Attrition Risk Analyst

Finds your at-risk roles and teams from HR data through Postgres MCP — tenure dips, comp drift, and workload signals — and delivers a de-identified risk digest to people ops.

A core-runtime agent for people analytics, built on a strict de-identification rule: the digest names teams, roles, and patterns — never individuals, and it never writes a person's name into the output. It connects read-only to your HR database via the official postgres-mcp server, discovers how you store headcount, tenure, compensation, reviews, and workload, and computes the group-level signals that precede attrition: teams whose voluntary exit rate is trending up, roles where comp has drifted below market band midpoints, teams with review cycles that lapsed, and workload concentration (a few people carrying disproportionate hours/allocations). The digest goes to Slack for the people-ops review. Runs as an agentic loop and adapts to your schema.

Quick Start

ad init --template attrition-risk-analyst my-attrition-risk-analyst
cd my-attrition-risk-analyst
npm install
npm run dev

Tools

• analyzeAttritionRisk: Compute group-level attrition risk signals and deliver the de-identified digest
• notifySlack: Deliver a notification to a Slack channel

Required Secrets

• DATABASE_URI — Postgres connection string to your HR database (user must have SELECT only; prefer a view that excludes direct identifiers) (Get one)
• SLACK_BOT_TOKEN — Slack bot token with chat:write scope (Get one)

MCP Servers

• postgres (stdio)

Try It

periodDays: 365

channel: #people-ops

Architecture

• Topology: core-only
• Archetype: autonomous
• Model: openai/gpt-4o-mini
• Category: people-ops
• Industry: saas

Deployment

ad deploy

Generated by AgentDeploy Blueprint Framework