Can We Predict When Our Staff Is Leaving?
Analysing and Predictive Modelling of Staff Turnover
MySupplyChain is part of a logistics MNC with about 650 staff locally. They are quite successful in all their business units: air and sea freight, land transport, container handling and supply chain management.
During the monthly management meeting, the new Sales Director, let’s call him Donald, is trying to learn more about his new employer and asks many questions. As it is HR’s turn, his question is about the annual staff turnover. “About 29% annually” answered the HR Director, Susan. Donald did not comment since he did not want to put his new colleague in an awkward position. After the meeting, he took Susan aside and asked her: “As far as I know this is a bit on the high side, right?” Susan knew her stuff too and nodded to his comment.
Understanding the Business Problem
Very often, Susan and her team are pressured by line departments due to the high staff turnover and the associated negative effects on the departments. The suggestions – and sometimes demands – towards HR include the amendments of salary and benefits.
However, Susan does not easily give in for two reasons. Firstly, such amendments would change the cost base drastically. Secondly, Susan is quite sure that salary and benefits are not the most important drivers for people leaving.
Blaming HR for staff turnover is like blaming the rain when the roads get flooded.
UK
UK
Susan explained “We have just started a project to look into this. Our staff turnover has always been around 12%. But over the last couple of years it has gone up. We are concerned because every exit means a certain disruption to the business operations. It requires our clients and subcontractors to get familiar with new faces.
And, of course, we need to train the new staff, and this costs money and time. We are worried about our branding as an employer, too. We have asked a consulting firm to help us with the analytics”.
The business questions formulated by the management team were:
- What are the drivers for staff turnover?
- Can we predict staff attrition, i.e. who is at high risk of leaving?
- How can we reduce the attrition to less than 15%?
Getting the Data
When trying to develop a useful model for workforce analytics, you will sooner or later run into a prevalent problem: A large variety of workforce data is easily available. However, the available data is often not the data you need for the specific analysis you want to perform. Is this different for our staff turnover analysis?
Scanning through exit interviews and brainstorming potential drivers for staff attrition brings a long list of items in the categories: organisation, compensation and opportunities, job nature, personal factors, and leadership-related factors.
It becomes obvious that not all drivers, all potential causes for attrition, can be verified directly. For some, it is quite hard to get any data. For some, data from outside the organisation is needed in order to perform an analysis because within the organisation, there is no difference.
Namely, everyone in the organisation receives the same package of dental benefits. It can well be that people leave because of that item “Dental benefits are poor”. However, if having only data from internal staff who all receive the same dental package, we cannot check whether people who left had a worse package than people who have been staying. Again, this is because this item does not show any variation, no difference whatsoever.
In order to analyse variation, you need to have variation.
UK
UK
After working on the list of 50+ potential drivers for staff turnover, we zoomed in on some demographics as well as some job related data like salary, job level, training hours we could download from the HR system. Additionally we highlighted some proxy measures for employee engagement scores like job motivation, supervisory practice and teamwork derived from the last employee engagement survey (EES).
Preparing the Data
We collected the data in an Excel table. Each row carries the information, the dataset, for one employee, where column ID marks the employee identifier. When collecting data, especially whilst involving non-HR staff, elimination of personal data like name and email address is a must. Column Left shows the outcome variable Y – representing staff turnover – and has only two levels, Left or Stayed. Column LeftWhen carries the year of resignation.
Columns Department and Section display organisational units. BusinessTravel can be 1_none, 2_some or 3_frequent. Age, Gender and MaritalStatus (1_single, 2_married, 3_divorced) stand for basic demographics. The numbering forces a certain sequence.
Column Tenure shows years with the company, DistanceToHome presents kilometres between workplace and home. In Education, the level of education is marked from 1 to 5 (1 = secondary school and below, 5 = degree and above). JobMotiv, Super, and Teamwork show the rating of the organisational unit in the categories Job Motivation, Supervisory Practice, and Teamwork & Collaboration in the last EES the staff has participated in.
Analysing the Data
As the outcome variable “Left” is discrete with only two levels, left or stayed, the analysis needs a “special” set of tools. Since the limits of MS Excel with its ToolPak are easily reached with this kind of analysis, the powerful analysis and visualisation software R was deployed.
