At the SHRM15 conference there were numerous presentations on the use of people analytics. This involves the use of big data to decide who is to be your best employee, who is likely to be your worst employee, or who is likely to be the next employee to leave. I participated in a panel discussion and Tweetchat on the use of analytics in the workplace hosted by IBM on a #smarterworkforce. As we rush to embrace and adopt the predictive analytical process we do need to keep in mind that there are, as of yet, some limits to the process and value of big data analytics.
Value of Data Is Determined By the Questions Asked
There is an old saying that applies to the use of computer and data: “Garbage in, garbage out.” It was originally an admonition about how you wrote a program that then transformed to a statement about the data you selected to analyze. That adage holds today for the use of big data as well. The conclusions that are reached will depend on the questions that are asked.
As Daniel Goleman pointed out in his HBR article What People Analytics Can’t Capture is that if you are measuring someone on their results you need to pay attention to how they achieved those results. He relates the story of a manager who was good at getting employees to perform and achieving the desired results. Data collected on this manager would show that his groups were consistently high performers and achievers. However, further review of the details of this achievement showed that this manager intimidated and brow-beat employees to get the desired results. He was able to get these kinds of results year after year because no one stayed and he always had a new group to work with. If you asked a question about performance you got one answer, if you asked one about turnover, you got a different answer.
Algorithms Need Hard Data
The best algorithms need hard data on performance to give an accurate picture. If we looked at financial performance only we would get a different picture of performance than if we looked a turnover data and tenure of employees. A manager who gets good performance and keeps employees could be deemed to be better than a manager who gets great performance but has a new staff every year. Right now we may not do a good enough job of parsing the data on the “softer” issues. Goleman points out that Google realizes the limitations of using big data on promotions and refuses to use them for promotional purposes. That should speak volumes about the state of the science.
We Are Still Working On the Science
We have to remember that algorithms are built by people. The decisions on what data is collected are made by people, at least for the time being. This makes them susceptible to biases. Without an awareness of these potential biases, decisions could be made on conclusions derived from the wrong set of data based on the wrong set of questions.
I am not saying we need to eschew the use of big data, in fact we cannot, but we can’t have a healthy skepticism of their conclusions if we don’t understand the questions asked. Learning about the limits of big data analytics means learning to use them more effectively. One of my quotes from IBM #Smarterworkforce roundtable discussion at SHRM15 was “As we are moving into analytics we need to find a balance and keep the human in human resources.”
Think about it.