To all recruiters — use machine learning to hire better candidates

Photograph by Luis Llerena (https://unsplash.com/@albertosaure)

Currently, most companies invest a lot of time and resources into finding the right people for a job. This can range from spending more money on advertising the position to contracting more headhunters to find more candidates. With the amount of information that is available in the hiring process, machine learning can unearth much more efficient methods for identifying strong candidates. Machine learning is the study of pattern recognition and computational learning theory in artificial intelligence. By creating a “model”, which is essentially a trained data set made from example inputs, companies can create accurate predictions or decisions as outputs rather than following a static procedure. This model can be used to make smarter choices in the recruitment process.


Proposal Overview

This business proposal examines the role of big data analytics in talent acquisition based on your company’s desire to improve existing hiring processes. As your company continues to invest in human capital, we strongly recommend bringing big data analytics to your Talent Acquisition department as it can drastically improve the hiring process by reducing human capital costs and providing more accurate hiring decisions.

This proposal begins with an introduction to business analytics as they apply to HR. It continues by delving into how workforce analytics can be applied to the recruitment and selection process specifically, with case studies on how two very different companies have successfully implemented analytics. We provide these as context for the next section, ‘Recommendations for Implementation’, in which we explain more specifically how your company might implement analytics as part of your talent acquisition process. Finally, we make the business case for investing in talent analytics. In this section, we will provide evidence of how investing in this area will positively impact your bottom line and help your company continue to grow.

Introduction

“Big Data” is a term which, over the past few decades, has become ubiquitous in nearly every industry as our capacity to handle millions of pieces of data has drastically improved; however, the concept of big data analytics is still foreign to many Human Resources departments. Within Human Resources, big data analytics has yielded especially promising results in the talent acquisition process where companies handle thousands of resumes per year. Using machine learning, data scientists can produce faster and more accurate hiring decisions than professionals alone.

When you sought our expertise on improving your current HR processes, we knew that technology would be the strongest area for investment — your investment in technology across other business functions signal that it is already a clear priority. As your company continues to invest in human capital, we strongly recommend bringing big data analytics to your Talent Acquisition department as it can drastically improve the hiring process by reducing human capital costs and providing more accurate hiring decisions.

Background

At the turn of the 21st century, “big data analytics” for businesses became more popular than ever before. Big data analytics integrated new technology for managing and analyzing data into the business world and provided a means for businesses to make sense of millions of pieces of data at once. In the wake of these advances, companies have responded in one of two ways: to invest in analytics for many facets of their business, or to avoid analytics because they misunderstand the topic (Fitz-enz, 2009).

Now, nearly a decade later, survey-based reports find that firms are currently spending an estimated $36 billion on data storage and infrastructure (LaRiverie et al., 2016). This quick and drastic increase in the use of big data analytics has provided a tremendous opportunity for businesses, and companies which invest in analytics are using their data to gain a competitive advantage on their peers.

In its early stages, analytics investments were primarily focused on business functions such as finance, which by nature involve data analysis. Human Resources departments are now more frequently beginning to use analytics to streamline and capitalize on their existing processes. HR processes are generally looked at in terms of an employment cycle: talent acquisition, onboarding, training, talent management, employee well-being, and termination.

As analytics have been introduced in HR strategy, companies have begun to utilize them in every part of the process. As talent analytics can provide high return in every part of this cycle, talent acquisition is perhaps the best place to begin implementing data analysis. Recruitment is the first stage in the HR cycle, so the effects of investment in this area will be felt quickly throughout the company.

Companies which invest in data analytics are already seeing great returns and a distinct competitive advantage in the hiring process. Two prominent companies have transformed their hiring processes through the use of data analytics: Google and JetBlue. Though the companies are drastically different in many ways, their commitment to talent analytics sets them apart from their peers. The following case studies highlight how the companies succeeded in implementing analytics as well as some of their most important discoveries since implementation.


Case Study: Google

Technology giant Google has been heralded as a leading company in HR analytics. Google’s HR Analytics team has been known to study issues ranging from compensation to diversity training programs, sometimes going as far as to invite academics to collaborate and provide their expertise. Though many of the “People Analytics” projects the company has invested in are highly confidential, a handful of case studies and articles have been published which highlight the company’s findings.

