Follow
IBM
On a journey of Discovery and Innovation

Employers, can you figure why your people leave?

The decision to quit a job is a very important and complicated one.  How do employees take this decision? Are there set patterns of behavior on employee attrition? Can employers decipher these patterns, and use them to reduce attrition? I believe, based on my work with some leading companies, employers who are smart enough to ask the right questions about their employees will be able to predict their employees’ behaviour, sometimes even before the employees themselves realize that they would eventually be making a decision on their employment.

There are so many reasons why employees leave a job. The common hypotheses that I have heard making rounds in various companies are that of high performing employees leaving the job due to inadequate recognition or compensation, employees being in the same role for a long duration leaving the job due to lack of growth opportunities, employees not being able to work with their first line reporting quitting the job, employees unable to manage work-life balance quitting their work to concentrate on life, etc… While the above is a small list, every HR department would have hundreds of such hypotheses. Which among the above list are correct? Most of them seem logical; then does it mean that they are all true? The best way to separate out truth and perception is to collect data and statistically prove or disprove them. With the advent of Big Data Analytics collecting diverse data from within and outside the company and arriving at meaningful insights has become a lot easier.  Big Data Analytics can tell us which two or three top reasons contribute the most and which hypothesis that we had is actually wrong. This will help HR executives focus their energy on high-yielding projects.

The approach to come up with the best predictive model for explaining attrition varies by the industry and I have seen it vary across companies in the same industry as well, depending on their operating model. The first step in coming up with a predictive model is to list the data that needs to be collected. Companies mainly depend on surveys such as employee satisfaction surveys and exit interviews to gauge the reasons on why employees quit. While they provide useful insights they are limited by the one-size-fits-all approach that is taken for such surveys. Adding to the fact is that overall satisfaction could be one of the many factors which could explain employee attrition and not the only factor that would determine employee attrition. Existing studies indicate that “age, tenure, overall satisfaction, job content, intentions to remain on the job, and commitment are consistently and negatively related to turnover. Generally, however, less than 20% of the variance in turnover is explained.”

Other than the ones listed above, I see there is lot of industry and context specific ‘employment’ attributes that could affect employee attrition. The most common factors among them include compensation and employee performance within the company. There are quite a few environment related attributes that contribute to employee attrition as well, such as industry growth, availability of skills and time taken to train new employees.

Once the data is collected, we should look at correlating it with the employees and alumina and see the differences. There are a few predictive analytics tools which can help churn data and decipher the actual reasons behind employees leaving. The tools themselves decide on which data points are more important and which don’t contribute to any predictive insights. There are various kinds of predictive models that could be developed. Decision-tree based predictive models are easy to understand and provide a rule based output which can be implemented easily. Sometimes the need is not necessarily which employees will attrite but the attributes of employees who attrite, the attributes when put together can provide us the answers on why they attrite.

In particular, a marketing firm that we did work for, the employer was highly worried about high performing employees leaving work. The predictive analytics engine on the contrary came up with rules saying that high performers were many times less likely to leave the organization. The causality could be explained by the fact that they were comparatively very well paid to other employees. The rule also indicated an abnormal amount of new employees who were falling into the low performance category and leaving their organization at a very early stage.

It was pretty clear for anyone seeing the results of the analysis that the company was not concentrating on developing skills in new employees and giving new employees adequate time to pick up the skills needed for high performance. The new employees sensing the bad environment attrite even if the company is ready to retain them for a longer duration with them being tagged as low performers. Our recommendation to the employer was to come up with a different set of performance measures specifically for new employees and increasing the duration between their joining and the first evaluation. Sometimes we have to go ahead and give diametrically opposite recommendations to the organizations than what they were expecting, because data, when asked the correct questions, always provides the correct answers.

Srinivasan Govindaraj, Advanced Analytics SME, Business Analytics Growth Initiatives, India Software Labs, IBM India

Disclaimer: “The postings on this site are my own and don’t necessarily represent IBM’s positions, strategies or opinions.”

Post Your Comment
Name
Required
Email
Required, will not be published
Comment
All comments are moderated
 