Business Travel
The amount of business travel required for the position was raised as “driver” for staff turnover. However, graphical as well as statistical analysis confirmed that this was just a myth.
Organisational Units
It was mentioned a couple of times, that organisational units have a different staff turnover. This hypothesis was obvious at the plot and could be supported with statistical analysis.
Marital Status
During interviews with HR and with HODs, they claimed that single staff is much more likely to change job, i.e. the organisation than married staff. This graphically obvious assumption proved to be statistically true.
Training Hours
Furthermore, it was mentioned that staff may be less likely to leave if they perceive the organisation investing in them, i.e. training and developing them. The available data supported this hypothesis – although the dot plot not being very convincing.
Always combine graphical with statistical analysis. Relying on only one of them might be a mistake.
UK
UK
Designing a Model
After testing potential drivers for staff attrition one by one, it was necessary to combine them in a model. This way, the interaction between factors will be taken in consideration and the resulting contribution for each factor can be evaluated. For this analysis, we used R and R Studio.
For example, this model served to predict the probability of an HR staff member to leave who is 35 years old, married with 4 days of training record last year. The risk for this staff leaving was only 5%.
Predicting the Probability of Staff to Leave
This model is designed using R and is converted in MS Excel so that it can be used by every manager at any point in time.
It is necessary to feed the newest data into the model regularly to adapt it to changing conditions.
Calculating Prediction Accuracy
This prediction is the result of statistics, i.e., it is in no way perfect. The prediction accuracy shows 85.2% and is calculated on the basis of data of the last three years. This means 544 staff who would be predicted to leave following our model, left in reality, whereas 99 staff who would be leaving following the model did not resign, etc.
Setting Your Priorities
Most importantly, employees want to have the feeling that they are being appreciated, i.e., developed.
Secondly, they value a boss who knows what it takes to support and lead them.
Additionally, age makes an understandable difference. Younger employees are more eager to change and find a better spot if triggered to do so. This is in line with marital status, because Singles are mostly the younger people who are more flexible to change their jobs.
Lastly, the department itself does not make a difference but, so it seems, the working climate and the supervisory practice they find in the department do so.
Making a Business Decision
Summary
- Lack of development options, especially training, provided to them. If we were able to provide a career plan and attractive development options to staff, we would have a very high chance to keeping them longer (Tan, 2014). There is a caveat. It could also be, that supervisors do not invest development options in staff who seems to be on the lookout for other jobs. Statistics will not be able to inform about the direction of the correlation between variables Training-Hours and Left.
- Inadequate supervisory practices (Tan, 2020).
- Younger staff are more likely to leave than older staff.
- Singles are more likely to leave the organisation than married or divorced staff.
- There is a difference in turnover from department to department. Drivers are probably to be found in employee engagement survey results.
The Impact
It was mentioned a couple of times, that organisational units have a different staff turnover. This hypothesis was obvious at the plot and could be supported with statistical analysis.
With this knowledge, crafting a programme to reduce staff attrition and increase their retention is possible.
Consequently, in MySupplyChain, HR recognised that a demanding workload may leave little room for an employee to spend time on learning, which may hold them back from advancing their careers. Hence, instead of justifying a learning budget increase, HR recommended time allocation for learning.
They started by creating fortnightly brownbag sessions to enable employees to learn during their lunch time. This way, they build up the habit for learning, and at the same time, protect time for employees to participate in training. Besides the brownbag sessions, they have also introduced an annual learning festival, which runs over a few weeks, with varieties of topics and learning options for employees to take part at their own convenience.
Summary
Collecting and analysing data beats anecdotes and rumours. For example, based on the anecdotes collected by the supervisors, distance to home was said to affect employees’ retention. Despite introducing transport allowance to retain the staff, it did not achieve much for some staff. By performing in-depth workforce data analytics, HR discovered that the issue lied with the recruitment process. Interviews were conducted at the headquarter, whereas the actual working place was at the other end of town.
The Impact
If you wish to download the complete set of data for this staff turnover case and follow through each analysis step you might want to get Data Analytics for Organisational Development.
This will enable you to perform your analysis with your own data up to building a powerful model for predicting organisational change indicators.
If you want us to lend a hand in your analysis, get on your Zoom and contact us. We will be there for you.