In his Harvard Business Review article, David Garvin describes what is perhaps People Analytics’ most famous endeavor, Project Oxygen, which “sold Google’s engineers on management.” The project sought to understand and improve management practices at Google by collecting data–starting with exit interviews and ending with employee survey data–to understand what makes managers effective. Initially, some within the company speculated that the engineer-driven organization actually didn’t need managers. However, after analyzing data points from hundreds of employees, the People Analytics team was able to not only identify what makes managers successful at Google, but also to highlight the importance of managers, even in a very flat organization (2013).

Google’s approach to talent acquisition has been equally successful. The company used analytics to calibrate the ‘ideal’ number of job interviews for a candidate from ten down to five, saving hours of time and millions of dollars on recruiting costs (Davenport et al). Laszlo Bock, head of Google’s People Operations function since 2006, goes into great detail on the evolution of Google’s hiring processes in his book Work Rules! He describes how the company started out slowly–they recruited and hired only the best, and still live by the mantra that you should only hire someone who is better than yourself in some meaningful way. However, as the number of applications to get a job at the company skyrocketed, so did the amount of data the company had for predictive analytics.

Now, as Bock describes, the company has realized that there are four distinct attributes that predict whether someone will be successful at the company: general cognitive ability, emergent leadership, “Googleyness” (a combination of intellectual humility, conscientiousness, comfort with ambiguity, and evidence of taking courageous or interesting paths in your life), and role-related knowledge. In Bock’s time at Google, the company has stopped hiring for Ivy League degrees and started hiring for ambition. He shares that the company now has more faith in a top student from a large state school than an average Ivy League graduate.

Beyond hiring for the right people, Google has, “produced an algorithm to review rejected applications…that has helped the company hire some talented engineers its screening process would have otherwise missed” (Derose). This example illustrates that the company goes to great lengths to ensure their predictive algorithms are based in solid science, and settles for nothing less than perfection. Google employees hold themselves to very high standards, and as a result, employees and company executives trust the decisions made with the assistance of big data analytics.

Case Study: JetBlue Airlines

Of course, not every company has Google’s data-savvy nature, nor do they have software engineers on hand to develop algorithms for the People Operations team in their free time. A handful of other companies have used data analytics to master HR processes. For example, JetBlue Airlines has seen tremendous results from using data analytics in their hiring processes. In the Summer of 2015 at a Wharton Business School conference, two of JetBlue’s talent experts shared some of the secrets to their company’s success.

JetBlue has catered their analytics to their business needs. Instead of hiring engineers for intellectual ability, JetBlue hires flight attendants for helpfulness. An article from the conference shares that, “Last year, though, the company learned an interesting lesson… ‘For our flight attendants, we had always looked to find the nicest people we could possibly find to be in the sky with [customers],’ Then some customer data analysis, performed in conjunction with the Wharton School, yielded unexpected results: ‘Being helpful trumps being nice. Being helpful even balances out the effect of somebody who is not so nice,’… ‘People will tell you they know the right kind of person for a given job. But what we think isn’t always what is best.’” (Wharton)

JetBlue and Google have both used data analytics to revitalize their hiring processes in a way that works to eliminate biases, make processes more efficient, and ultimately save the company time and money. Both have also successfully leveraged their unique company cultures and missions to design an analytics program that answers the right questions. Although Google emerged quickly as a hub for talent analytics, JetBlue proves that any company can find a way to apply data techniques to improve their hiring processes. Like Google and JetBlue in their pre-analytics days, your company has a tremendous amount of untapped information at its fingertips, and it is time you use it to your advantage.


Suggestions for Implementation

We suggest focusing analytics efforts on the Talent Acquisition department. There are several potential applications for analytics in the hiring process at your company, and the first issue for recruiters in your workplace is always going to be finding the right person for a role. Your recruiters have access to massive networks, but no effective way to leverage the connections without committing more time and resources. Rather than hire more recruiters, machine learning can be used to recognize data points from applicants’ resumes in order to narrow down the field of search. Instead of spending hours sifting through resumes, your recruiters could be more committed to research on the personal level. A similar process could be used to determine which candidates are less likely to leave their position based on information from their previous history–yet another task recruiters could potentially avoid. For every tedious task replaced by machine learning algorithms, your company would experience a huge return on investment: investing in analytics will save both time and money.

Beyond a resume, the interview process is also very important for companies when determining the right candidate. Most companies have interview questions (both technical and non-technical) that are usually reused among interviewees. Reusing the same questions may seem like a poor choice at first as the chance of a candidate seeing the problem before increases over time. However, reusing questions also allows the interviewer to properly assess a target candidate with a controlled source. Over time, interview results could be fed into a machine learning model to predict whether a candidate should be hired.

