Author: Carlos Grajales
More than half a dozen people, including statisticians, mathematicians and engineers are sitting together in a table, discussing strategies to tackle their new analytics challenge . Like many data analysis teams, they use a multi-disciplinary approach to develop an analysis guideline that relies on the latest statistical models, grounded on sound mathematical theory and implemented in the latest software. To be fair, their work and process is not truly remarkable, as you couldn’t find any difference from the level of sophistication their methods employ, at least compared with the common analytics department in most industries. The reason why this particular team in San Francisco stands out is simply because they are not the common analysts from the Marketing or Sales department, they are the full-time analysts devoted to the Human Resources Department. They don’t model sales, they model personnel.
The idea on itself is not quite revolutionary. Way back to the previous century, some companies started gathering HR metrics, specialized quantities to measure important aspects that matter to Human Resources Specialists, such as “Retention Rates”, “Total Headcount”, “Employee Satisfaction” or “Time to hire”. These companies relied on descriptive statistics to assess the performance of the different roles the personnel management team had within the company . During those times both metrics and analyses stayed relatively simple, relying mostly on basic charts and using more advanced inference only for very well defined projects (like the very common example you had in school, where you compared the “impact” of a new training procedure by contrasting means). Then came the 21st century, Google, Big Data and it wouldn’t be too long before the Human Resource department started asking for statisticians. The idea has reached quite a following, to the point that it now has its own trendy data science name: “People Analytics” or “Talent Analytics” .
Technology firms are usually among the first in trying anything data-driven and this wasn’t the exception. One of the most notorious and reported projects of Personnel Analytics took place in 2009 in Mountain View, where a group of statisticians within the Google complex started working on a Project code-named “Oxygen” . This new project was very similar to other the analysis team had worked on; basically identifying patterns that could help them identify the best way to do something. The key difference this time was the goal itself. Project Oxygen wasn’t about finding the best ways to suggest search results of the best way to present ads: it was about finding the best way to create Managers . By statistically analyzing performance reviews, feedback surveys and nominations for top-manager awards, the team tried to identify the most common habits of effective managers. By the end of the year, the team had identified 8 key aspects a successful manager should have, among them “Being productive and results-oriented” or ”Empower your team” and “Have a clear vision and strategy for the team”. I know, nothing quite revolutionary, every coach on earth could actually tell you something like that, unless you consider the least important managing aspect Google identified: “Technical Expertise”. Apparently, Google found that employees valued way more bosses who made time for one-on-one meetings, bosses who took an interest in their lives and who navigated through problems by asking them questions instead of shoving solutions to the team. To be precise, the Google analytics team realized that it is indeed important for your manager to understand your work and the labour of the area, but the difference between a regular and a great manager is creating a deep connection with the staff.
As part of its “stating the obvious” section, Project Oxygen also proved some other common knowledge theories: better managers tend to have better teams, which perform better and are happier. They also have employees who are retained for more time. Even if you believe all these results are common sense, being founded on a data-driven approach allowed the company to put all these insights into work. Based on the analysis, Google updated their management training programs, implemented personalized coaching for managers with poor performance and started working on creating better management teams .
And since we are talking about Google, it is worth discussing their popular and unique employee benefits, many of which are also based on statistical analyses and experimental designs. Google’s PiLab, which stands for Google’s People & Innovation Lab, has performed many interesting analytical projects, such as the project for improved management that we just recently discussed . PiLab is also responsible for many other HR related projects, such as experiments designed to find ways to improve the amount of money employees save for retirement, a Retention Algorithm that identifies the employees in risk of leaving the company  and a study aimed to increase consumption of healthy food choices among their personnel . Not only have these projects produced tangible results that improved the life and work of Google’s employees, they have also gathered media attention and produced good-will for the company.
In fact, other industries have also been following this news. Nowadays, using statistics for HR management isn’t something exclusive of Technology firms. Most big companies now rely on insights provided by data to manage their human resources, even quite older companies where the problem is mostly losing their capable workforce due to retirement. Black Hills Corporation is a North American energy conglomerate, with over 130 years of existence. As a big company, it has some challenging HR requirements: their workforce is rapidly aging and some of their positions require some very specialized skills and a long road of training to reach full competence. To overturn this, Black Hills used statistical forecasting models to estimate the number of employees that would retire per year, per type of worker . This in turn allowed the company to create a detailed workforce planning, which signaled the possible shortcomings in talent the company could face. Thanks to the model, plans were designed to prevent the potential talent shortage, by identifying the expected timeline of departure of key positions and selecting with time the places where those hires would come from.
Nowadays, using statistics for HR management isn’t something exclusive of Technology firms. Most big companies now rely on insights provided by data to manage their human resources, even quite older companies where the problem is mostly losing their capable workforce due to retirement.
