively one and two years from now, and that the number of beds grows constantly at a 4 percent annual rate between expansions. Supposing the ward currently has thirty beds, the growth would be as shown in Table 1.

TABLE 1. DEMAND METRIC BASED ON NUMBER OF BEDS.

Now let’s assume that the number of nurses grows at the same rate as the number of beds, but the number of doctors grows at a rate that is 50 percent of the growth in number of beds (i.e., they can attend to two patients at a time). Assuming the ward currently has ten nurses and five doctors, by associating the above metric, we would project the demand for nurses and doctors to be as stated in Table 2.

While historical data might be effective to project future needs for roles that are stable, they cannot be the main driving factor that determines future workforce needs in a dynamic environment. For the process to be effective, corporate strategy must be translated into workforce requirements. What is needed is an understanding of the nature of the work required to meet strategic goals. The main source of information is not historical data but the organization’s strategic plan. For lack of better methods, focus groups, surveys, and the Delphi technique are often used as tools to accomplish this task. We believe that a more rigorous approach is needed.

TABLE 2. RESULTING DEMAND FOR NURSES AND DOCTORS.

We consider both internal and external factors that affect how a particular workforce mix performs against the goals of the organization. Relevant factors include the business environment, market and economic conditions, and technology enhancements. We assume that the exact way in which these factors affect the workforce mix is not known with complete certainty, but it can be estimated and turned into probabilistic models. Specifically, it is possible to model the relationships between a workforce mix and a set of factors to estimate the degree by which corporate goals are satisfied.

Let us consider an example of a claims handling process in an insurance company. For this example, we assume a three-year planning horizon. We also assume that we have five different types of claims, ten tasks necessary to handle those claims, and employees in five different roles and skill levels to perform those tasks. Table 3 shows the expected volume of each claim during the next three years.

TABLE 3. EXPECTED VOLUME OF CLAIMS, BY TYPE.

In order to handle each type of claim, several tasks need to be accomplished. Table 4 shows that, for instance, Tasks T01, T03–T07, and T09–T10 are required to process claims of type 1. Each entry in Table 4 indicates the number of minutes of each task that are necessary to process a single instance of the corresponding claim.

Next is the availability and commitment of claims processing personnel to the five types of claims included in the planning model. The example considers five types of roles, and their availability and commitment are given in Table 5.

Personnel in all roles are scheduled to work 200 days per year and eight hours per day. The percentages under each year reflect the commitment of personnel in a role to the five claim types in the model. Variations from 100 percent reflect indirect activities such as overhead not included in the list of tasks. To complete the input data, we need to determine task data consisting of the distribution of a task by role. Each entry in Table 6 indicates the percentage of the task that is performed by a worker in each of role category.

TABLE 4. TASKS REQUIRED PER CLAIM TYPE.

For instance, Task T01 is performed by personnel in Roles 1 and 2 in equal parts. Other tasks require work by other roles, and the assumption in this example is a worker in one role will not perform an activity that requires the skill level of another role.

For each year, the total number of minutes per task is distributed (using the percentages in Table 6) among the different roles. The total number of minutes is simply the sum product of the demands and the task requirement per claim. For instance, the total number of minutes of task T01 in Year 1 is given by:

TABLE 5. AVAILABILITY AND COMMITMENT BY ROLE.

TABLE 6. PERCENTAGE OF TASK PERFORMED BY EACH ROLE.

T01 total = 20,000 (Claim 1) * 22 (min) + 15,500 (Claim 2) * 33 (min) + 50,000 (Claim 3)*

18 (min) + 37,000 (Claim 4) * 28 (min) + 43,000 (Claim 5) * 26 (min) = 4,005,500 minutes

The distribution of the total number of minutes among the skill levels involved in performing task T01 is taken from Table 7. In the case of task T01, the calculation is:

Role 1 time = 4,005,500 (min) * 50% = 2,002,750 minutes

Role 2 time = 4,005,500 (min) * 50% = 2,002,750 minutes

The time requirement per role is simply the sum of the times per task required for that role, as shown in Table 7. The total workload for Role 1 in Year 1 is given by:

Role 1 workload = 2,002,750 + 246,000 + 120,325 = 2,369,075 minutes

Workforce requirements are based on both time requirements and talent availability. A summary of the forecasted requirements is given in Table 8 and calculated as follows. A total of 2,369,075 minutes are required of Role 1 in Year 1, as shown in the Table 7. To find the number of workers needed, the availability of a single Role 1 worker in Year 1 is calculated.

