Nobody resigns on the day they resign. The decision gets made quietly, months earlier, and then it waits for a trigger: a recruiter's message, a bad meeting, a bonus that lands light. By the time the letter hits your desk, you're negotiating against a decision that was finalized half a year ago. You will lose that negotiation almost every time.
The survey paradox
Engagement surveys have a built-in flaw: the people you most need to hear from are the least likely to tell you the truth. Someone who has already decided to leave has nothing to gain from honesty and a reference to lose. So they click through, give you sevens, and keep interviewing. Gallup found that 42% of voluntary turnover is preventable, and that 45% of leavers say nobody even talked to them about how their job was going in their last three months. The information gap isn't on the employee's side. It's on ours.
Behavior leaks the truth
Feelings hide. Behavior leaks. Six months before a resignation, the patterns start to shift, and they're visible to anyone who looks. Their collaboration network shrinks; they work with fewer people. Discretionary contributions stop first: mentoring, documentation, ideas nobody asked for. Their planning horizon shortens; they stop signing up for things that finish two quarters out. Individually, each of these has an innocent explanation. Together, and trending, they don't.
This is exactly the kind of problem agents are built for. A human manager with eight direct reports and a day job cannot track pattern drift across months. An agent can watch continuously, compare against each person's own baseline, and raise a hand when the trend turns. Me against me, not me against some company average. The same standard I hold for improvement works for detection.
Prediction without trust is surveillance
But understand this: the moment your people feel watched, you've traded a retention problem for a culture problem, and the second one is worse. There's a right way to do this. Aggregate before you individualize. Propose before you act: the agent should suggest a check-in to a manager, not score a human being. Every action needs a human confirmation and an audit trail. And be transparent that the goal is to keep good people, not to catch bad ones. This is also where regulation is heading, from NYC's bias audit law to Colorado's AI Act. Build the guardrails in now or retrofit them under subpoena later.
What to do with six months
Early warning only matters if you use the time. Gallup's data shows a manager having one meaningful conversation a week makes employees four times more likely to be engaged. That's the intervention. Not a counteroffer, a conversation. Six months is enough to fix a role, move someone toward work they love, or fix the manager problem you've been avoiding. The exit interview is where the truth goes to die. Move it six months earlier and it becomes the cheapest retention program you'll ever run.
HappyTeams reads these signals continuously, and proposes the action before it's too late.
Request a Founding Spot →