Skip to content
Customer Success Intelligence

Project Overview

AI-Powered Customer Health Risk Detection

A customer-success automation system built with n8n, Google Sheets, Gemini agents, Slack and a Next.js web interface.

Risk signals
4
AI agents
4
n8n flow
1

The Problem

Churn risk is usually spotted too late

Customer success teams often identify churn risk too late because customer-health information is spread across operational systems.

This project centralizes those signals and turns them into actionable risk analysis.

Solution

Detect, reason, recommend, automate

  • Detect

    Evaluate customer activity, NPS, support status and renewal proximity.

  • Reason

    Use specialized AI agents to classify risk and interpret signals.

  • Recommend

    Generate an appropriate next action for the account manager.

  • Automate

    Support downstream workflows such as task creation and Slack alerts.

Architecture

How a request flows through the system

The browser never talks to n8n directly: a server-side API route proxies every analysis request.

  1. Web application

    Next.js Web App

    Customer is selected in the analyzer

    /api/analyze

    Server-side proxy route

  3. n8n workflow

    n8n Webhook

    Receives the customer ID

    Google Sheets

    Loads the customer record

    Deterministic Risk Calculation

    Business rules compute health metrics

    Risk Classifier Agent

    Risk Signal Extractor Agent

    Merge Agent Analysis

    Combines both agent outputs

    Action Reasoner Agent

    Decides the follow-up

    API Response

    Structured JSON result

  5. Web App

    Risk level, signals and recommended action are displayed

AI Agent Architecture

Specialized agents on top of deterministic rules

Each agent has a single, narrow responsibility. Their outputs are merged into one structured analysis.

Deterministic calculations remain part of the system. Customer-health metrics are computed by explicit business rules in n8n before any agent runs. The agents receive those results as input to classify, explain and decide, rather than recomputing them.

  • Risk Classifier Agent

    Determines risk level, category and confidence.

    • riskLevel
    • riskCategory
    • confidence

  • Risk Signal Extractor Agent

    Identifies factual risk indicators from customer data.

    • riskSignals

  • Action Reasoner Agent

    Decides whether action is needed and recommends the follow-up.

    • priority
    • recommendedAction

  • Slack Composer Agent

    Being reintegrated

    Formats approved decisions into clear Slack messages.

    • Slack message

Tech Stack

Built with production-grade tools

  • n8n

    Workflow orchestration

  • Next.js

    Web app and API proxy

  • TypeScript

    Typed application code

  • Tailwind CSS

    Interface styling

  • Google Sheets

    Customer data source

  • Google Gemini

    LLM behind the agents

  • Slack

    Team alerts

  • Docker

    Local n8n runtime

  • Vercel

    Web app hosting

Risk Signals

Four indicators from the customer sheet

Each signal maps to a column of the Google Sheets customer data.

  • Inactivity

    Time elapsed since the customer last logged in.

    last_login_date

  • NPS

    The latest Net Promoter Score reported by the customer.

    nps

  • Support Escalation

    Whether the account currently has an escalated support case.

    support_escalated

  • Renewal Window

    How close the contract is to its renewal date.

    renewal_date

Current Prototype Status

What works today, and what comes next

Completed6

  • Customer data retrieval
  • Deterministic risk detection
  • Multi-agent risk reasoning
  • Responsive web interface
  • API proxy
  • Vercel deployment

In Progress3

  • Production n8n public endpoint
  • Final task / Slack reintegration
  • Production hardening

Try the Customer Health Analyzer

Run a live analysis on one of the sample customers.

Open the Analyzer