Comment
D Gurumurthy, Head - HR
Dear Sir, Leaving of the people from jobs are for different reasons. Firstly when the alternatives are available, they will leave. Otherwise they will continue in the present jobs when they are only dependence on job. The other persons leaving the job are for different reasons of their personal reasons. Certainly better prospects are now available to the persons who are having high caliber persons. People leaving the jobs will tell so many reasons. But only very few will leave others will continue in the same jobs even though they are in discontentment. It became common phenomena in present trends. D.Gurumurthy
Nice article, and has pointed in many comments getting the right data is the huge problem in this domain. So if the collection of data is organized as a continuous process through out the tenure of all employees this would help to could get better insights than just relying on the exit interview or other general feedbacks. More companies are already involved in this practice and yes since its an evolving area combining the predictive models with the Domain business rules should be the approach.
In my view, presently in many MNC company's most are resigning because of salary. Both employee and manager, even sometimes senior management are underpaid(30% to 40% less than market standard). Company think that because of brand name people will be stay back, But currently employees are more smarter, and they know the value within them.
Srinivasan Govindaraj
Yes in a many studies, we have found that relative salary levels (including incentives) to be one of the top 3 predictor variables of importance. However the most important insights employers get is around the interaction of salary levels and experience range. This usually points out groups that need focus.
I am still trying to think whether EXIT INTERVIEWS extract the real reason for a change. No professional wants to burn the bridge with the organization that he is leaving (If he/she sees, even a remote possibility of returning back to the same Organization sooner or later). Considering this scenario, THE GENUINE reason for leaving the company is still undiscovered. Reasons like, Going for Higher studies, Have a Dream Offer with me, Want to go back to my Native will get recorded more frequently.
Srinivasan Govindaraj
Yeah as stated in the article exit interviews have the above fallacies, which is why we would recommend usage of more operational data from within the organization as noted in the comment below on Ashok's.
I agree with the power of Analytics to find the real reasons and find patterns for employee attrition. The problem lies in companies ability to plan the exit interview in a manner that employees can provide the real data. The challenge will also be in terms of how to leverage the outcome of such analysis as most of the time the root cause of problems like high attrition lies in foundation of the company.
Srinivasan Govindaraj
Hi Ashok, Yes many times the root cause might be in a core policy of the company, This analysis will highlight it and probably highlight likely solutions as well. The other point I wanted to add is that, exit interviews is one of the starting points, but we rely heavily on the data from the time the employee was employed with the organization before he changed his mind. The data would include but not limited to employee's work history, his performance, rewards etc... when (s)he was employed with the organization.
I was involved in an exercise to create a predictive analysis model with dependent variable as "estimated cost of medical treatment" and independent variables as disease, hospitalization duration, procedure etc. The data came from IRDA site that has about 1 million actual records of hospitalization in India with real expenses and diseases. We tried many combinations of variables to get a good R square, the maximum that we got was about 40-50%, good enough I guess? As Srini mentions in his article, less 20% of variance in turnover could be explained by the chosen parameters? Other comments have already brought some valid points on quality of data, subjective parameters etc. I would just like to state that earlier HR consultants such as Marcus Buckingham, Henry Mintzberg even Peter Drucker used personal interviews as a means to gather reliable data and then applied their vast experience and judgement (subjectivity) to interpret the results and discover causality. We don't have so many Druckers in world but can Big Data analytics serve as an alternative?
Srinivasan Govindaraj
The idea of using big data to supplement experts is a good one. I think what the current set of predictive analytics tool set can do is elevate the thinking of business analysts, by providing insights from data and also recommend possible actions. The speed with which these tool set can transform data to insights and insights to actions can impact the day to day decision making of business analyst very positively.
Dear Srini, Though attrition happens regularly, it includes a mix of high performers, poor performers, average performers who don't like company policies, associates going for higher studies, associates leaving job to take care of family / joining spouse abroad etc. The only crowd which the companies really worry about are, the 'High Performers', who find that they have overgrown their Salary / Role / Challenging environment. And there is a specific but a small set of attributes they look for, when they switch company. So my doubt here is, do we need these Analytics and Big data technologies for this relatively minor population for that company ? ... Will it be worth the investment for the company ...or will they be capturing the same data what they would have captured by a casual Exit interview.
Srinivasan Govindaraj
Return on Investment is a key metric that needs to be evaluated before deploying any Big Data Analytics project. Estimates show that employee attrition costs a minimum of 30% of an employee's remuneration package to upwards of 150% of employee's salary (http://en.wikipedia.org/wiki/Turnover_%28employment%29#Costs). Hence for large enterprises even a reduction of 1% attrition will result in multi-million dollar savings. Some companies (like the Public Sector Undertakings mentioned in Apurva's comment) where attrition numbers are so low, might not need the use of Big Data to solve the employee attrition problem. But they might need Big Data Analytics to solve different kinds of problems.
As Apurva pointed out, there are lot of qualitative inputs that go into the resignation decision. We can still create the derived variables and try to get the predictive capability but the viability of the model would be in question. We tried something similar in our company where we modelled the likelihood of the candidate to join the company given an offer. And results were extremely positive with respect to this requirement and we could do with fewer qualitative variables and more quantitative variables.
Srinivasan Govindaraj
Thanks for bringing out the analytics use case on Offer Acceptance. That is a very important use case in HR Analytics as well. It will help plan organizations to roll out the optimum number of offers. The point on Qualitative inputs is very valid. As qualitative surveys from peers / managers can reveal more than what is usually captured in the systems. In one of the projects, we had implemented a survey of managers on their employees on a continual basis. This was then used as an input to the existing data about the employees. We did see a good improvement of the model over and above what we were able to predict only with the operational data that we used before.
Whether you make use of big data and predictive analysis or just a simple trend analysis using MS Excel - it always goes down to the quality of data. Quality of data, especially when an employee leaves is never good and is highly questionable. The employee is an emotional churn and worried if his/her move will bode well for his future - no matter how good or lucrative the new job offer is. The companies also do not do a great job at making resignations hassle free - the employee is offered counter offers and reasons why he or she should not resign and continue with the organization. Due to all this churn and influx of ideas in an employees mind, the answers he provides in exit interviews do not truly reflect his views - usually the employee doesn't provide true answers just to have lesser hassle during this time. If you were to ask your friends or colleagues who left the company - more than 80% of the time it is due to higher pay, better job prospects and career growth. Companies deploy fancy analysis terms like Six Sigma, Big Data etc just to overlook the truth because usually there isn't much most of the companies can do about it. Why do some companies like Google, Apple, Facebook etc have far lesser churn that other companies like IBM, Infosys, TCS etc. Or for that matter compare why our own companies like SAIL, NTPC, ONGC, Indian Oil have far lesser churn - it is because either some companies spend significantly on their employees or the employees do not have much options in the market or other companies.
Srinivasan Govindaraj
Hi Apurva, I completely agree with you, on most points that you have stated here. In fact, most of the time in building good predictive models goes into data preparation, i.e. getting all possible data some of them could be external sources relevant to the subject of investigation (e.g. in this case overall attrition trend for the industry, change from previous years etc...), ensuring their quality is good (deal with missing data, correctness of the data, usability etc... ) , correlating it with multiple sources provides you real insights on how accurate some of the subjective inputs are e.g. does the exit interview mean anything or do we need other sources to supplement it. Big Data Analytics provides you the tools and techniques to do a thorough analysis with multiple data points in a way it was not possible before. It gives an objective outcome, which will help in decision making and also a method to act on the insights in real-time. This we have seen, gives the competitive edge to the companies committed to their cause.
Apurva, I completely agree with you that compaines like IBM, Infosys etc have far more attrition rate as compared to other firms. Throughout my career, I have worked with big MNCs including IBM and I must say that no matter whatever data analysis approaches they publish in market place, these companies never value their employees. Moreover the questions asked in exit interviews have no relevance to improving the employee benefits or employee retention policies.
This entirely depends on whether you cherrypick people for ultra high levels of technical excellence and then go to huge lengths to retain them (eg: google, facebook etc have free five star hotel quality food / coffee etc on their campus, free rides to / from work, rest areas where people can sleep if they are coding all nighters, bring pets to work to reduce stress etc). Compare that to a typical "cube farm" located miles out on the highway where the only food is subsidized but tasteless in the office cafeteria, coffee is not free, etc. Then compare the companies that have forced curve fitting for appraisals versus those companies that have a more holistic appraisal model where there is no compulsion to rate 25% of your employees in the bottom two categories of appraisal, and where bonuses etc are not tied to high appraisals [which tends to foster a vicious competition within a team, rather than cooperation]. So if you compare just the exit interview results, or attrition / churn rates in companies, but don't dig into employee hiring, appraisal and promotion processes, you'll find yourself acting out the story of the blind men and the elephant. Or maybe blind men and several elephants, if you use big data.
As for PSU jobs as someone else asked - ultimate job security (at times, for far less work than in the private sector), lots of extra perks such as subsidized housing, etc.
 
