The AI Playbook: 6 steps for launching predictive AI projects

5 ‘extraordinary turnarounds’ for global well-being

Close proximity is the key to knowledge spillovers among startups

Credit: Laura Wentzel

Ideas Made to Matter


People analytics, explained


When the COVID-19 pandemic struck, some companies were better prepared than others to reorganize and mobilize their employees. Those organizations had effective talent analytics strategies in place, according to Emilio J. Castilla, a professor of management at MIT Sloan.

Castilla defines talent analytics — also known as people analytics — as “a data-driven approach to improving people-related decisions for the purpose of advancing the success of not only the organization but also of individual employees.”

Not all companies understand what talent analytics really is, and therefore can’t have an effective strategy, Castilla said. But when done right, talent analytics allows you to know your people so well that you can keep them and your organization productive and effective, even in a constantly changing environment.

Castilla cited one large service organization that, pre-pandemic, had already been comprehensively collecting and coding high-quality data about the skills, capabilities, and knowledge of its employees, managers, and even past applicants.

When the health crisis hit, management at the firm was able to quickly identify talent and reorganize its workforce to operate successfully in a new world. If it needed more help in particular areas, it was able to hire and onboard new employees fast. It had existing technology platforms already in place that enabled it to retrain people, to keep everyone connected in the age of social distancing, and to give employees the right information to make fast decisions in a quickly changing environment.

“They did not even experience the disruptions that many other companies experienced,” Castilla said. “They already had well-prepared and tested ways of reorganizing work.”

In his MIT Sloan Executive Education course, “Leading People at Work: Strategies for Talent Analytics,” Castilla lays out four simple steps to successful talent analytics:

  1. Generate important people-related questions and problems to address.
  2. Plan a project to gather and analyze relevant data to tackle questions and problems.
  3. Report the results.
  4. Use those results to drive company improvement for all.

The steps seem straightforward, but require careful thought and a wise strategic perspective to obtain helpful results, Castilla said.

Say your company has difficulty attracting and retaining the best talent, and senior leadership wants to identify the best way of improving employee attraction. Castilla suggested that you first conduct a review of existing research, frameworks, and case studies to find out what other companies have done and which approaches have proven most successful.

Next, go narrow in defining your own terms and objectives. For example, is your goal simply to increase the number of applications so you can be more selective? Castilla guessed not. “Numbers of applications may not be necessarily the best measure,” he noted. “What you likely mean is you want to maximize the attraction of qualified applicants who accept your job offer right away, are motivated, and stay productive in your organization long-term.”

Be careful with nebulous terms like “the best” or “star performers,” Castilla said. It matters a lot how exactly you measure what makes a hire a “high performer,” he cautioned.

Avoid data bias

Talent analytics encompasses a broad range of carefully collected information, starting with information gathered at the time of application, followed by salary, role, project allocation, performance-review history, engagement surveys, and career goals, among other data.

As you start to formulate what data to collect, be comprehensive. There is a tendency to unconsciously reinforce initial assumptions or biases by collecting only certain types of data. Make sure you identify, and then gather, the variables that you would need to properly test your hypotheses, while accounting for other alternative key factors.

For example, one large international financial institution wanted to evaluate its return on investment of employee referral programs in its call centers. Were employees recruited in that manner more productive and did they stay longer than employees recruited in other ways?

Read more analytics coverage

Some leaders questioned why the company even needed a referral program. Shouldn’t call center employees be happy to refer friends and family without any financial reward? Other executives worried that such programs might foster nepotism rather than improving the quality of the workforce.

When making data decisions in this instance, Castilla said, “You need make sure to collect pre-hire and post-hire data that will either support or refute your hypotheses while controlling for information about the applicant collected at the time of application.” That data might include relevant qualifications collected from resumes or from semi-structured interviews of your candidates as well as performance and turnover data. “The analyses of such data can later help you identify improved ways for implementing your recruiting practices,” Castilla said.

Quantity and quality matter

Although companies say they want to be strategic in talent analytics, they often have sparse or poorly collected data, Castilla said. “As a result of that poor data quality, any analysis of such data will lead to poor results and, therefore, undesirable and poor business recommendations.”

In particular, companies often tend to select on the dependent variable — that is, they only analyze observations in which the phenomenon of interest is observed and exclude those in which the phenomenon is not observed, Castilla said. To best test their hypotheses, he encourages companies to collect a comprehensive set of variables that control for potential alternative explanations for their problems and concerns.

It would be a frustrating exercise if, at the end of the project, you found certain recruits were more productive than others but didn’t quite understand why, Castilla said. Maybe it was because of their educational level or their years of particular types of experiences or their set of abilities and skills or the way they were recruited into the company, but you won’t know if you didn’t collect and included these factors in your analyses.

In the real-life call center case described by Castilla in “Social networks and employee performance in a call center, the analyses of the data confirmed that workers hired via employee referral were initially more productive and less likely to turnover than non-referrals. How long they stayed and continued to be highly productive, however, depended on post-hire factors.

Specifically, if the employee who referred them helped train and mentor them and provided other support, that could result in increased productivity and job satisfaction and reduced turnover. However, if the original referrer left the company, that could have a negative impact on the performance of the referred hire, even if they had been working for the organization for a long time, Castilla’s study found.

Missteps to avoid

As you reach step 4 — using data results to drive company improvement — anticipate all the potential consequences of your plan. Companies often underestimate the new issues that their planned actions may introduce.

For example, your program to focus on increasing attraction and retention might negatively affect a corporate goal of maintaining or increasing the diversity of your workforce, Castilla said. In some of his prior work, Castilla found there may indeed be unintended negative consequences behind the adoption of certain organizational practices precisely aimed at promoting meritocracy in the workplace.

“I’m pretty sure your goal is not to attract and retain talent no matter what, and you would be concerned about undesirable potential outcomes behind your people-based decisions,” he said.

Another common mistake is not including human resource experts in your high-level strategy discussions. Castilla has been involved with projects where top management tasked HR with collecting and preparing certain data for analysis, but neglected to keep HR in the loop as strategic objectives evolved. “HR didn’t know management had abandoned the initial plan, nor why there was a change in strategy,” he said. “It’s a huge mistake when HR is disconnected from what’s happening at the strategic level.”

‘Huge and sustainable’ returns

In short, when done strategically, talent analytics goes way beyond solving one particular problem. “Your returns should be huge and sustainable,” said Castilla, with the long-term goal of advancing not only the success of your business but also the success (and motivation and welfare) of your employees.

“Organizations that invest in creating and developing a talent analytics approach that is strategic, people-centered, and long-term-oriented are better prepared to address challenges like the one we are currently living in,” Castilla said.

For more info Tracy Mayor Senior Associate Director, Editorial (617) 253-0065