TABLE 7. WORKLOAD FORECASTS.

TABLE 8. WORKFORCE REQUIREMENTS.

Role 1 availability in Year 1 = 200 (days/year) * 8 (hours/day) * 60 (min/hr)*

60% (availability) = 57,600 minutes /worker

Therefore, the number of workers is:

Role 1 requirements = 2,369,075 (min) / 57,600 (min/worker) = 41.13 workers

Note that the workforce requirements exhibit 0.86 percent growth from Year 1 to 2 and 16.8 percent from Year 2 to 3. This is driven by the projected growth in the demand for claims. However, due to task requirements and labor commitments, both growth patterns are not proportional as the growth in claims demand from Year 1 to Year 2 is 9.7 percent and from Year 2 to Year 3 is 19.6 percent. However, the growth in workforce requirements for those years is 0.86 percent and 16.8 percent, respectively.

A Simulation Optimization Approach to Demand Planning

The demand for products or services is the main source of uncertainty in the example described above; this is particularly true for new products or services. To deal with this uncertainty, a Monte Carlo simulation is used to treat the demand assumptions as random variables instead of point estimates. Each random variable follows some specified probability distribution function. In our example, we assume that the demand for Claim 1 follows a normal distribution with a mean value given by the expected values shown in Table 5. Similar distributions are used to model demand of the remaining claim types with larger standard deviations representing greater uncertainty. Because, typically, demands among products are not independent (for example, during a strong economy, it may be expected that the demand will be relatively high for most or all claims), correlation among demands should be considered in the model.

The uncertainty of key data makes the Monte Carlo simulation an attractive tool to embed in a simulation optimization process as depicted in Figure 2.

FIGURE 2. DEPICTION OF A SIMULATION OPTIMIZATION PROCESS FOR FORECASTING WORKFORCE REQUIREMENTS.

The workforce mix in Figure 2 consists of the number of workers required for each period of the planning horizon, at each skill level, and in each location. The optimal mix—that is, the one that satisfies the corporate goals to the highest degree—becomes the workforce demand that will be used in subsequent steps of the company’s workforce planning exercise. The optimization procedure can be a general-purpose optimization algorithm based on metaheuristic methods, such as OptQuest,⁴ embedded in most commercially available Monte Carlo simulation tools.

For the process to work properly, the degree of satisfaction of corporate goals must include some trade-off component, such as return on investment. In other words, the process should take into consideration the cost of the workforce. For instance, the optimization model may be configured to maximize the degree of satisfaction of corporate goals at minimum cost.

Defining the Key Attributes Relevant in Forecasting Employee Behavior. The second step in defining the model is to identify key employee attributes to consider. Attributes describe the characteristics of an employee, such as age, gender, ethnicity, work experience, education, and performance rating. Attribute values are used to classify employees for the purpose of assessing differences in retention behavior and the impact of different HR decisions on different groups of employees. For instance, we may want to track employees by two attributes: gender and age. Then, within gender, we have two values: male and female; within age, we have four values: veterans, baby boomers, Generation X, and Generation Y.

At the core of the agent-based model resides an accurate predictive analytics module, based on decision tree analysis, which we call Retention Rate Tree Analysis. More sophisticated than classification and regression trees, this Retention Tree not only partitions historic data into groups with homogenous retention behavior, but also recognizes trends in behavior within partitions. Starting from a root node, the tree splits on attributes most relevant in explaining differences in retention behavior. The tree-building algorithm selects the best attribute to split on at each level and branch, as well as the optimal splits of the attribute, based on preselected information gain metrics.

