AI Customer Support & Escalation System
A message-triage system for an e-commerce brand that answers only what it can prove from a live spreadsheet — everything else, and anything from a distressed customer, goes straight to a human with the order already looked up.
Build Status
Built and live-tested
3
Channels unified into one pipeline
34
Scripted messages run against the real system, real Gemini, real spreadsheet — zero misfires
0
Tolerance for a wrong automated answer, by design
Inside the Build
The complete workflow canvas.
Scenario Replay
See how the logic responds.
Choose a real-world situation to follow the path from trigger to outcome.
Trigger
“Where is order #1048?”
Capture message
Load order data
Run safety gates
Send verified reply
Outcome
The customer gets a fast answer backed by a live spreadsheet row.
Operational Change
What changes after automation.
Routine replies
Repeated manually in a shared inbox
Answered in seconds from verified data
Urgent messages
Wait behind routine questions
Escalated before intent handling
Agent context
Order details searched by hand
Customer and order context preloaded
The Challenge
A growing online seller had one shared inbox for Messenger, Instagram, and web-form messages, and staff were answering the same routine questions dozens of times a day while urgent complaints sometimes waited behind them in the queue.
The System
We built an n8n decision engine that checks six ordered gates — starting with whether the customer sounds distressed, before it even looks at what they're asking — and only lets the system answer when it can point to a specific row in a live spreadsheet. Anything it can't prove, or anything urgent, goes to a human inbox with the customer's order already pulled up.
Guardrails
Distress and urgency are checked before automation answers.
Replies require a matching source-of-truth record.
Unsupported requests always fall back to a human.
