Every resolved chat, scored by AI
An AI evaluator scores 100% of resolved chats into MySQL — per-agent quality, churn risk, and upsell-ready leads, instead of a 1% manual audit.
Three QA people could only spot-check a few hundred of 1,000+ chats a day, so managers couldn’t see whether customers were handled well, pitched the upsell, answered in time, or left frustrated — let alone the conversion picture. It was a grey area.
A chat is marked resolved — the trigger fires with the full transcript, the customer, and the agent.
A webhook hands the whole conversation to a Google Sheet that manages every chat.
Reads the chat against a fixed rubric and returns a structured JSON verdict.
Scores response time, whether the upsell was pitched, tone and clarity, customer sentiment, and whether anyone was left hanging.
Once the AI verdict exists, the script pushes the JSON into MySQL.
Fail-safe by design — each step only fires when the previous one actually produced data, so there are no empty or partial writes.
A reporting layer, built with a data analyst, turns it into per-agent scorecards and customer insight.
Who’s about to churn, who’s upsell-ready, and how each agent is actually performing.
Managers went from auditing ~1% of chats by hand to an AI scoring 100% of them — per-agent quality, customers to save before they churn, and upsell-ready leads, all from data that writes itself.
- Crisp
- Zapier
- Google Sheets
- AI evaluation
- Apps Script
- MySQL
- BI report
Client details anonymized — the work is real.
Want something like this for your stack?