The promise of big data for HR.


The move toward "big data" has enabled companies to make

large advances in understanding their customers and markets. Now we hear

many people advocate for "big data" for HR. How much is hype

versus reality? Where are the insights going to be? In this article, the

author reviews recent big data developments and maps out the likely

paths for HR.

What's Going on Outside of HR

Outside of HR there is a revolution taking place in understanding

consumer behavior and business processes. Huge databases are aggregating

information in ways never before possible, enabling an incredibly

detailed description of how people live their lives and purchase goods

and services. In short order, these pictures will become even more

detailed, as the appliances and systems in our homes and our cars become

even more linked to the internet and report our minute-by-minute


For example, I recently took a trip to Chicago from Los Angeles. On

the plane on the way to Chicago, I purchased wifi service with about

four hours to go in the flight. I was presented an offer of three hours

for the price of two for $15, two hours for $10, or one hour for $7. On

my return trip home, I waited until there were about two and a half

hours left in the flight, and was presented with a different offer: two

hours for $15. This type of dynamic, location-and-time specific pricing

occurred without me ever giving any personal information. Innovations in

product offering and marketing like this are truly breakthrough.

On the business process side the evolution is also impressive,

though more incremental. Technology today enables companies to remove a

lot of guesswork in managing the business. They can accurately monitor

their supply chains, sales, and equipment functioning. Real-time

congestion maps and route optimization software cut down on driving time

and idling in traffic. Corporate headquarters can monitor the uptime of

manufacturing lines 24/7 across the globe. The end result is more

efficient allocation of resources across the enterprise. In contrast to

the breakthroughs in product development and marketing, these

innovations in business process management do not amount to a

revolution, yet they are still very important for business success.

While these advances are impressive, still there are gaps in

understanding consumer behavior and business processes. Understanding

what drives a lot of purchasing decisions is still part art and part

science. There are a million new product and service ideas innovators

dream up every year, yet only a small fraction are ever successful

commercially. Trial and error are still the best ways of measuring

what's going to be a success in the marketplace. That's the

art side; predictions are abysmally inaccurate.

On the science side, our existing models of consumer behavior do an

adequate job of predicting how people will respond in many situations.

The growing mountains of data have not fundamentally changed our

understanding of what drives many purchase decisions. We know that

someone who buys peanut butter is highly likely to want jelly as well;

using sexually explicit imagery in advertising will attract

people's attention; and warmer clothes sell better during the fall

and winter. We already have solid models of consumer behavior that make

those predictions highly accurate.

On the business process side, there are many barriers to improving

organizational effectiveness. Most come down to people: how do we get

them to do the right things in the right places at the right times? For

example, we know customer service drives a lot of revenue, and

innovation can increase market share. Yet we can't just tell

employees to provide top notch customer service or be innovative: they

have individual motivation that may or may not be aligned with the work

systems, job design, and rewards. Improving employee engagement and

motivation would go a long way toward improving both sales and margins,

and turn the big data promise into a reality.

The Promise of Big Data for HR: New Data vs. Access and Analysis

To improve what is happening on the employee side, we need to

expand and improve what we know. Three ways we can do it in this new big

data world are through collecting new data, using existing data more

effectively, and better strategic analysis.

Collecting new data: Some of the biggest gains on the consumer and

business process side have been made by gathering new data on what

people do and how things operate. When many people hear "big

data" this is the probably the first thing that comes to mind: the

massive collection and digitalization of information that previously was

not available for easy analysis. On the employee side, there are some

examples of new data collection that provide business process insights.

Monitoring call What is enterprise search? center workers' time per call and customer ratings

can increase both productivity and quality. Converting medical records

to electronic format can help improve health care delivery. These both

are examples of how technology can enable the direct observation of

business processes in ways that reduce errors and increase efficiency.

The information collected primarily is about when, where, and how work

gets done.

A different type of new data collection will come from social media

platforms at work. Companies are experimenting with adapting in-house

versions, such as Yammer, that can be used for communication and

peer-to-peer networking. Like the early days of cell phones and the

internet, we are at an inflection point where it is clear that

businesses are almost certain to redefine key processes around the usage

of social media platforms at work. The challenge at this point is we

don't fully know what those innovations will be, though we can make

educated guesses.

Knowledge management is one potential contribution. Social media

platforms can provide a user-driven exchange of information that spans

functional, geographic, and cultural boundaries. Personalized

professional profiles, like those found on LinkedIn, may aid project

assignment, promotion, and career planning. If companies build

successful in-house platforms where employees share their experiences

and goals, there will be improved matching of employees, teams and


Better access to existing data: The new data described above

provide insights into when, where, and how employees do their jobs, yet

do not explain why employees do what they do- their motivation and

engagement. On this front, there is not much new data to be collected

that hasn't been collected before: we know the questions to ask and

have collected these types of data for two generations using interviews

and surveys. What is different today is the ease of getting the data:

data collection methods have improved greatly, providing potentially

deeper insights on larger groups of employees at more frequent


The tools for conducting electronic employee surveys and providing

real-time feedback continue to improve. Today anyone can design a low-

or no-cost survey. Of course, designing an effective survey requires

more than just throwing a bunch of questions up on Survey Monkey. Yet

the technological barriers to conducting extensive surveys have been

removed for good. Today the only real barriers are created by survey

fatigue and the challenges of rallying the organization to take action

on the data. The costs of fielding a survey from a technological

perspective are too small to be a barrier.

