Review calls and improve your assistant
Your assistant gets better from real calls, not from guessing. Review the first calls, fix one thing at a time, and re-test — the same loop, every week.
Step 1 — Review the first 10 calls
- In the portal, open Call History, SMS History, and Voicemail for the last week — three separate pages. Call History has CALLS and AI TRANSCRIPTS tabs, with Date, From, To, Duration, and View Log columns. SMS History shows Date, From, To, Message, and Status for every text the tree and the agent sent. Voicemail is its own page, with an audio player for each message.
- Open “View Logs” on a call to see what actually happened: per-event rows (Date, From, To, Input, Digits, Detail) for the phone-tree path, and for AI calls the transcript as Caller / Assistant bubbles plus a “Tools used:” list. The tools list is your best diagnostic — it shows which tool the agent actually used. The AI Transcripts tab lists Voice and simulated sessions with turn counts and the provider and model used; its empty state reads “No AI transcripts for this organization yet.”
- For each call, note: what the caller wanted, whether the assistant understood it, and where the caller hesitated or repeated themselves.
- Sort the notes into three piles: greeting problems, knowledge problems (hours, pricing, area, booking), and handover problems (message or transfer).

Call History only reaches back about 25 days — the phone records come from Plivo, which allows up to about a one-month window — so review recent calls promptly rather than saving the review for month-end.
Step 2 — Fix the biggest pile first
- Greeting: shorten it, or move the most common request to option 1.
- Knowledge: correct your notes first — notes win over website content — then fix the website if the snapshot is pulling the wrong fact. The snapshot refreshes nightly, or use the “Refresh website knowledge” button in the AI panel after a website edit.
- Handover: check that messages are being returned, and that transfers reach a person, not another menu.
Step 3 — Turn a real call into a test
- Replay the exact caller question in Simulated Chatbot, with your fix in place. Pick the right mode: Voice Agent for an agent question, Phone Tree for a menu question (reply with menu choices, as a caller pressing keys would), and Phone Line for exactly what a caller gets.
- If it still fails, the fix is in the wrong place — the notes, the tool toggle, or the booking connection — not in the wording.
- Only after the dry run passes, let real callers hear the change.
Tip: Do not change five things at once. Change one, re-test, and you will know what worked.
Checklist
- First 10 calls reviewed
- One fix made, and re-tested in Simulated Chat
- Usage page checked — are you near your AI minute limit?
