The numbers worth watching
| Number | Why | Bad sign |
|---|---|---|
| Qualification rate | The share of conversations that turn into a real enquiry. | Falling while volume rises — the agent is catching noise. |
| Handover rate | How often a person is needed. | Very high means the agent cannot answer; very low means it is not escalating. |
| Abandonment | Conversations that end mid-way. | Rising after a change to the brief. |
| Average length | Call minutes are the metered unit. | Creeping up with no change in outcomes. |
| Unresolved | Questions the agent could not answer. | The best list in the module — it is your content gap. |
The unresolved list deserves the emphasis. Every question the agent could not answer is a page your website does not have or a document the agent has not been given, and fixing it improves the agent and your site at the same time.
Reports
Last 8 weeks
How metering is recorded
Call time is written to a usage ledger as individual events rather than as a running total. Totals in Phony and in the shell are both computed from that ledger, so the two cannot drift apart — which is the failure a stored counter eventually produces.
When a call is transferred to a person, the segments belong to a single conversation and are metered once. It is worth knowing because the alternative — counting each leg — would double-bill exactly the calls that went best.
The same figures appear in Admin → AI usage alongside every other module's spend, and on the billing page as a line rather than a surprise.
What this does not do
The views are fixed. Export for anything else.
Usage is visible; nothing notifies you when it moves.
Reports stop at the outcome. What the lead became is a CRM question.
Metering is per workspace.
Questions
Can we export transcripts?
Yes, per conversation.
Is chat billed?
No.
Where do I see the bill?
Settings → Billing in the shell.