A more recent innovation is shorter "pulse" surveys that

are used for more frequent monitoring of employee attitudes. One use of

these surveys is more real-time measurement of key questions from an

annual employee survey on a quarterly or more frequent basis. A second

use is measuring progress and improving organizational change

initiatives: this can include open-ended questions that enable employees

to raise issues and contribute ideas. The pulse surveys can complement

actions taken in response to the annual employee survey, including

monitoring the progress of change.

A final innovation in reporting relates to manager feedback of

survey data. Today automated summaries for each manager of how their

direct reports responded are possible. The individualized feedback

reports can be used for benchmarking against other managers'

performance in the organization, and for comparison with previous

year's performance and trend analysis.

Analysis: There is very little new in data analysis, with two

possible exceptions. The first exception is social network analysis,

which provides a mapping of how information flows in organizations, and

the relationships that people rely on to do their work. While such

mappings can provide interesting ways of looking at the organization,

the time needed to do a social network analysis is substantial; the

benefit needs to justify the effort. One potential benefit is

understanding who influences organization attitudes, behaviors an

change, which could be used during a change process. Knowing key

influencers allows the organization to bring them into the loop on a

change process.

A second relatively new practice links employee survey responses to

other data sources such as employee turnover, sales, customer

satisfaction, and analyzing the relationships to identify drivers of

talent retention and business performance. Proponents of this approach

include those arguing that engaged employees are critical for business

success, particularly in customer-facing roles, and include the authors

of the The Service Profit Chain (Heskett, Sasser and Schlesinger) and

many consulting companies that provide survey hosting services.

Other gains on the analysis side may come from increased ease of

using statistical software routines. Examples of routines include

regression analysis, anova, factor analysis, and even calculating simple

correlations and means. HRIS and enterprise software providers are

building more and more functions into their offerings that can be

operated in turnkey fashion. This can reduce the need to rely on

statistical experts to do highly technical analysis. Yet knowing how to

specify hypotheses, clean data, set up and test models, and examine data

for anomalies are all critical competencies that anyone should possess

before attempting statistical analysis, even if turnkey software

automates some of the steps, such as cleaning the data of outliers.

Anyone can walk into a surgical room, put on a gown and pick up a

scalpel, but I would only trust a surgeon to operate on me. Setting up

and running a complicated statistical analysis is not brain surgery. Yet

doing it so that real, actionable insights are found requires a minimum

level of statistical competence and knowledge about modeling individual

and group behavior.

It's Not Just the Data, It's What You Do with It

Which brings us to the real "bottom line" when we're

talking about the pros and cons of the current big data era" the

critical need to ask the right questions. Fancy analysis does not

necessarily mean deeper insights, unless the analysis is directly tied

to insightful questions that require advanced statistics. Asking the

right questions means being clear about what causes what so you

don't confuse correlation and causation. And that in turn means

setting up and testing a causal model.

Causal models address "why" things happen. Figure 1

provides an example of a causal model. Generally speaking, there are

three main contributors to performance in a job or team, and they are

represented in the figure on the following page:

* Job design: the design of the roles and responsibilities of the

jobs or team, and how they are supposed to work together with other

roles, teams, departments/functions and business units.

* Capability: the individuals' or team's capability to

perform the tasks efficiently and effectively.

* Attitudes: how motivated and engaged the individuals or team are

to perform the tasks.

In order to do a complete diagnosis of what drives employee

motivation and productivity, all three types of contributors need to be

considered, even if not all three are ultimately included in the

measures used to conduct the analysis.

The challenges of setting up a "true" causal model can be

demonstrated using two examples that have received a lot of attention

and are closely related: employee engagement and linkage analysis. These

both are good examples of the promise and pitfalls of big data for HR.

Employee engagement. Employee engagement is a concept that is both

old and new at the same time. There is a growing volume of research

literature on the topic, which I will not describe in detail here.

"Engagement" has been used to refer to everything from

traditional measures like job satisfaction and commitment, to new

measurements that include thriving, energy, and discretionary effort.

The implied causal model says that higher employee engagement leads

to better business performance. On one hand, there is evidence to

suggest that having more engaged employees in certain customer-facing

roles can directly contribute to improved business performance; this is

the argument made by the authors of The Service Profit Chain--customers

will buy more if they are treated well. All else being equal, having

more engaged employees should be better for organizational performance.

