People IntelligencePeople Intelligence is not a crystal ball
A signal can be valuable without claiming to know the future.
Björn Slegers ·
"Can you predict who will leave?" We get that question regularly. It is understandable. If an organisation knows where risk is emerging sooner, it can perhaps also act sooner. But that question contains an assumption we deliberately choose not to adopt. That human behaviour is predictable enough to attach individual certainty to it.
People are not stable datasets. An employee who today exhibits characteristics historically more common before a departure might make a completely different decision tomorrow. A different manager. An internal project. A conversation that takes place at the right moment. A change at home. The same data suddenly takes on a different meaning.
That is why we believe the distinction between a signal and a prediction is important. Not as a matter of semantic caution. As a design principle.
Seeing sooner is different from knowing the future
A system can make it clear that something is deviating. That multiple factors are changing at the same time. That a pattern is occurring more frequently in one team than elsewhere. That can be particularly relevant. But the message should not be: "This employee will leave." Much more useful is: "Something is happening here that deviates enough to merit a closer look."
That phrasing leaves room for human interpretation without minimising the value of the data. Because the alternative is not appealing either. We do not want to go back to organisations where risk only becomes visible after someone resigns, goes on long-term absence, or a conflict escalates. The opportunity lies somewhere between sailing blind and false certainty.
A percentage can project more certainty than actually exists
A risk score of 78 per cent looks impressively precise. But the precision of the number does not automatically say anything about the quality of the underlying context. What data was used? How recent is it? What is missing? Is the pattern relevant for this specific organisation? And is it truly a predictive model, or a heuristic that combines certain signals?
These questions are important. A good intelligence system should not just show an outcome. It should also make clear what a signal is based on and how cautiously someone should interpret it.
Reason enough to take a look
That might seem like a modest ambition. We think the opposite. When an organisation can see sooner that something is changing, a manager can have the right conversation sooner. HR can investigate sooner whether the same pattern is occurring more broadly. Management can see sooner where capacity or organisational design is beginning to cause friction.
You do not need a crystal ball for that. People Intelligence does not become less valuable when it admits that people are unpredictable.
Perhaps it actually becomes more trustworthy.