AI-Powered Customer Scorecard
AirGap Labs
IT Infrastructure & Security | Fortinet Engage Preferred Partner
TECHNICAL CASE STUDY
Internal CRM & Automation Suite
Stack: Shopify Liquid • n8n Cloud • Supabase • Chart.js • Harvest API
Role: Principal, AirGap Labs • Irvine, CA
CASE STUDY
AI-Powered Customer Scorecard
Automated relationship health scoring with live drill-down analytics
Industry |
IT Infrastructure & Managed Security Services |
Platform |
Shopify Liquid + n8n AI Agent + Supabase + Chart.js |
Timeline |
Built from scratch; iterated with AI scoring layer added later |
Audience |
Sales team and account managers |
The Problem
AirGap's account managers had no systematic way to identify which customers were thriving and which were at risk. Customer health was assessed ad hoc in weekly meetings — relying on memory, gut feel, and whoever spoke up. There was no scoring model, no trend visibility, and no way to prioritize outreach across a growing book of business.
The team needed a way to surface risk early, identify expansion opportunities, and walk into customer conversations with data instead of assumptions.
The Solution
A Customer Scorecard dashboard that combines structured CRM data with an AI-powered scoring agent. Each customer receives a composite health score across four categories, displayed in a visual dashboard with drill-down tables and an AI action plan generator.
Scoring Categories
- Financial Health — invoice aging, payment history, open balance trends
- Engagement — ticket volume, response time, meeting frequency
- Project & Contract Activity — active SOWs, renewal proximity, project completion rate
- Relationship Depth — stakeholder coverage, escalation history, NPS signals
Key Features Built
- Stacked bar chart grouped by health tier (Healthy / At Risk / Critical) with per-category breakdown
- Drill-down table that filters to customers in any selected tier when a chart bar is clicked
- AI scoring agent in n8n: reads structured customer data, reasons across the four categories, and returns a scored JSON object with narrative justification
- Action Plan modal: one click sends a customer record to the AI agent and returns a prioritized action plan for the account manager
- Score history tracking — Supabase stores each scoring run so trend lines are available over time
- Role-gated to admin and sales rep metafield values
|
4 Scoring dimensions |
n8n + LLM AI agent |
Automated At-risk customers surfaced |
Yes Action plans on demand |
Technical Highlights
The AI scoring layer was the most architecturally interesting piece. Rather than a rules engine with fixed thresholds, the n8n agent receives a structured data payload and uses an LLM to reason across all four scoring categories simultaneously — weighting factors contextually rather than mechanically. The agent returns a normalized score plus a plain-English rationale that appears in the dashboard alongside the number.
The scoring pipeline handles varied n8n response wrapper shapes via a recursive extractor, which proved important as the workflow evolved through several iterations. The chart and table components share the same tier palette constant, keeping visual language consistent even as the scoring model was tuned.
Outcome
Account managers now open the scorecard at the start of each week to identify their highest-risk accounts. The AI action plan feature reduced the time to prepare for a difficult customer conversation from 30+ minutes of manual research to under two minutes. The health tier chart gives leadership an instant portfolio-level view that was previously impossible without a dedicated CRM system.
AirGap Labs • Irvine, CA • airgaplabs.com