The tree uses historic employee data as well as external macroeconomic data to conduct multifactor analysis in order to accurately predict future employee retention behavior. As a result, the agent-based simulation is not only able to simulate the behavior of individual employees, but is also able to determine the key factors impacting an individual employees’ behavior based on his unique set of attributes. An example of a rate tree is depicted in Figure 3.

The root node of the tree represents employees of a particular organization over some time span (e.g., 233,146 cases in five years), with an average annual retention rate of 90.85 percent; however, if we split the population into five tenure bands, we see great variability in retention behavior. In fact, we see there is a much higher than average attrition rate in the 0–3 years-of-service band, while all other tenure bands have above-average retention rates. Managers can target this particular concern with the implementation of a better onboarding program to stem the high attrition of new employees. Deeper examination of the tree could result in better insights into retention behavior.

FIGURE 3. RATE TREE DEPICTING RETENTION BY FIVE TENURE BANDS EXPRESSED IN TERMS OF YEARS OF SERVICE (YOS).

In Figure 4, the tree from Figure 3 has been expanded to include the organizational division as well as the job level, summarized in terms of bottom, middle, and top. (The figure depicts only part of the complete tree in order to make our point here.) Drilling down into Division “C” of the enterprise, we see that it is not every type of new employee that exhibits low retention, but only those at lower-level jobs, so a less costly onboarding program than previously considered may be sufficient, targeting only this subpopulation.

FIGURE 4. RATE TREE DEPICTING RETENTION BEHAVIOR BY DIVISION, TENURE BANDS, AND JOB LEVEL.

This feature, more rigorous in nature than other workforce planning solutions, enables decision makers and workforce planners to identify clearly actionable decisions and investments and evaluate their impact on the shape of their workforce of the future.

The retention rates at any node of the tree are used as retention probabilities in the agent-based model, to simulate the probability that an employee with the set of attributes implied by their tree node will stay in the organization from one period to the next.

Identifying Current and Potential Talent Practices. The third step in the planning process is to develop a comprehensive inventory of practices (i.e., formal or informal programs and policies) currently in place that impact attraction, movement, and retention, as well as any proposed modifications to current practices and any practices being considered for future implementation. In order to track costs effectively, costs per employee, as well as any program management costs, should be accounted for.

Determining the Impact of Each Practice on Employees with Different Attributes. The impact of each practice on an employee’s behavior is determined based on relevant employee attributes. In the absence of solid historical data and/or external benchmark data, we assume the organization can obtain anecdotal data and informed judgment as to the expected impact of different practices on employees with specific attributes. Like the development of the workforce profile, this is an area where it may be wisest to start simple and build sophistication over time.

As an example, if the organization were to implement a policy that allows for flex-time, we would predict a highly positive impact on the retention rate of female Generation Y employees, whereas we would expect little or no effect on the retention of male baby boomers. Ultimately, the set of attributes chosen to describe the employee population should be selected according to the following criterion: Does the impact of any of the current or potential practices vary significantly by this attribute? If the answer is yes, then the attribute should be included in the model.

Defining Current and Potential Sourcing Channels. In addition to considering the impact of talent practices, it is also necessary to consider the effectiveness of alternate sourcing channels in bringing employees into the organization. For each current and potential sourcing channel, the following parameters are defined:

A current distribution of the population with the channel, as defined by key employee attributes

A cost-per-hire for the channel

An effectiveness factor for that channel

A maximum number of new hires that can be obtained from that channel

These data can be obtained from the company’s recruitment history complemented by external, publicly available data for common channels (e.g., Bureau of Labor Statistics databases, university graduation rates, online job sites), but parameters related to effectiveness and cost will vary by organization.

The probability distribution of the population in a channel represents the likelihood that a new hire will have certain attributes. For example, according to the National Center for Education Statistics⁶ of the Department of Education, the distribution of the population of graduating seniors in all public colleges and universities in the United States, is shown in Table 9.