 
IBM
IBM has always been a company in a state of constant renewal and reinvention. Through economic upheavals and natural disasters, tech bubbles and recessions, it continues to engage clients, governments, local communities and universities to improve how the world works. Differentiated by values, strengthened by collaboration and experienced through the IBMer, today their solutions in Cloud, Big Data, Analytics, Mobility, Predictive Intelligence and others are making the world smarter.

Through its Blog posts, IBMers will explore some essential areas of business and life that are deeply interlinked with technology and would like to invite all to share experiences and comments as it continues on this journey of discovery and innovation.
 
 
 
Most Popular
IBM's Activity Feed
December 31, 2013 08:49 am by D Gurumurthy, Head - HR
Dear Sir, Leaving of the people from jobs are for different reasons. Firstly when the alternatives are available, they will leave. Otherwise they will continue in the present jobs when they are only dependence on job. The other persons leaving the job are for different reasons of their persona...
December 25, 2013 10:02 am by SL
Nice article, and has pointed in many comments getting the right data is the huge problem in this domain. So if the collection of data is organized as a continuous process through out the tenure of all employees this would help to could get better insights than just relying on the exit interview or ...
December 19, 2013 12:59 pm by Srinivasan Govindaraj
Thanks for bringing out the analytics use case on Offer Acceptance. That is a very important use case in HR Analytics as well. It will help plan organizations to roll out the optimum number of offers. The point on Qualitative inputs is very valid. As qualitative surveys from peers / managers can r...
December 19, 2013 12:46 pm by Srinivasan Govindaraj
Return on Investment is a key metric that needs to be evaluated before deploying any Big Data Analytics project. Estimates show that employee attrition costs a minimum of 30% of an employee's remuneration package to upwards of 150% of employee's salary (http://en.wikipedia.org/wiki/Turnover_%28emplo...
December 19, 2013 12:17 pm by Srinivasan Govindaraj
The idea of using big data to supplement experts is a good one. I think what the current set of predictive analytics tool set can do is elevate the thinking of business analysts, by providing insights from data and also recommend possible actions. The speed with which these tool set can transform da...