So far, we’ve discussed some of the successes within the personnel analytics field. Yet this story can’t be complete without a good share of failures as well. Modeling people is far from being easy and even the almighty Google can tumble with it. Todd Carlisle was Google’s People Analytics Manager for a while. He devoted some time during his tenure to develop a test of “googliness”, some weirdly-named measure of cultural affinity for potential candidates . What Carlisle did was ask every employee with at least 5 months within the company to fill a survey of about 300 questions (phew). All those 300 variables were correlated with the standard performance measures of Google. As you’d expect, some of them turned out to show significant correlations, looking as promising predictors of future performance. After some refinements, a final version of the Google Candidate Survey was used starting from 2007. The survey was applied to each potential candidate: based on its results a more thorough decision about the hire was made. But worry not, if you happen to get a job interview at Google, you wouldn’t likely take this test: it has been almost completely abandoned. After using it for a while, and testing it with real hires, the company realized that almost nothing in it was a good predictor of success within Google. What remains of the project are some departments that still use some of the questions during their interviews, yet the overall idea has been somewhat abandoned, victim of spurious correlations.
Any reader with some HR experience should identify similarities between Google’s failed approach and the use of Psychometric tests in job interviews. Although not necessarily related with HR, (Psychometrics involve the objective measurement of psychological traits for many purposes, ranging from diagnosis to education ), the use of sophisticated psychological measurements to aid in recruitments tasks has been a constant in major companies for a while.
Yet, as Carlisle himself found out, it is not easy to predict performance based on such questionnaires. In fact, some research suggests personality traits tend to exhibit lower predictive power than general cognitive ability tests . Even though companies spend millions on psychometric testing, its results are far from guaranteed.
These incidents also shed light on one of the major drawbacks of People Analytics: bias. It has been argued that flawed tests and models commonly used for personnel selection or promotion may be culturally biased, rewarding knowledge and practices common in certain cultures, yet not in others . An example would be a test developed with data of Latin American employees only. Within this group, the personality traits that correlate with good performance might not be the same traits that describe successful Asian employees, simply because personalities tend to differ from a cultural perspective. Not taking such considerations into account could derail the purpose of a Psychometric test. Besides, the used of multiple testing during job interviews can also raise concerns of privacy and of inappropriate use of personal information.
But even after taking into account the perils and fails of it, the advantages of analyzing people data within a company are becoming more and more relevant, as organizations are now using data to actually solve performance issues and profit driven situations from a HR perspective. For instance, a health care provider analyzed associations between infection rates and people issues reported in different hospitals, as a means to identify training or hiring issues that could lead to unsecure protocols being utilized . A financial company used analytics to model the cases of theft reported within the company. They found out what people who commit fraud have in common with each other and what type of environmental or hiring factors might contribute to such behaviours. With this, the company expects to dramatically reduce the incidence of such violations in the future, or at least be more aware of possible risks .
All companies rely on people for their day-to-day operations and considering the fact that personnel is one of the major expenses of major corporations, you probably assume, correctly, that People Analytics is a major powerhouse for companies. And as it happens with powerhouses, companies are willing to pay millions to gather the deep insights a statistician can bring to the HR team. I know, you are probably thinking in disrupting the consulting model by offering the latest models for the HR department, at least I did the moment I knew about it. But I have bad news for us: it is too late. Many data companies are already investing big on these trends, providing personalized solutions for the personnel operations of companies. IBM has its very own HR Analytics solution, focused on statistical solutions to improve employee satisfaction, productivity, design better training initiatives and even to optimize the selection process of new candidates . Oracle has developed a specialized Dashboard for the human Resources department, a platform that not only displays the latest metrics but also performs some basic association analyses to identify interesting correlations between performance metrics and employee compensations or satisfaction .
The many success stories and the increasing amount of available tools has only increased the pace at which statistics has taken over HR departments across many industries. At this rate, it won’t take long before many of the more common operations of the personnel department become intrinsically bind with the use of statistical tools. Personally, the next time I get rejected in a job interview I’m thinking of asking to take a peek at their hiring mathematical model, just to be sure.
 Bersin, Josh. The Geeks Arrive In HR: People Analytics Is Here. Forbes Website (Feb, 2015)
 Bryant Adam, Google’s Quest to Build a Better Boss. The New York Times (Mar, 2011)
 Kurkoski, Jennifer. Hello Science – Meet HR. Google Research Blog (Jun, 2012)
 Sullivan, John. How Google Became The #3 Most Valuable Firm By Using People Analytics To Reinvent HR. ERE Recruiting Intelligence (Feb, 2013)
 Kuang, Cliff. In the Cafeteria, Google gets Healthy. Fast Company Magazine (Apr, 2012)
 Collins, Mick. Change Your Company with Better HR Analytics. Harvard Business Review Website (Dec, 2013)
 Poundstone, William. Are you smart enough to work at Google? Oneworld Publications (January, 2012)
 Psychometrics. Wikipedia, The Free Encyclopaedia (March, 2017).
 Martin, Whitney. The Problem with Using Personality Tests for Hiring. Harvard Business Review (August, 2014)
 Kanengoni, Herbert. Bias in personnel selection and occupational assessments: Theory and techniques for identifying and solving bias. International Journal of Psychology and Counselling. Vol. 5(3), pp. 38-44, May, 2013
 IBM HR Analytics. IBM Website.
 Oracle Human Resources Analytics. Oracle Website