Harikesh Nair: Human Resources’ Big Data Moment
A marketing professor explains how workplace analytics can transform how companies hire, evaluate, motivate, and retain employees.
How do you motivate employees? | Reuters/Mike Segar
How do you motivate employees? | Reuters/Mike Segar
We’ve all seen by now how big data and analytics play out in consumer-facing industries, in which consumer behavior is now tracked and measured better than ever. But quietly, behind the scenes, there is a revolution also taking place within the workplace itself. Thanks to the widespread adoption of software and database systems within companies, and improvements in tracking technology, companies can now track, measure, and assess employees much better than before.
This opens up opportunities for social scientists to peer within the firm — treated for a long time as a “black-box” — and study its inner workings with real data. By measuring effort and productivity, we can now manage and reward talent better. More important, we can now quantitatively assess employee productivity and answer some fundamental questions of management: Just what exactly makes some employees very productive and some unproductive? Is it innate ability, training, incentives, peer effects, managerial investment, or a combination of all?
This is human resources’ big data moment. And because it touches on the key resource a firm has at its disposal — its people — it touches every functional area within the firm. I believe that quantitative analysis of within-firm behavior and the associated insights it will provide — what we refer to as workplace analytics or people science — will transform how we evaluate, motivate, hire, and retain people in companies in the future.
In a sign of things to come, some forward-looking firms have already put in place initiatives that use data to assess employee productivity and improve employee hiring and retention. Google’s People Science team, for instance, has quantitatively analyzed what makes some Googlers better managers and what types of pay (salary vs. bonus) employees value. Biogen has established a People Strategy and Analytics team that uses predictive models for understanding patterns of attrition, performance, and recruiting among its employees. Workplace analytics capabilities are now being incorporated into enterprise resource planning software, such as Oracle’s Human Capital Management suite, that helps firms collate employee data, build models using it, and use it to source, acquire, and retain key talent.
For several years now I have been conducting research in this area with Sanjog Misra, a marketing professor at UCLA Anderson School of Management. We have focused in particular on developing new methods that leverage the large quantities of within-firm data that’s now available to answer questions about the design of incentives within firms. We have also looked at how firms can align quantitative incentive design with other functions like marketing and employee hiring and firing.
Designing a Better Compensation System
In one study, conducted in collaboration with a Fortune 500 contact lens manufacturer, we looked at internal data to understand how to design better compensation incentive systems within a large sales force.
Although there is a large amount of academic theory on the question of whether and how incentives like compensation work, much of this theory has been just that: theory. The availability of data on contracts and outcomes on employees has now provided unprecedented access to help us understand this problem empirically and to see which theories work and which don’t.
This is important because most of the marketing in business-to-business contexts is done by salespeople. Compensation policy is one of the key levers available to influence them. Though it is complex, getting the structure right is critical.
At the initial stages of our collaboration with the firm, the company used an incentive plan that involved a salary and a commission that paid commissions on sales if the agent’s sales per quarter crossed a quota and fell below a ceiling.
Like most firms, the company faced significant challenges in formulating and optimizing the right quota-plan for its needs. For starters, the quotas needed to be fine-tuned to reflect the significant differences in sales agent productivity. A quota that is too low is always “beat,” providing little room for incentives. A quota that is too high demotivates agents because they feel it is unattainable.
The company also needed to fine-tune the ceiling, which helps the firm from paying out large commissions due to reasons unrelated to the sales agent’s efforts – for example, the number of new prescriptions can be suddenly high merely because a new Wal-Mart opened in the agent’s territory. The challenge is that if the ceiling is set too low, the company reduces the scope for incentive pay; if the ceiling is too high, the company may end up paying out too many commissions.
A third challenge was determining the right periodicity of the new plan – that is, how often commission awards were paid out. Commissions were paid out based on the total sales achieved during an entire quarter. The result is a potential inefficiency: If quotas are very low and easy to beat, agents may find it optimal to shirk in the early months and make up sales later in the quarter. The shirking may be high for the most productive agents, as they know they can easily make up the sales in the last month. This suggests that paying out commissions based on a weekly or monthly sales achievement cycle may reduce shirking. But how much the improvement may be was hard to predict.
