Chapter 3. Supporting Human Capital Decision Making
If you are reading this book, then you are likely aware that the use of metrics in human capital management has been receiving a good deal of attention recently in a variety of forums. Many in Human Resources are ready to jump on the metrics bandwagon and are looking for specific directions on how to begin. Getting started with any new initiative can be a daunting task. The development of a metrics approach is seen as finding the best human capital related metrics (the what to measure) and deciding which measurement and analysis techniques to use (the how), as well as the methodology for interpretation (what does it mean), followed by making recommendations for appropriate action (what to do about it). With such a broad cut, the goal of developing an all encompassing metrics plan is the kind of project that can easily be measured in person years.
We have left out the most important dimension in terms of defining a metrics strategy: the why. Metrics should not be gathered, analyzed, and reported on for the sake of doing so. There needs to be an underlying motivation. Looking across the various disciplines that define most of the functional areas in modern business organizations, the underlying motivation for gathering data, analyzing, reporting on trends, and making recommendations, relates to decision support. Frequently the individuals or department responsible for providing decision support are not the same individuals making the actual decisions.
Boudreau and Ramstad have drawn a powerful analogy between where Human Resources is today in terms of human capital analytics and where Finance was 50 years ago in terms of analyzing accounting related data. In an oversimplification, accounting can be described as the function charged with the recording and reporting of all the financially related transactions that occur between an organization, its partners, customers, employees, and so on. These numbers can then be aggregated and reported on in literally millions of different ways. As expressed in the Generally Accepted Accounting Principles (GAAP) set by the Financial Accounting Standards Board (FASB), these accounting principles can be thought of as the rules governing how this is done. FASB (or any other agency) dictates exactly how the accounting numbers should be interpreted in any given situation and how that interpretation should drive financially related business decisions. This is a critical point of comparison. Accounting can produce a myriad of metrics, but without some guiding framework, or as Boudreau and his colleagues have succinctly put it, “a decision science,” drawing any reasonable conclusions from this sea of numbers is nearly impossible. In the maturing discipline of Accounting and Scientific Financial Management, it does not make sense to ask, “What makes a good accounting number?” If it was asked, the answer would be in the form of another question, “What decision are you trying to support?” Modern finance as a decision science has evolved considerably in just the last five or so decades and grew out of the need to make sense of the numbers that were being generated from the accounting process.
It is important to understand the distinction between the roles of making decisions and supporting those who make the decisions. In the end, the quality of any decision support system, and the underlying metrics and analytics that it encompasses, is best measured by the quality of the supported decisions. It is worthwhile to step back and examine this concept of decision making before we plunge in to build a framework to support it.
Meta-Level Decision Science, as it relates to the way humans make decisions in general, is far too high a level at which to start, instead we will take a cue from the world of finance and use that as a context within which to begin. In managerial finance, the core of what is considered to be the decision making process is referred to as micro-normative, which is defined as a process where a decision maker is selecting between alternative courses of action under uncertainty (probability) while seeking to maximize some objective function (utility theory). This process is carried out in the context of a causal model, which explicitly or implicitly infers causal relationships or links between observable and unobservable aspects of the real world system against which the model is built. We will refer to the combination of models, measures, metrics, and methodologies used to help support the decision maker as a framework. Let’s examine each of these in turn.
Decision maker. While this may seem intuitive, it is important to understand that identification of the decision maker(s) is critical to building effective decision support tools and that different decision makers may, and usually will, have different perspectives and requirements. There are literally thousands of decisions made daily in most modern organizations where the process has become much more diffused and decentralized over the past few decades, as the idea of having decisions made by those with a vested interest in the outcome has gained momentum. This decentralization has profound implications for decision support.
Framework. What is a framework? For our purposes here, it is defined as a collection of measures, methodologies, tools, models, and guidelines organized to facilitate meaningful and informed decision making. Typical discussions of framework development in Human Capital Management (HCM) analytics generally start by focusing on the measures (metrics) immediately. “What makes a good metric” is often the place where practitioners attempt to start. This bottom-up approach is starting from the wrong end. The value of a particular metric cannot be determined without knowing the decisions that need to be made. A single metric will often be applicable across a variety of decision processes. The place to start is with the decision itself.
