← All cases

AI Customer Support

AI-powered WhatsApp support, measured in seconds

A school's business WhatsApp began responding with a median time of 96 seconds, complete with a monitoring dashboard, imported history, and two response modes — without changing the number or the phone.

96 s
median response time (before: hours)
4 tabs
dashboard: conversations, queue, metrics, pairing
2 modes
responds on its own or suggests replies to the human agent
Client
Online English school
Period
2026, in production
Stack
WhatsApp (WAHA/GOWS) · Cloudflare Worker + D1 · Gemini · TypeScript

The problem

A lead who writes on WhatsApp and waits hours gives up. The school had human support during business hours and a queue that no one measured — so no one knew how much they were losing.

What was built

A connector between the school’s business number and a Worker that logs every message, measures the time to response, and triggers a language model in two modes: responds on its own for routine questions, or suggests the reply for the agent to approve. The dashboard shows conversations, the queue, metrics, and the pairing QR code — all on one page, with per-teacher login.

One detail that matters: the previous history was imported (backfill), so the metrics were born with a past, not from scratch.

What was measured

  • Median of 96 seconds between the student’s message and the reply. The average was misleading (17 hours, skewed by abandoned conversations) — which is why the dashboard shows the median, paired with a 24 h ceiling.
  • Running in an isolated container, separate from the demo environment, with automatic session recovery.

What the client didn’t have to do

Change numbers, buy a new device, or migrate to a support platform. The phone number stays the same.