Not only do machine learning models expedite the decision-making process, but training a machine learning model on interview data would actually allow a better assessment of job fit. Under this model, a hiring committee would use a series of interviews to decide whether or not to hire an employee. After the employee’s interviews, the recruiter would go back and label a candidate as either “would have been a good hire” or “would not have been a good hire”. Then the labels would be retroactively adjusted via a feedback loop to compare those who were actually hired with those who would have been a good hire.

After the algorithm is designed, errors are corrected to help train the model, which could determine which decisions are false negatives and which are false positives. False negatives in this case would be people that were rejected through the model classification, but would have actually been a good hire. The issue with this is, after rejecting someone, the recruiter will never know if they truly were going to be a good hire. A possible solution to this problem could be to re-interview previously rejected candidates. This would help the model train with more accurate data. Similarly, this case could produce false positives. A potential hire who was classified as a “good hire” could later be determined as someone who shouldn’t have been hired. Digging deeper, there are two more subcategories: candidates that took the offer and ended up being bad hires and the ones that didn’t take the offer and would have been bad hires. In the second case, we experience the same challenge as with false negatives: the recruiter has no way of knowing whether someone was a bad hiring decision as they haven’t worked at the company.

In addition, it is important to take into account the cost of each decision made by both the company and the model. The cost of a false positive is high–the company has invested time and money into an individual who doesn’t posses the skills or commitment to maintain their position. This can also damage the product that is being worked on (timeline for tasks) or the productivity and morale of coworkers. The costs of a false negative are opportunity costs–the company missed out on a great potential hire that would have maybe furthered development more. With this in mind, recruiters tend to favor false negatives and hope that candidates re-interview for the same position in the future, which allows for more reliable predictions on data.

While there are both strengths and weaknesses involved in applying machine learning algorithms in the hiring process, the result is ultimately a huge improvement over your current manual processes. While algorithms may produce false positives and negatives, human error has always been a threat to the integrity of the hiring process. Implementing these algorithms would not override human decision-making entirely, but it would help simplify and guide the hiring process, and the return on investment is surely high.

Monetary Benefits of Implementation

There are many hidden costs in the process of hiring an employee: recruitment, advertising, and selection all come with their own set of costs. Implementing machine learning algorithms requires some up-front investment, but over time your company, like Google and JetBlue, would reap the reward. Using these methods can save your company tremendous amounts of money over time, and the return on investment could potentially be huge.

For example, the costs of replacing a worker can be substantial. Across establishments, replacing an employee costs, on average, about $4,000 overall. The price ranges from about $2,000 for blue collar and manual labor workers, and as high as $7,000 for professional and managerial employees (Dube et al). Machine learning algorithms would help us minimize the costs of false positives and negatives caused by human error in the hiring process. Even now, your company would save substantial amounts of money, but over time, these savings will scale with your company growth. Imagine the same type of savings in every stage of the hiring process–the possibilities for machine learning algorithms in this area of Human Resources are endless.

Conclusion

As you begin to invest more in how you hire, develop, and retain individuals in your organization, it is of great importance that you integrate analytics into your decision-making processes. Your company’s investment in technology is clear to us, based on your investment in technology across several business functions. As you continue to expand and your number of employees grows, it is important that this investment extends to your HR function. Though we believe you should eventually implement analytics into the entire HR cycle, talent acquisition is the logical first step. We hope that you will consider implementing our proposal to ensure you continue to hire applicants who will be successful in the workplace.


Authors

Burak Aslan

Allison Traylor

Sources

Bock, L. (2015). Work Rules!: Insights from Inside Google that Will Transform how You Live and

Lead. Hachette UK.

Derose, C. (2013). How Google Uses Data to Build a Better Worker. The Atlantic.

Dube, A., Freeman, E., & Reich, M. (2010, March). Employee Replacement Costs.

Fitz‐enz, J. (2009). Predicting people: from metrics to analytics. Employment Relations

Today, 36(3), 1–11.

Garvin, D. (2013). How Google Sold Its Engineers on Management. Harvard Business Review.

LaRiviere, J., McAfee, P., Rao, J., Narayanan, V., Sun, W. (2016). Where Predictive Analytics Is

Having the Biggest Impact. Harvard Business Review.

Wharton Business School Press. (2015). Should Hiring Be Based on Gut — or Data?.

Knowledge@Wharton.