TABLE 9. DEMOGRAPHICS OF GRADUATING SENIORS AT U.S. COLLEGES AND UNIVERSITIES.

These data are entered into a sourcing channel labeled, for instance, “General Colleges and Universities,” so that during a simulation, when a new hire is drawn from this population, the likelihood of hiring a minority female, for example, would be about 16 percent.

The cost-per-hire figure for the channel is expected cost to the organization to hire a new employee through that particular channel. It includes setup costs (e.g., travel costs, setting up a booth at a job fair), advertising costs, recruiting costs (e.g., recruiters’ time, managers’ time in interviews), agency fees, employee referral fees, relocation expenses, and signing bonuses. If the organization cannot calculate cost-per-hire for each channel but has a good estimate of average cost-per-hire by job level, then each channel’s cost-per-hire figure can be computed by multiplying the cost-per-hire times the effectiveness factor.

The effectiveness factor represents the effectiveness of the channel in yielding qualified candidates. It is multidimensional and can consider such factors as percent of jobs filled, offers as a percent of interviews, first-year retention rates, and offer acceptance rates. Effectiveness can be measured in many ways, but it is important that the calculation be consistent across all channels.

Finally, the organization estimates the maximum number of new hires it expects to get from each channel, for each role. Ideally, this information will be based on authoritative data sources, such as the Bureau of Labor Statistics, or from verified data consolidation vendors⁷ and adjusted to reflect expected future state. In many cases, however, they are based on the best judgment of in-house recruiting experts.

The data in the recruitment channels is used to simulate new hires.

Defining Assumptions with Respect to Promotion and Movement Within the Organization. The last step in setting up the model relates to the mobility of employees within the organization, in terms of promotions, job and location changes, etc. One of the attributes associated with each employee is job level, which may be defined generically for the entire organization or as defined career paths within job families. Mobility is modeled based on eligibility rules or historic rates to predict the likelihood that employees with particular attributes will move within the organization during the measurement time frame.

Table 10 shows an example of a mobility probability table for a services company. In this example, employees are described by tenure, job level, and performance rating, and an advancement probability is computed for each employee according to their attributes. The probability represents the likelihood that an employee with the attributes shown in the first three columns will change jobs or locations during the upcoming period. These data will be used to simulate advancement of employees through the organization.

Once the model has been populated with these data, different scenarios can be tested to predict the outcome of various talent management decisions. These decisions relate to:

1. Changes in talent practices. Assuming a limited budget, the organization must prioritize the talent practices it will implement, maintain, change, or discontinue and the level of funding for each. A key application of the model is to determine the budget allocation that results in the highest possible level of readiness while meeting defined representation goals.

2. Allocation of recruitment budget. The model considers how budget dollars are allocated across sourcing channels in simulating movement into the organization. A key application of the model is to determine the budget allocations that will most likely enable the organization to achieve readiness, cost, and representation goals.

TABLE 10. PROMOTION/ADVANCEMENT RATES.

3. Economic/business outlook and other environmental parameters. Economic forecasts, projected unemployment rates, the financial strength of the organization, and demand and supply gaps for certain skills affect employee retention decisions. How these factors are defined is unique to each organization, depending on the relevance to their workforce.

OPTIMIZING AGENT-BASED SIMULATIONS

Modeling all possible combinations of practices, recruitment budget allocations, and environmental parameters is virtually impossible, even for a small number of options.⁸ Therefore, a procedure is needed that allows the user to focus on the set of scenarios that produce the best possible results. That is the essence of the optimization component of the technology. The optimization engine uses the most advanced global search algorithms to efficiently find the best solutions to simulation problems. This enables the user to focus on evaluating a limited number of potential solutions that the optimization has selected to yield the best results.

Figure 5 shows the results of an optimization of an SWP case for an engineering services firm. The performance curve represents the readiness level, and each dot on the performance curve represents an improving solution in terms of readiness. The goals for this optimization were expressed as follows.

