AI Can Now Tell You When Your Staff Are Going To Quit

AI Can Now Tell You When Your Staff Are Going To Quit
Image: iStock

With the benefit of hindsight, it’s often easy to see why people leave a job. Sometimes that’s by thinking back and recognising the signs of dissatisfaction after they leave, through exit interviews, or by examining performance reviews and seeing long-term changes that you’ve missed in the day-to-day hubbub.

But what if you could predict when people are planning to move on? That could help you change things to retain the person or perhaps help you plan for an orderly exit. Culture Amp has been looking at this and has added some new predictive analytics capability to its platform and boasts it can predict who is going to leave a job and when with 80% accuracy.

The cost of hiring is high. Aside from fees for job ads and recruitment agencies, there’s the productivity hit you take when a new person is learning the ropes, training and equipment costs and contractor support if there’s a gap between when a person leaves and they’re replaced.

According to Toby Roger from Culture Amp – a company that focusses on employee feedback – some estimates put the cost of employee turnover at over 100% of the incumbent’s salary. Being able to predict when an employee is going to leave can help reduce that cost.

Culture Amp says in its “Using predictive analytics to forecast employee turnover” report that people leave their jobs over things like not feeling a sense of belonging and company leadership with many deciding they’ll bail from a new job within six weeks of starting with about one in ten churning out of a role within the first six months.

Using the data Culture Amp has collected through its employee feedback platform, it has built an analytics model using insights from abut 2.5 million people. Some of the signals it uses are easy to understand. For example, when the answer to whether an employee sees themselves still working with you in two years’ time suggests they’re on the move, that’s easy to understand.

But the new analytics tools also looks across employee surveys longitudinally to find more subtle changes in employee responses.

The benefits of this are profound. As well as helping you better predict, and potentially prevent or slow down, employee churn, it can help wth strategic and tactical planning.

For example, if the Head of Sales is setting an ambitious strategy for the year, but you have the foresight to see 30% of the sales team are going to leave; suddenly the conversation changes. People problems can disrupt your organisation’s overall strategy. If you can predict people problems, you can circumvent business problems.

During my time in corporate life, there were many times I regretted when a colleague or team member moved on. There were plenty of times when the exit interview revealed that something could have been done to prevent that person from moving on. Often, when we’re on the treadmill, trying to keep up with deadlines and other pressures, we miss the signs that someone is unhappy.

And while many of us have probably filled in employee satisfaction surveys over the years, it’s not that common for companies to use that data for more than a short time as they take a survey, produce a nice report, make a few small changes and then resume business as usual.

You can access the full report here.


  • The AI hasn’t met me yet then. Given Employee surveys have the tendency to be identifiable surveys, those questions my future with the company are always answered to present the position of a lifer.

    The rest gets answered truthfully. They do not need to know what I may or may not be thinking about my future with the company.

    • I’m sure the AI can detect survey answer incongruity.

      * Plum spots are taken and you need someone to die to progress
      * You work your arse off for no gain or recognition
      * Expected to travel to third world hell-holes
      * Never going to leave! Five year plan has you firmly in this company.


  • It is not much different than spying on employees phone calls and emails. It could also be considered an invasion of privacy, having a computer analyse one’s behaviour to determine the probability of leaving.

    So, what happens then? Does the employer fire you if the AI says there is a probability of over x%? Does the employer not promote you? Either way, you’re screwed if the AI gets it wrong.

  • To be clear, we never reveal individuals who are likely to leave, only groups of employees (e.g. Sales team, 45-55 year olds, etc.) that meet a certain size criteria in order to protect confidentiality.

    Here at Culture Amp, we highly value the trust that is central to quality feedback, and our aim is to use feedback to help organisations improve culture. We respect the privacy and security of every individual in our platform and at no point do we allow for an individual’s feedback to be accessible to anyone in the company without their consent. Like I mentioned, we only show results at an aggregate level of sufficient size to prevent individual identification.

    Additionally, we do not use invasive behavioural or passive monitoring. Instead, we use what employees willingly tell their company in various surveys, and the demographic and organisational information associated with those response, in aggregated groups.

    Our predictive analytics are not designed to identify individuals but to allow companies to identify the things that make groups of people leave so they can improve things and reduce turnover in the future. It’s our way of using machine learning to make work a little bit better.

    Tim Hancock, Lead Data Scientist, Culture Amp

Show more comments

Log in to comment on this story!