Tune from real answers

The agent is live and talking to real customers. Now it's about tightening what it says and seeing whether it's actually working.

Open the Training tab regularly. Every time a human takes over a conversation to correct the agent, a pending proposal shows up here — approve the good ones and the fix applies everywhere at once.

Coming up: a full deep-dive on the Training tab is next in this path — see the "More from Academy" panel.

Expand what it can do

Add more guidelines as edge cases show up — they live under "Behavior" in the Build sidebar:

Guidelines sits under "Behavior," alongside Scenarios.

Open Guidelines Settings →

Guidelines — hard rules the agent should always follow, organized by category.

Then connect whatever integration fits your use case — a Shopify or WooCommerce store, a booking calendar, a CRM.

Check the numbers

Set up follow-ups for conversations that go quiet, and check Analytics for a first read on volume, response quality, and where customers drop off.

Follow-ups configured for at least one common drop-off point
Analytics checked — volume, response rate, and any recurring gaps

Was this helpful?