The company wants to maximize readiness on a three-year planning horizon, while making sure that nonwhite and female employees would represent at least 30 percent of the resulting workforce. In addition, the company imposes a $4 million annual recruitment budget, a $10 million annual program management budget, a $100 million annual compensation budget, with the total annual HR budget not to exceed $105 million.

FIGURE 5. OPTIMIZATION RUN SHOWING 100 ITERATIONS.

In this example, readiness represents the extent to which workforce demand can be filled by the available workforce at the end of Year 3. In order for demand to be filled, the model matches employees to jobs based on job requirements and employee qualifications. Readiness, r, is measured on a 0 to 1 scale, and is calculated as follows:

where p_i represents job i in the workforce demand scenario, and E represents the set of jobs whose requirements can be matched with a current employee; thus, the numerator denotes the sum of all jobs in the demand scenario that can be filled, and the denominator represents all available jobs.

The best solution found is shown in Tables 11 and 12. From these, we conclude that if the company implements the programs marked “YES” in Table 12, and allocates the $4 million annual recruitment budget as depicted in Table 13, then it should expect to achieve a readiness level of 96.3 percent at the end of three years. The total investment in personnel costs and expenses is $94.01 million, of which $3.27 million is spent in sourcing new hires and $90.74 million in compensation, benefits, and other retention programs.

TABLE 11. SELECTED HR PROGRAMS IN THE BEST SOLUTION.

TABLE 12. RECRUITMENT BUDGET ALLOCATION ACCORDING TO THE BEST SOLUTION.

Recruitment Channel	Budget Allocation
General Universities	10%
Social eNetworks	5%
Ethnic-Serving Colleges	75%
University Job Fairs	0
Online Job Sites	0
Company Website	0
Recruitment Agencies	0
Ethnic-Serving Agencies	0
Network Contacts/Referrals	5%
Publication Ads	5%

Further inspection of this solution would show that women are expected to grow from 24.7 percent of the workforce to 39.8 percent; minorities from 25.5 percent to 43.5 percent; the age composition of the workforce varies from 35.6 percent to 40.2 percent in Generation Y, 23.8 percent to 42.6 percent in Generation X, and baby boomers from 40.6 percent to 25.2 percent. Average annual turnover is 6.7 percent, and new hires represent 19.4 percent of the workforce. It is also possible to drill down within each job level to analyze similar trends.

Workforce planners can define an optimization objective—i.e., what goal the model will optimize (typically related to readiness, cost, diversity representation, etc.)—and other key measures of success. They can also define parameters that govern the simulation, including length of the planning horizon, programs to optimize, changes in business strategy, and environmental factors (e.g., economic outlook, talent availability), as well as constraints (e.g., budget limitations).

In Figures 6 through 8 we compare solutions obtained from optimization to those obtained from traditional “what-if” analysis. Scenario 1, denoted as Base, corresponds to conducting “business as usual”; in other words, no new talent practices are added or modified, and investment in current recruitment channels remains the same. In the second scenario, denoted What-if, the user has made decisions to add or modify an HR practice or to reallocate recruitment investments. The third scenario, denoted Optimized, refers to the best solution found by OptiQuest.

FIGURE 6. READINESS RESULTS UNDER THREE DIFFERENT WORKFORCE PLANNING SCENARIOS.

As you can see from Figure 6, although the starting readiness level is about 85 percent, both the Base and the What-if scenarios perform poorly in terms of readiness (reaching levels of 60 percent and 83 percent at the end of Year 3, respectively), while the Optimized results in a readiness level of 97 percent at the end of Year 3.

FIGURE 7. TREND CHART OF NEW HIRES UNDER THREE DIFFERENT WORKFORCE PLANNING SCENARIOS.

FIGURE 8. TREND IN FEMALE EMPLOYEES UNDER THREE DIFFERENT WORKFORCE PLANNING SCENARIOS.

In Figure 7, we show the trend in external hires. In Optimized, we observe a small upward adjustment from 131 new hires in Year 1 to 137 new hires in Year 2, in order to account for initial turnover. The trend then becomes stable at 137 new hires in Year 3. Since turnover is much higher in the Base and What-if scenarios, the necessary adjustments are larger, and the number of new hires each year is highly volatile.

