top of page

Predicting Employee Attrition using Machine Learning & How to reduce Attrition Rate

Updated: Aug 22, 2022

As an HR, one is always told to keep a check on the Attrition Rate of the company. But what does this actually mean and why does it matter so much? In this article, we explain the why and how behind maintaining a good attrition rate and most importantly, predicting the attrition rate of your company. This is coupled with key insights from eZO.Mentors.

What is the Attrition Rate?

If we would have to explain it to a 5-year-old, Attrition rate, or Churn rate, is the rate at which employees leave. Thus, simply put, it refers to the unpredictable and uncontrollable, but normal, reduction of workforce due to resignations, retirement, sickness, or death. Attrition Rate is the percentage of employees leaving your firm in a given period of time, usually a year. Therefore, it goes without saying that HRs desire a low attrition rate, as it means employees are satisfied in their jobs and hence the company has been able to retain them.

Attrition rate can be specified in 4 major types, based on the reason for which people are leaving.

a) Voluntary Attrition – It is when an employee decides to leave the company through resignation, and in some cases absconding.

b) Involuntary Attrition – It is when the company decides to part ways with the employee through Termination, Retrenchment, and other reasons.

c) Internal Attrition – It is when employees move internally from one department to another.

d) Demographic-specific Attrition – When a specific group of people (based on age, gender, ethnicity, etc.) leave. It could maybe because of a toxic type of corporate culture that could be harming the company and its employees.

It is well understood that the first and last types of attrition are the major causes of worry for HR.

Why is it an important matrix?

While for our parents, most of their careers ended in the same company they started their careers with, but as times are changing, we find today’s professionals changing companies every 4-5 years. This has further increased the importance of the attrition rate for HRs and made it crucial for them to keep the attrition rate in check.

Having a clear view of your employee attrition rate is essential to help HRs understand where the company stands in terms of candidate retention. If the employee attrition rate is high, for instance, it can mean that the company is not providing enough benefits or the best work environment to keep its top-performing employees.

By measuring and analyzing the attrition rate, you can identify the problems you need to solve to make sure that you keep your employees from leaving you. It also helps in recruitment planning as a prior estimate of the required workforce can ease the last-minute hassle of hiring.

How to calculate the attrition rate for your firm?

To calculate your company’s attrition rate, divide the number of employees who have left your company by the average number of employees over a specific period of time (you can calculate this on an annual, monthly, or quarterly basis).

Impact of COVID?

Many IT firms in India reported a rise in their attrition rates in a few quarters that followed the global pandemic. In the third quarter of FY22, Infosys reported an attrition rate of 22.5%, while Wipro reported 22.7%. During the same period, the attrition rate at TCS was 15.3%, and Mindtree reported 21.9%. The overall attrition rate of the IT industry after COVID stood at around 25%, which is quite alarming.

Since the beginning of the epidemic, the number of start-ups obtaining unicorn status in India has increased. With growing investment, these start-ups are recruiting the greatest people in the computer sector by offering high pay packages and other advantages, making it difficult for large IT organizations in India to retain their top employees.

How do firms manage to cope with high attrition rates?

Major IT firms in India are battling double-digit attrition rates. They expect this trend to continue for a few coming quarters. These companies have doubled down on hiring freshers and are trying their best to retain existing employees. The Indian IT services industry will hire around 360,000 fresh graduates in FY22, a study by UnearthInsight, a cognitive platform based out of Bengaluru, said.

Companies have announced various options to retain existing employees. For example, upskilling, promotions, salary hikes, employee stock options, and long-term incentives are being offered to existing employees to halt them from quitting.

What you can do as an HR to reduce the attrition rate of your firm?

Some of the ways in which voluntary attrition can be minimized and retain the employees for a higher time are as follows:

  • Recruit and hire the right people

  • Offer competitive compensation & benefits like salary, exposure and projects, higher education, etc.

  • Promote employee engagement

  • Give recognition to employees

  • Offer career progression opportunities

  • Defining clear key performance indicators

  • Employee retention policies for better and longer staying employees.

  • Transparent, periodic, and systematic Review

MONTE CARLO MODEL for predicting the attrition rate

There are many ways to predict attrition rate for a company, which in turn can help in planning the hiring process. However, the big question is – Can we use past data to predict attrition rate? Answer is Yes, past data when clubbed with simulation models such as Monte Carlo Model can help you predicting your company’s hiring needs.

Monte Carlo Modelling, which gets its name from the town of Monaco-famous for its casinos. This is because this simulation too, is based on the game of chance. Monte Carlo Modelling, in mathematical terms, is a simulation used to predict the probability of occurrence and thus forecast outcomes when the possible outcomes are highly random and variable.

