Employee attrition is a true headache for any business. It’s never a fun thing to have talented, value-adding people decide to walk away. However, if we know who is likely to surprise us with a resignation letter within the next weeks, we can take steps to make sure they reconsider. But how?
HR professionals collect enormous amounts of data about employees every day: from their background and work history, to demographic data and results of various satisfaction-level questionnaires. But how can we use this data to our advantage?
In the AI era, Machine Learning is the answer, yet expecting these HR professionals to make time to learn Calculus and how to write code isn’t exactly realistic.
The good news is, this isn’t necessary. With Magicsheets, you just press a single button and get the predictions right where your data is: in the spreadsheet.
Everyone can keep focusing on their jobs, and still harness the power of Machine Learning.
To see how our magic works in practice, let’s look at an example using real HR data. You can follow it step by step directly in Google Sheets.
Predicting Employee Attrition with Magicsheets
For this example, we will be using a real HR data set from IBM (you can view it here).
Let’s imagine that we are collecting a variety of data about our employees, such as their previous work experience, the perks they have received, and how they rate their experiences working at the company.
Based on this data, we are hoping to be able to predict:
- Which of our current employees are likely to leave?
- Which of the new hires are likely to leave the company (in the mid- to long-term)?
What are we measuring to determine this? 📏
For each employee, we can collect various types of data, such as:
- Personal data: who the relevant employees are. (Here we include things such as age, gender and educational background.)
- Position details: department, position, and level within the company.
- Work experience: previous work experiences, as well as years spent with us, how long since we last promoted them, and how much time they have been working with their current manager.
- Remuneration & perks: everything given to the employee by the company, including salary and training.
- Work environment: how much traveling is required, as well as work time and over time culture.
- Self evaluation: this is data collected from the employees themselves from interviews or questionnaires, including satisfaction levels, work-life balance, etc.).
- External evaluation: how well the employee is doing in performance ratings, as well as how involved they are in the job.
- Attrition: we know which employees left in the past, and we take note of this.
You can take a look at all data fields for this example in this template.
We need to measure all of these factors in order to predict attrition as they all play into each other. For example: even a very well-paid employee would be likely to leave if they rate their experience as low, due to excessive over time and travel.
Predicting existing employee attrition 👩💻
Now that we have all the data we need, we can simply keep it all together in its spreadsheet. Using the Magicsheets add-on for Google Sheets (which you can get here: it only takes ~10 seconds!), we can now start predicting attrition for our designated employees.
After making sure the “Attrition” column is the final one, it’s time to fire up the Magicsheets add-on.
Note that for this example, the final 10 employees have no “Attrition” field as this is what we are going to try and predict.
- Select “Class”. This means we are predicting a “yes / no” (“will stay / will leave”) result.
2. Select the data. This is the same as good-old Excel. We just select the data and press the ‘select data’ button.
3. Select the location of the output. This is the “Attrition” outcome we are trying to predict for those employees we want to make predictions for (a.k.a the final 10 employees whose “Attrition” field is blank)
4. Press “Predict” and let the magic happen…
Now that we can see the predictions directly in our sheet, we can determine which of our employees are most likely to leave.
We can also see the “score” of our model. In this case, the score represents the predictions’ accuracy: roughly how many of the predictions are correct. For example, an accuracy of 83% tells us that for 100 employees, 83 of our predictions would be correct.
For a small data set like this one (using 290 data points to build the model in question) an accuracy of 83% is a fantastic score already!
Predicting new hire attrition 👩🎓
Let’s imagine we are looking to predict attrition for new employees in particular. Because they are new hires, some data fields might not yet be complete. For example, we might not have any data concerning their job satisfaction level.
Therefore, we will need to consider fewer fields when making our predictions. In our template, you can view all the data fields we are considering for this in the “New hire attrition” tab. The same sequence of steps will be followed to make these predictions:
- Select “Class”
- Select all data fields
- Select the output location
- Press “Predict”
And voilà, you have your results!
Using your own data 🔢
To make predictions using your own data, you can simply follow the same steps to get the predictions you need. Just make adjustments where you need, to make predictions suitable to your company’s unique data.
In many cases, you might not need as many fields as we have used in our example, but that is completely fine. Magicsheets can work with very few, as well as hundreds of data fields.
In addition, don’t worry if you don’t have thousands of data rows. Magicsheets can work with both very small and very large amounts of data, just input the numbers, open the add-on and off you go!
Can I trust these numbers? 👀
At this point, you might be thinking: “That all sounds great, but can I actually trust these numbers?”. One way to assess the quality of your Magicsheets prediction, is to look at the “accuracy” level in the add-on. The higher the accuracy, the better the model that runs under the Magicsheets hood and the more precise the predictions.
How can I make my results better? 👩🏫
Not all results are equally accurate. In the Machine Learning world, your results’ quality will largely depend on the data set used. Here are some suggestions to make the most out of your data:
1. The size of the data set can make predictions more accurate:
You can definitely make a good prediction with only 20–30 employees’ data, yet the results are often likely to be more accurate if you have 200–300 employees’ data for example.
2. Select the features that make the most sense to what you’re looking to predict:
Magicsheets can generally pick out the data fields (the columns) that matter most for the desired results. However, sometimes we need to use our own judgement to decide which predictions are most applicable.
For example, “personal hobbies” might be an interesting conversation starter, yet won’t be as useful to determining attrition — or it might! This is where the HR professional’s best judgement comes into play.
Predicting other things 💡
Magicsheets can be used to predict much more than just employee attrition, however. This is just one HR-focused example of how Machine Learning magic can be used to help you power up your business with our data-driven solutions.
For more ready-to-use case examples, visit our template gallery.