FEATURES / ATTRITION RISK

The resignation letter is the last thing that happens, not the first

By the time someone gives notice, the damage is done — knowledge walks out, teams scramble, projects stall. Gitrevio detects the behavioral shifts weeks or months earlier, giving you time to act.

How it works

# Attrition risk assessment — March 2026
HIGH RISK
Sarah Chen (Backend) Score: 0.78
Signals: review frequency -45% (8 weeks), commit
scope narrowing, cross-team interactions -60%,
1:1 skip rate increasing
Contributing factors (Shapley):
engagement decline 38%
reduced scope 27%
social withdrawal 22%
schedule pattern change 13%
Estimated departure window: 4-8 weeks
Impact if leaves: -38% review capacity, 3 orphaned services
MEDIUM RISK
James Park (Frontend) Score: 0.52
Signals: velocity stable but code review depth
declining, meeting attendance sporadic
Estimated departure window: 2-4 months

Attrition risk is not a single metric — it is a constellation of behavioral shifts detected across multiple dimensions simultaneously. No single signal is conclusive. A drop in commit frequency alone means nothing. The model looks at the pattern.

Each engineer's baseline is calibrated individually. A senior architect who reviews 20 PRs a week dropping to 12 is a very different signal than a junior developer whose review count fluctuates naturally.

The model updates continuously. Risk scores aren't a monthly snapshot — they adjust as new data arrives, so you see emerging patterns in real time.

What the model watches

Eight categories of behavioral signals, each composed of multiple underlying metrics. The model weighs them relative to each individual's established baseline.

Review withdrawal
Declining PR comments, shorter reviews, fewer review requests accepted. The clearest early signal — disengagement from the team's shared work.
Scope narrowing
Working on fewer repos, avoiding new features, maintenance-only commits. A shift from building to coasting.
Social disengagement
Fewer cross-team interactions, slower Slack response times, reduced participation in technical discussions.
Schedule shifts
Changed working hours, decreased overlap with team core hours, sporadic availability patterns.
Quality plateau
Code quality stable but not improving over time. A stagnation signal that often precedes departure by 2-3 months.
Meeting behavior
Declining meeting attendance, camera-off patterns, reduced participation in planning sessions.
Learning cessation
Stopped exploring new parts of the codebase. No new file touches, no new repo contributions, no new technology adoption.
Output reduction
Fewer commits per week, adjusted for sprint load and project phase. Raw velocity decline after controlling for external factors.
Mentorship withdrawal
Stopped answering questions, fewer pair-programming sessions, reduced onboarding involvement.

Shapley attribution: know why, not just what

A raw risk score of 0.78 is not actionable. You need to know what is driving the risk so you can address it. Shapley values decompose the score into contributing factors with mathematically rigorous attribution.

This transforms a vague sense of concern into a targeted conversation. Sarah's risk is driven by engagement decline and reduced scope — not compensation or work-life balance. A challenging new project might help more than a raise.

Different root causes demand different interventions. The Shapley breakdown tells you which lever to pull.

# Shapley factor decomposition
Sarah Chen — Risk score: 0.78
Factor breakdown:
engagement decline ████████████░░░░ 38%
reduced scope ████████░░░░░░░░ 27%
social withdrawal ███████░░░░░░░░░ 22%
schedule change ████░░░░░░░░░░░░ 13%
Suggested intervention:
→ New project assignment (addresses top 2 factors)
→ 1:1 conversation re: growth trajectory

Privacy and ethics

This is about organizational health, not surveillance. Attrition risk scoring exists to help managers have better conversations and make proactive decisions — not to monitor individuals.

Risk scores are visible only to direct managers and above. Individual activity patterns stay private — only aggregated risk signals surface. Engineers can see their own profile and understand what data informs their score.

No keystroke monitoring. No screenshot capture. No individual hour tracking. Gitrevio works from commit metadata, PR activity, and review patterns — artifacts the team already produces as part of normal work.

What you can do with early warning

Proactive 1:1 conversations
Address concerns before they calcify into a decision to leave
Project reassignment
Move someone to work that reignites their engagement
Mentorship pairing
Connect disengaging engineers with senior mentors
Career development
Create a growth path that makes staying the better option
Compensation review
When pay is the driver, act before a competing offer arrives
Team restructuring
Reorganize around people's strengths and interests

Detect attrition risk weeks before it becomes a resignation.

Get started free