The analysis of new hires is not complete unless we also analyze the composition of turnover. By choosing the correct set of talent programs and practices, the Optimized scenario improves retention of the right kind of employees, described by a certain type of attribute. Say, for example, that the organization wants to encourage female employees to stay; then the organization would be interested in investing in programs designed to increase the retention of female employees—for instance, a comprehensive healthcare program. Such a program would also likely increase the retention of other types of employees, but its expected impact on female employees would be higher.

In Figure 8, we chart the trend in female employees for three years. In the Base scenario, we see the number of female employees decreases steadily if the organization continues with its current talent programs. In the What-if scenario, we have specifically chosen certain programs designed to reduce turnover of female employees; however, it takes two years for the downward trend to be reversed, because the hurdle that has to be overcome through hiring is large. Given budget restrictions, the programs chosen under the What-if scenario do not produce the biggest impact per dollar invested. On the other hand, the Optimized scenario shows an immediate upward trend in the number of female employees. Here, the investment in talent programs is chosen to produce the greatest impact in terms of the female retention goal. This is analogous to the financial arena, where an investor seeks a portfolio of securities that results in the highest return for a given cost.

We now illustrate the benefits of this type of analysis with a real-world application at CH2M HILL, a global engineering services firm with a total workforce of about 30,000 employees, which is using this technology to align its talent strategy with its business goals.

CONCLUSIONS AND FINAL REMARKS

Agent-based simulation optimization software for Strategic Workforce Planning is based on a bottom-up approach to predicting talent behavior patterns. This technology focuses on accurately predicting individual employee decisions based on demographic and career characteristics, as well as work-related and external stimuli. Individual behaviors are then aggregated into cohorts of interest so that large-scale effects can be analyzed. As the case from CH2M HILL shows, this approach is fundamentally different from traditional top-down or trend-based approaches used by most workforce planning tools in the market, and it produces more statistically reliable results.

In addition to improved accuracy, the agent-based approach enables the evaluation of the impact of alternative investments in talent programs, benefits, and sourcing practices based on the unique set of attributes of your workforce, the work environment, and the state of the economy. The optimization capability takes this process one step further by enabling the planner to identify the best set of retention and sourcing programs that are most likely to achieve the organization’s goals within budget.

We believe no other approach to Strategic Workforce Planning can achieve the same level of accuracy at such granularity and with more confidence in the predicted outcomes.

References

1. The Conference Board 2011 Survey of 704 CEOs worldwide.

2. See, for example, Wayne F. Cascio and John Boudreau, Investing in People: Financial Impact of Human Resource Initiatives (Upper Saddle River, N.J.: Pearson Education, 2008); Jac Fitz-enz, The New HR Analytics: Predicting the Economic Value of Your Company’s Human Capital Investments (New York: AMACOM, 2010).

3. See www.OptTek.com for more information and documentation about the OptForce SWP system.

4. See www.OptTek.com for information and documentation about using OptQuest.

5. Compiled from Watson Wyatt webcast “Advanced Workforce Planning: Securing the Future,” Human Capital Institute, November 20, 2008; “Customizing the Employment Offer,” CLC Solutions (Washington, D.C.: Corporate Leadership Council, December 2002).

6. See http://nces.ed.gov/.

7. See, for example, EMSI or Workforce Locator.

8. If there were only twenty PPIs and two alternatives for each PPI, there would be about 1 million different combinations to choose from (ignoring any budget constraints).

9. Arlene Dohm, “Gauging the Labor Force Effects of Retiring Baby-Boomers,” BLS Research, 123 (2000). Retrieved from http://www.bls.gov/
opub/mlr/2000/07/art2full.pdf.