Still another aspect relates to the broader question of how to design incentive systems that do not create their own distortions in behavior. Other scholars have pointed out that incentive systems have hidden costs because smart agents can game the system. For instance, many companies pay sales agents commissions only if sales exceed a quota. But if a sales agent feels he or she has no way of making the quota, or has already beaten it, he or she will tend to reduce effort. Or the sales agent may push customers to buy at a time when it suits the agent, which may result in a lost customer.
Previously, many thought these distortions were intellectually interesting, but perhaps not too big in practice. One insight from our recent empirical work is that such distortions may be so large that they could in some instances actually overwhelm the gains from incentive provisions altogether.
A New Plan Yields Strong Results
To analyze the problem, we built statistical models of the sales agents at the company to create several scenarios involving changes to the three key features of the plan: the quota, the ceiling, and the quota horizon. The models combined economic theory on how workers responded to incentives with real data on observed past behaviors to develop a predictive analytic model for each agent in the firm.
We worked with the firm to narrow the range of plans to a set that were feasible for implementation. A new plan was selected in consultation with senior management, sales managers, salespeople, and legal and human resources teams.
The plan that was implemented featured low quotas and no ceilings. It also included a monthly incentive based on a straight commission (a straight commission is a scheme where there is no ceiling and the commission rate does not change with the sales achieved). It was put in place across the United States in January 2009.
The results were extremely strong. The companywide effect of the new compensation scheme was about a 9% increase in overall revenues. Comparing 2009 versus 2008, the new plan generated an average increase in revenues of $79,730 per agent per quarter.
Overall, our results suggest that the new plan was a success on several other dimensions as well. First, it was more efficient. One way the bad incentives in the old plan could have worked is to simply induce a shift in sales away from early months in the quarter to the later months, with no effect on overall quarter-level sales. The results from the new plan showed that the old plan’s effect was not simply to shift sales across months of the quarter in this manner, but to also reduce the overall sales in a quarter. In the new plan, sales went up in every month of the quarter compared to the old plan. This shows that the shifts across the months seen in the old plan were also accompanied by a net reduction in total achievable sales. In other words, the old plan was inefficient.
Second, most agents increased their effort and output.
Third, the new plan eliminated the large swings in sales in the old plan. These had been driven by the incentives the old plan induced for agents to change their effort when they are close to or far away from quota. Importantly, eliminating this volatility also reduced inventory holding costs and streamlined supply-chain and capacity planning. Finally, data from surveys conducted at the firm showed that employee satisfaction with the new plan was high, arising primarily from the reduction of quotas and the subjective assessment of productivity under the old regime.
How Data Aids Decision-Making
Overall, we believe the kind of approach we developed has the potential to significantly improve the practice of compensation design. It is rigorous and practical, utilizes internal databases, and is built on sound theory. It also showcases the value of combining models with large datasets for improved decision-making.
This kind of research also leads to another key theme in organizing workplaces: the alignment between functional areas within the firm; in this case, between sales, marketing, and hiring and retention of employees. In many instances, incentives are not balanced within a firm because these decisions are split across various units, and each has different goals.
It does not have to be this way. In recent research that followed this study, we discussed how hiring the right set of sales agents affects the company’s ability to provide incentives – while at the same time providing the right set of incentives to help hire the right set of agents. In this sense, the right strategy is to make them co-dependent.
The advantage of being close to the data is that we can quantify the extent to which such alignment helps improve outcomes for the firm.
As these studies have demonstrated, analytics and data have the potential to transform the study of work. This is the promise of the new people science. Firms headed by forward-looking business leaders who understand the value of science and its ability to improve practice will benefit the most from this promise. Such leaders will be early adopters of these tools and will use them to make their workplaces more efficient. And in the long run, most firms will likely follow suit.
For media inquiries, visit the Newsroom.