Yet that is NOT the same thing as saying that if organizations improve

measured employee engagement, that business results will improve; that

is the causal conclusion that cannot be universally supported by either

data or logic. The answer can be found by paying proper attention to

what is being measured and the true causes of business performance.


As George Rose, EVP People and Organization, Sony Pictures notes,

the first problem is that: "Focusing on employee engagement alone

is not enough. I have known some groups that register as highly engaged

on annual employee surveys yet are clearly not high performing. You have

to get to the root of the issues that are the true causes of

performance." A group of employees can seem very

"engaged" in survey measures because they are happy with the

work they are given and the direction provided by their supervisors. But

if their supervisors do a poor job focusing the team on tasks that

support the business and promote accountability, there can be a

disconnect between measured employee engagement and actual business

results. What's missing in this case is the information on goal

setting and accountability

A second problem comes from comparing engagement across employees.

Some jobs are more interesting to do, provide greater autonomy, and are

involved in making key decisions (for example, many technical and

professional roles); people who work in these jobs are much more likely

to register high levels of engagement in a survey because the work is

intrinsically motivating. Other jobs are more routine, less interesting,

and even unpleasant to do (for example, past due bill or garbage

collection); registered engagement here should be lower. To compare

engagement across roles like these requires baseline information on how

intrinsically motivating each job is. We need a way of knowing when

attention should be focused elsewhere on other areas that could be

improved, and which might contribute even more to improved business

performance than more employee engagement.

The challenge of drawing a direct link between measured employee

engagement and business performance can be seen in Figure 1. Measures of

engagement at the individual level fall into the "attitudes"

bucket, and thus have the potential to impact business performance. Yet

the roles of capability and job design have to be considered as well:

* Capability: A group of lower-capability employees could have high

measured levels of engagement if they are not pushed hard to perform

beyond their comfort zones; yet once they are held accountable for

performing at a high level of productivity, measured engagement could

plummet as they resent having to work hard at tasks that are difficult

for them to accomplish. In this case, high measured engagement predicts

lower business performance, and vice versa; thus engagement in this case

does not "cause" performance.

* Job design: Even a highly competent group of employees could be

held back from contributing to business success because of lack of

alignment or resource support. In such cases, there may be no link

between employee engagement and business performance.

Linkage analysis. The statement that employee engagement improves

business performance comes from a type of linkage analysis. In its

simplest form, linkage analysis links people measures (employee

attitudes and capabilities) with business performance metrics. The

implicit assumption is that making improvements on the people side will

improve business performance: people metrics "cause" business


Linkage analysis is appealing is because it is easy to construct an

argument for why the people measures should matter for business

performance. Yet, just because something should contribute to improved

organizational effectiveness and performance, there typically is no

guarantee that it will--and often many scenarios under which it

won't. While many people measures are important contributors to

organizational effectiveness, they are part of an interdependent system

that relies on multiple HR contributions to be successful. For example:

* Training plays an important role in closing competency gaps, so

training incidence should be correlated with improved performance; yet

training is often not the most important contribution.

* Competency demonstration should also be correlated with improved

performance; yet some of the most important competencies that drive

performance cannot be measured well and thus are missing from the set of

measurements (for example: decision making ability).

* Coaching is an important part of effective feedback and

performance management, yet it is rarely the most important or critical

barrier to performance; instead, it is a contributor while rarely, if

ever, being "the" cause of improved performance.


The problem with linkage analysis that focuses on only one of the

measures above is that the analysis is highly likely to show a

statistical correlation between the people measure and organizational

effectiveness and performance. Yet as the discussion above demonstrates,

such correlation cannot be attributed to true causation.

Some organizations have recognized that engagement alone is not

enough to help explain or understand the observed relationships between

people data and organization performance. Jack in the Box has adopted a

model similar to the one in Figure 1. As Mark Blankenship, SVP and Chief

Administrative Officer, explains, "Employees can be engaged, but if

they aren't aligned with the organization's goals (through

communication, compensation and accountability) and they don't have

the right team or tools in place to perform (through equipment,

training, processes, hiring of the right talent), it's difficult to

determine what organization decisions to make to further improve


These two examples--employee engagement and linkage

analysis--highlight the general challenges of big data facing HR and

organizations today: a lot of analysis is driven by ideas that

"make sense" yet which do not provide actionable insights.

Having new, easier-to-access data opens the door to interesting

analyses. Yet causal models have to be tested for the analyses to

provide actionable insights to improve performance. When people measures

and business outcomes are statistically correlated, alternative causal

models need to be considered and ruled out before taking action. Only

then will we have confidence that big data is moving HR forward and not

just providing an alluring distraction.

Dr. Alec Levenson is a senior research scientist at the Center for

Effective Organizations, Marshall School of Business, University of

Southern California,

Write a comment

Comments: 0