Traditional methods of monitoring how likely someone is to leave a business have included surveys, one-to-one meetings and reviews. However, in a busy environment with a high staff turnover, these methods do not appear to deliver the results. Imagine you are a retailer with 500 sites and thousands of staff with a high operational turnover; how can you better understand who is likely to leave and why so that you can take action to improve the situation? In this article, we look at how an AI can predict if someone is going to resign and how the data in your workforce management software can be used to train the AI tool.
How do I use AI and WFM data to improve employee retention?
We believe that the best approach is a three-step process:
- Use AI tools linked to your workforce management (WFM) system to identify the people most likely to resign.
- Use AI tools to identify the reasons that those people are likely to leave.
- Have a workforce manager address the reasons for dissatisfaction.
A systematic approach to these three steps will see continuous improvement in your HR processes, improving operational turnover and keeping it low.
What can an AI tool tell me if it is trained on our WFM data?
An AI can accurately forecast how many people will likely leave your organisation in the next 30, 60 or 90 days. They can also deliver insights per person on why they are likely to leave.
With access to this data, managers can intervene with potential leavers, address concerns, and improve retention rates.
What data sets in a workforce management platform make it ideal for estimating if someone is about to leave?
An AI is looking for patterns which correlate to an outcome. In order to find those patterns, the AI needs lots of data from which it can make them. A workforce management platform contains ideal data points for establishing how likely someone is to resign. These include:
- Start and leave dates: This is important because it allows the AI to look for patterns that occur before someone leaves. With lots of historical data, including the shifts worked in the run-up to someone leaving, patterns can be identified that occur regularly amongst leavers.
- Shifts Worked: Some WFM platforms have billions of shifts recorded, a big enough data set to train an AI.
- Pay rates: A WFM platform usually tracks salary per hour, but it could also see if that number was less than other people in the same role or if the level of overtime offered to an employee was dropping.
- Lateness: With in-built time and attendance software, a WFM system will have the data to see if someone is late more often than they used to be.
- Working time directive breaches: The EU Working Time Directive was built to ensure people get the correct rests and breaks, most WFM platforms track breaches, and this data is valuable for predicting resignations.
- Sickness: Has the employee shown a marked increase in sickness recently?
- Unsociable shifts worked: How many overnight or weekend shifts has the person worked recently?
- Long shifts worked: How many shifts over ten hours has the employee worked?
There are many more data points that we can use to develop our AI, but hopefully this is enough to convince you that your Workforce management platform holds all of the information you need.
How is the AI trained to understand if someone is about to leave?
In AI terms, generating a probability of someone resigning in the next 30, 60 or 90 days is called a “Classification problem”.
To truly understand how an AI can predict if someone is going to resign, we need to understand what classification is. A Classification is a form of supervised Machine Learning, where the computer is taught using data that is already “labelled”.
The labels highlight data points that are correlated to a high likelihood of someone leaving. These labels don’t always exist in the workforce management platform but can be created using “data engineering” techniques which automatically pre-calculate the labels.
What examples of labels are used to predict how likely someone is to resign in the next 30 days?
As mentioned earlier, workforce management software includes all of the shifts worked, but it will treat a Tuesday 9-5 shift no differently than a Saturday night 22:00 to Sunday at 06:00 shift. They are both lines of data in a database.
This is a good example of some data we might want to process using data engineering to create a label. We could create a flag for unsociable shifts. To do this, we would pre-calculate against a set of rules defining an unsociable shift and then flag the Saturday night shift as unsociable. We could similarly flag all shifts over 10 hours or all shifts that breach the working time directive in some way.
These are pretty simple flags to produce, but a more complicated one might be ‘shift equality”, which might even need an AI or algorithm of its own to rank each staff member on how fair the shifts that this person has been allocated are compared to other staff.
This process is fully automated using data engineering tools.
Once we have labelled all the data, how do we train the AI to look for patterns?
In classification, the model is fully trained against historical data using several different models. This is often done in the cloud with companies such as AWS or Google Cloud.
The outcome from each model is then evaluated on historical test data that the AI hasn’t been exposed to yet.
An algorithm is then used to evaluate the best model, which is then used to perform predictions on new unseen data. Again this process is fully automated.
What factors correlate highly with a high operational turnover?
In our experience, the top 5 factors that correlate with a high probability of someone resigning that can be derived from data in your workforce management solutions are:
- Working Time Directive Breaches
- Shift Equality
- Shifts over 10 hours
- Pay relative to others in the role
- Sickness and absence
All of these can be evaluated using an AI, and a percentage chance of the person likely to resign in the next 30, 60 or 90 days is produced. The AI can then attribute how much of the percentage probability is contributed by each of these factors.
What benefits can an AI tool using WFM data bring to retention?
Now that we have covered how an AI can predict if someone is going to resign using your WFM data let’s look at what benefits this process can bring to your retention and the wider organisation below.
- Improved operational turnover: By discussing issues with staff based on the insights given by an AI tool, managers can address problems before an employee leaves.
- Increased employee engagement: Focusing on why people leave will ensure you continuously improve your HR and operational processes to retain staff and will help all employees, leading to higher engagement.
- Recruiting planning: An accurate forecast of likely resignation numbers by role and by location will allow advanced notice of recruiting requirements and allow you to maintain the right number and mix of staff.
- Improved management: Benchmarking the data between departments will help identify any manager needing support and coaching, which will improve retention in the long term.