The Problem
In many teams, people do not leave suddenly.
Before they quit, their behavior slowly changes:
- A very active person becomes quiet.
- A detailed person writes short, fast, low-quality messages.
- Someone who liked to argue and discuss stops sharing opinions.
- People reply slower, join fewer meetings, or do the minimum only.
Managers may see it too late - when the person already decided to leave.
The Idea
The startup would build AI agents that:
- Learn the normal behavior of each person over time.
- Look for changes in this behavior.
- Give early signals that someone may be disengaged or unhappy.
The system does not say:
“This person will quit.”
Instead, it says:
“This person’s behavior changed a lot compared to their usual pattern.
Maybe it is time to talk and ask how they feel.”
What the AI Looks At
The AI agents would look at safe, allowed data from tools like:
- Slack / Teams (message count, reply time, basic tone)
- Jira / Asana / GitHub (tasks, pull requests, activity)
- Calendar (meetings attended)
They would track things like:
- Engagement: number of messages, comments, reactions
- Speed: how fast someone replies
- Style: from long and detailed → to very short, cold, or distant
- Participation: joins fewer meetings, talks less, turns camera off
Important: the AI compares people only to their own history,
not to others. A quiet person is not a problem if they were always quiet.
How the System Works
We can imagine a set of small AI agents:
- A Communication Agent - watches message volume and tone.
- A Work Agent - watches tasks and code changes.
- A Baseline Agent - learns what “normal” looks like for each person.
- A Risk Agent - marks big changes (for example: “High change, last 3 weeks”).
- A Coach Agent - gives suggestions to managers, like:
- “Schedule a 1:1 talk.”
- “Ask if priorities are clear.”
- “Check if they feel overloaded.”
The main output is simple:
“In your team, 3 people show a big drop in engagement this month.
It may be good to talk with them.”
Who Is It For?
-
Companies and teams that want to reduce employee churn.
-
HR and People teams who need early signals of problems.
-
Later, the same idea can be used for customers (client churn):
- Less logins
- Fewer feature uses
- Different tone in support tickets
Ethics and Privacy
It can be dangerous if used like spying.
So the product must be very careful:
- Clear communication about what is tracked.
- No secret monitoring.
- Simple explanations:
- “Your engagement is down 40% vs last 3 months.”
- Not a tool to punish people, but a tool to start human conversations.
- If possible, employees also see their own data and trends,
so it helps them, not only the company.
Ethical use can be the main value of this startup.
Why It Matters
People rarely quit in one day.
They first disconnect emotionally, then mentally, then they leave.
If managers see early signals, they can:
- fix misunderstandings,
- change workload,
- give recognition,
- or accept that leaving is the best option
This startup idea is simple to explain:
“We use AI to notice when people quietly disconnect,
so you can talk to them before they walk away.”