Model. Central to the journey of advancing from simple anecdotal evidence, based on past observations, to a useful framework that allows for prediction of potential future outcomes, is the use of a model. In technical terms, a model is a selective abstraction of a real world system, optimally with enough detail to make it possible to gain an understanding of the real world and test alternatives working with the model. Building a model involves the process of capturing selected characteristics of the real world system and the processes it encompasses, and then combining these into an abstract representation of the original. As long as the model is a reasonably accurate representation, conclusions drawn from such an analysis may be validly extrapolated back to the original system. While this description may make modeling sound formidable, in reality it is the very process that humans go through whenever they are trying to understand any complex system.
It is the complexity of the real world system in its entirety that makes it necessary to develop these abstractions. The model is dependent on the perspective from which it is developed and viewed. My model of an automobile is abstracted from the point of view of the driver interested in getting from point A to point B and is very different from the model that my mechanic utilizes in his daily job of maintaining vehicles. We will be describing techniques where modeling involves defining inputs, entities, actions (including interactions), links, effects, and outcomes. It is these last three items that become the most critical in our development of models. How does a particular action by a particular entity within the system relate to effects upon another entity, cascading through other effects and ultimately manifesting as an outcome?
Take as a simple example a model of the cause and effect relationships related to employee tenure. We’ll start with a basic question, “Does salary have an effect on employee tenure?” We may already have a pretty good idea that it does and can move to a hypothesis: “Higher pay in a given job category will lead to longer tenure.” Inherent in that hypothesis are a number of potential courses of action, themselves related to making a decision about increasing pay levels in a particular job category. One of the critical relationships in this model is between pay and tenure, which we believe is a causal link.
Causality. The concept of cause and effect is inherent in the way humans understand the workings of everything around them. The ability to relate input factors related to outcome effects is central to building a framework, and being able to predict ahead of time what may happen. Haig Nalbantian and his colleagues at Mercer Human Resources Consulting provide a useful description of the continuum from observed “fact” to “predictable outcome” as it relates to human capital metrics and analytics. Without well understood causal relationships, the task of measuring and interpreting metrics would be essentially pointless as only those measures that were direct would be of any value. The vast majority of the time we are looking to equate the change in one measure with that of another in a cause and effect precedence relationship.
Courses of action. In general, the result of a decision is translated into taking a specific action. In this definition there is no such thing as not deciding or not acting. After all, in any given situation one of the alternatives is to continue the status quo. While this may seem like not deciding, it is choosing a particular course of action. Courses of action generally have definable subcomponents, which may be distinguished as mutually exclusive between different courses. Simply stated, this is when we say, “we can do A or B, but not both.” Other situations involve overlapping subcomponents as in “We can do A and B, or A and C.” The process of decision making generally involves weighing the alternative courses of action and choosing among them based on the optimization of some objective function.
Utility theory/objective function. Central to the concept of utility theory is the objective function. The objective function embodies the quantification of the goal we are attempting to achieve. Examples include employee retention, return on investment, profit, shareholder value, and so on. Through our causal modeling process we can determine the relationships between what we can control (such as salary, working conditions) and the components of the objective function that we are trying to maximize. If it were as simple as discerning and quantifying the relationships and building the equations, the process of making the decision would be mathematical and could be carried out easily. Unfortunately, it is far from that simple. Identifying, much less quantifying, the linkages in our model is difficult to do within reasonable levels of certainty. One seldom knows exactly what is going on, and it is that uncertainty that is usually the crux of the decision making process.
Uncertainty. A typical dictionary definition of uncertainty is “The condition of being uncertain; doubt.” Meaningful decisions are almost always in situations where one is not certain of the outcome; we never have access to the whole truth about our environment. An effective decision making process, and therefore any supporting tools or frameworks, must explicitly deal with this uncertainty.
In the end, decision making comes down to making choices between uncertain outcomes based on partial data from an abstraction of the system that does not provide complete information.