10. The Corporate Executive Board, “U.S. Trends in Entry-Level Candidate Availability by Function,” (2006).

11. See, for example, http://enr.construction.com/toplists/ProgramManagers/
001-050.asp; www.Verdantix.com.

12. See http://www.lloyds.com/News-and-Insight/Risk-Insight/Lloyds-Risk-Index.

Dr. Marco Better is the Director of Custom Solutions of OptTek Systems, Inc. He obtained his Ph.D. in Operations Research from the Leeds School of Business of the University of Colorado at Boulder. He also holds a B.S. in industrial engineering and an M.B.A. Dr. Better has over fifteen years of professional work experience in the automobile, banking, and telecommunications industries, both in the US and in Latin America. His current interests lie in the application of optimization and data mining technology to solve complex problems in industry.

Dr. Fred Glover is a Chief Technology Officer in charge of algorithmic design and strategic planning initiatives. Dr. Glover is a leading figure in the field of meta-heuristics, a name he coined in the 1980s – an area that is now the subject of numerous books and international conferences, focusing on the development of models and methods enabling the solution of complex nonlinear and combinatorial problems that lie beyond the ability of classical optimization procedures. He also serves as the MediaOne Chaired Professor in Systems Science at the University of Colorado, Boulder, where he holds the title of Distinguished Professor of the University of Colorado system. He has authored or coauthored more than 350 published articles and 8 books in the fields of mathematical optimization, computer science, and artificial intelligence, with particular emphasis on practical applications in industry and government. Dr. Glover is the recipient of the distinguished von Neumann Theory Prize, an elected member of the National Academy of Engineering, and has received numerous other awards and honorary fellowships, including from the American Association for the Advancement of Science (AAAS), the NATO Division of Scientific Affairs, the Institute of Operations Research and Management Science (INFORMS), the Decision Sciences Institute (DSI), the U.S. Defense Communications Agency (DCA), the Energy Research Institute (ERI), the American Assembly of Collegiate Schools of Business (AACSB), Alpha Iota Delta, and the Miller Institute for Basic Research in Science.

David Sutherland joined CH2M HILL in July, 2008, and is now the Director of Workforce Insight at CH2M HILL and has accountability for global workforce reporting, metrics and analytics, and strategic workforce planning. Headquartered in Englewood, Colorado, CH2M HILL is a global leader in consulting, design, design-build, operations, and program management for government, civil, industrial, and energy clients. The firm’s work is concentrated in the areas of water, transportation, environmental, energy, facilities, and resources.

Prior to joining CH2M HILL Dave led Avaya’s Workforce Planning function where he had responsibility for delivering metrics on all aspects of the global workforce, creating companywide headcount forecasts and driving the assessment of skill levels for critical positions to ensure hiring, development and retention strategies addressed identified gaps. Prior to Avaya, Dave was a special agent with the FBI and also spent five years with Prudential working in financial analysis, business measurement/process reengineering, and human resources measurement consulting roles.

Dave has conducted numerous speaking engagements over the past ten+ years at a variety of workforce planning and measurement forums including APQC, PwC/Saratoga, The Learning Forum, Human Capital Institute, Inform, SHRM, and The Conference Board. Dave also spoke on a workforce analytics subject-matter expert panel at the 2011 HR Technology conference. He currently serves on the executive committee for The Learning Forum’s Workforce Planning Council and was appointed to HCI’s Executive Advisory Board to serve on the Workforce Planning Board Committee. He graduated from Boston College with a degree in finance and is a Six Sigma green belt.

Dr. Manuel Laguna is a Senior Research Associate of OptTek Systems, Inc. He is Professor of Operations Management in the Leeds School of Business of the University of Colorado at Boulder. He received masters and doctoral degrees in Operations Research and Industrial Engineering from the University of Texas at Austin. He has done extensive research in the interface between computer science, artificial intelligence, and operations research to develop solution methods for problems in areas such as logistics and supply chain, routing and network design in telecommunications, combinatorial optimization on graphs, and optimization of simulations. His research has appeared in numerous academic-journal articles and books. He is the editor-in-chief of the Journal of Heuristics and is on the international advisory board of the Journal of the Operational Research Society.