Demo page. The widget itself is everything between the markers TMSC WIDGET START … TMSC WIDGET END: one <div> + <style> + <script>, no dependencies (vanilla JS + SVG). A mini model of our Customer Support Copilot: incoming requests (email / chat / WhatsApp / in-app) → AI reads the conversation → detects intent and tone → finds knowledge-base articles → types a grounded reply and suggests actions → auto-resolve, a draft for approval, or escalation to a person. All data is synthetic (customers, orders and amounts are invented). On open it plays an ~19-second auto tour (request → analysis → knowledge base → the reply typed live → escalation → a new chat resolves itself); any click inside the console stops the tour, ⟲ Replay starts it again. Palette and fonts follow twinmind.au (Newsreader + Inter, blue #0392fd). prefers-reduced-motion is respected.
✦ Customer Support Copilot
AI COPILOT · watching the queue
Conversations handled▲ +12.4%
—
last 6 months · all channels
Auto-resolved—
—
deflected without a human
First response▼ from 3h 20m
—
median, first reply
CSAT▲ +0.6
—
post-conversation rating
Agent hours saved▲ this half
—
reallocated to complex cases
Copilot standing by…
Where every conversation goes · last 6 months
0 handed to an agent with a draft already written — 22.0%
0 escalated to a human specialist, fully summarised — 8.8%
Humans touch only the conversations that truly need them — about 3 in 10.
Sentiment lift · how people feel, arrival → resolved
+0 more customers leave the chat happier than they arrived
😠 Angry 14% → 3%
😟 Frustrated 27% → 8%
😐 Neutral 41% → 30%
🙂 Positive 18% → 59%
on arrivalafter the copilot
Where conversations come from
Live chat 44% Email 31% In-app 15% WhatsApp 10%