An employee flight risk is someone whose engagement signals suggest they are likely to leave in the coming months. Most organisations only find out who their flight risks were at the exit interview — which is why, for too long, businesses have been flying blind on talent retention and turnover forecasting.
This post is the story of how our data science team built the employee flight risk model behind wotter’s leaver predictions — what we tried, what we rejected, and why the result now flags risk with 86% accuracy.
What makes an employee a flight risk?
The classic signs an employee is a flight risk are familiar to every HR team: discretionary effort tails off, feedback goes quiet, scores on questions about future and progression start to slide. The problem is that humans spot these patterns inconsistently and late — and a spreadsheet-based employee flight risk assessment matrix is really just a checklist of hunches.
The cost of missing them is well documented. CIPD analysis puts average UK turnover at around 35%, and its turnover and retention factsheet is blunt about how expensive each avoidable departure is once you count recruitment, onboarding and lost output. Could you afford to lose and replace your highest performers? What about an entire critical department?
And the cost is cultural as well as financial. HR and talent teams need to understand why people leave to craft a culture of retention — the key to a happier, more productive workforce and lower turnover costs.
The data advantage: our proprietary signal
Predicting employee flight risk well is a data problem before it is a modelling problem. Thanks to wotter’s rich trove of proprietary survey data, our data science team was perfectly positioned to extract predictive signals of employee turnover. That data is powered by our Wott Question algorithm, which asks the right question, to the right person, at the right time — the engine behind our continuous surveys.
How our employee flight risk model works
In our early days, predicting employee turnover relied on brittle, rule-based heuristics. We reframed the problem: can we use our proprietary datasets to predict the probability of an employee leaving a company within the next six months?
Why we rejected the standard approach
The textbook answer is binary classification — sort everyone into “stay” or “leave”. It oversimplifies employee churn, because people are more complicated than that:
- An otherwise happy employee might still leave for a better offer elsewhere.
- An unhappy employee may stay put out of a dislike of change and uncertainty.
- Layoffs and restructuring happen, in which case employees do not have a choice at all.
Workplace culture is complicated. Different people value different things, so predicting who might leave requires an approach that captures human nuance while still giving HR and talent teams reliable numbers to act on.
Learning why people disengage — not memorising who left
To solve this, we use a state-of-the-art machine learning model: a system that continuously learns from a series of smaller patterns to make highly accurate future predictions. Crucially, we use specialised techniques to make sure it learns the underlying reasons people disengage rather than memorising old data. That stops the system developing tunnel vision, so its predictions hold up in the real, unpredictable world.
Calibrated confidence, not raw guesses
There is a catch with most standard AI models. A typical system might flag a group as an “80% flight risk”, but it cannot tell you how confident it is in that guess. Is it truly 80%, or a shot in the dark?
This is where wotter goes a step further. We apply an extra layer of mathematical calibration on top of the model, so instead of a raw, unverified score you get a rigorous confidence level with every prediction. That is what makes an employee flight risk score trustworthy enough to act on.
A model that moves with your culture
It doesn’t stop there. The dynamics of workforce culture change continuously, so we monitor our data for those shifts before updating the model — ensuring you consistently receive reliable predictions in the face of an ever-changing world. It is AI in HR done the careful way: predictive analytics in HR only earns trust if it keeps earning it.
What this means for your business
The results speak for themselves: wotter clients experience staff churn at a rate 40% below the national average. Our flight risk model flags risks with 86% accuracy, and Wottsby and the wotter platform turn those flags into proactive action.
The end result:
- 20% lower employee turnover costs annually, keeping the CFO happy.
- A more productive workforce, keeping the CEO happy.
- An award-worthy culture, keeping everyone happy.
If you want to see what this looks like on a real dashboard, our leaver analysis pairs employee flight risk scores with churn, stability and demographic breakdowns — the prediction and the “why” behind it, side by side.
What’s next?
This one is for our wotter clients: want to know where your workplace culture will be in a month? We thought so. Core24 Predictions are coming soon.
Not a client yet? Book a demo and see your first flight risk insights for yourself.
Best,
Jasper Wilson
Lead Data Scientist
Frequently asked questions
Common questions on employee flight risk and predicting turnover. Wotter is an employee engagement platform — our flight risk model spots who is at risk of leaving while there is still time to act. Anything else, drop us a line at hello@wotter.group.