Let us simplify it with the help of an example, say a SaaS based firm based in India is trying to predict its attrition rate. It has a total of 1000 employees and historical data shows that on an average 150 employees leave every year, which means it has an attrition rate of 15%. But this figure is an average, and the actual figures may vary like it might be 50 in some years or even be 200 in some. Also, hiring and firing is usually in sync with months. Firms tend to not hire employees by the end of financial year, thus the attrition in those months is also quite low.

We can see the number of employees leaving the firm is not a constant number, which makes it difficult for HRs to predict the hiring needs. For this, we use Monte Carlo.

Monte Carlo Model says that there are two possible choices for an employee, to leave or to stay, where the probability of them occurring is 15% and 75% respectively (based on the attrition rate). Thus, if we can predict the choice of every one of 1000 employees, we can eventually predict the number of employees that may leave by the end of this year. But running these calculations on every individual employee is time taking and also not feasible when the number is big. Thus, we run simulation.

Let’s say we only take 10 employees into consideration; the histogram will look like this:

The average stands at 129, which is close to 150, but still, we cannot make anything out of this graph. Let us try hundred simulations.

Even though this graph has its average, 144, nearer to 150, but it still doesn’t look like something which can help us in predicting the outcome. For this, we need to increase the number of simulations to let say, thousand.

As we can see, the results become more and more reliable as the number of iterations goes up. But why stop here? Let us imagine how ten thousand iterations look! After running 10,000 simulations, the average stands at 150.11 which is in the range of past average. The graph looks as below:

This graph resembles a Bell Curve, which indicates that the data is normally distributed. Now, even if the past figures are unknown, HR can make sense of the above graph and predict that the attrition would be in the range of 100 to 200. This is because there is low frequency before 100 and after 200 and the chances of attrition going beyond these numbers is less. Thus, as HR, we can be sure that there would be a minimum of 100 people leaving the company and can be prepared for the same. This helps is saving it saves a lot of time and resources for HR and eases the process of hiring.

But here the challenge is that we do not know the reason which is the most influential in increasing of attrition. Let us take this model one step further and add a few more series, based on the reason why employees are leaving. To avoid many complications, let us divide these reasons into 5 main categories as mentioned below:

  • Inadequate Growth

  • Pay Gap

  • Work culture

  • Study/ Change of field

  • Personal

Next, we need to assume the turnover rates of these categories. Continuing with our earlier example of a 15% turnover rate, we assume the category rates and run the simulation. A histogram graph like the below can be formed for each category (this one is for Category 1, inadequate growth).

Thus, upon running 10,000 simulations of all the five categories, we get the expected number of resignations for each category. We can take an average of these quantities to know the approximate figure of turnover.

CategoryInadequate GrowthPay GapWork CultureStudy/ Change of FieldPersonal ReasonsSum Average Turnover30.6239.5639.5719.4720.16149.37

The sum here (149) aligns with the turnover in our original example (150). Thus, it is safe to say that out of 150 people who may resign this year, 30 would be leaving for inadequate growth, 40 due to pay gap, and so on. Thus, on a given sample, we can identify that 3% of the people leave because of inadequate growth or 2% of them leave to pursue higher studies. Different categories can be added to make the model more complex and accurate, thus, this model aids in determining the reason behind the resignation of an employee.


Another model to determine the turnover rate is Markov Chain. Markov Chain takes a sequence of possible events in which the probability of each event depends on the state attained in the previous event. Thus, this chain runs by taking the current result as a base for the next simulation. Each event’s probability can be determined by the historical data and thus we can land on the expected values for each category.

Lets us understand this with the help of an example. Suppose the same firm, that we took as an example in the Monte Carlo Model, wants to determine the state of its employees after a year. For this, we need the current state of its employees as well as the probability associated with each state.

Here, we have divided the reasons into 6 major categories as follows:

1. Remains in the current job role

2. Promoted

3. Looking for a change

An employee might have reached to this stage due to inadequate growth, benefits, compensation, manager mismatch, working culture and environment, no client facing, not able to maintain work-life balance, wanting to pursue higher studies, or any personal reasons such as mental stress, health, etc.

4. Resigned

If an employee resigns and joins a new company, this is might be due to better experience, salary, role, promotion, etc. that he might be offered in the new company.

5. Retired

6. Terminated

For example, consider the below table for the current state of employees.

Here, to simplify the model, we have categorized the reasons for a person to leave the job into different stages. For example, the state of resignation can be associated with different reasons such as pay gap, work culture mismatch, growth opportunities, switching to a different field, no client facing, mental stress, etc.

The next step is to find the current percentage of people in each stage, i.e., the initial status.

Now that we have both the initial numbers and the probability associated with each stage, we can multiply both the matrices to find the new % of people in each stage.

On running 3 iterations, we get the following values. Thus, it can be said that if the initial values were for the month of January, the following is the result for the month of April.

Markov Chain model helps in understanding the reason behind why a person is no longer with the firm. It also gives an idea to HR about the potential future loss of human resources to the company, by defining the stage at which each employee is.

57 views0 comments

Recent Posts

See All


bottom of page