The readiness checklist
A panel sits above the studio tabs whenever something is outstanding, listing what is unfinished. Each line is a button that takes you to the tab that fixes it. When nothing blocks the agent and nothing is worth flagging, the panel disappears — a checklist that stays on screen after it is finished is furniture. Two severities. A blocker stops the agent answering anybody, and turns the panel amber. A warning is worth fixing but the agent still works.
Drafts only and Test mode are deliberately not on this list. Both are legitimate ways to run an agent, and a checklist that nags about a choice you made on purpose is one people learn to ignore.
Every reason a message goes unanswered
These are checked in this order. Any one of them ends the reply, and most of them leave no trace in the inbox beyond the absence of a message.1
Something else already handled the message
On WhatsApp, a running flow that consumed the message stops the agent being asked. On every channel, a tap on an interactive button or list reply is answered by whatever sent it, and a message with no text at all — a photo on its own — never reaches the agent.
2
No agent covers this channel
No switched-on agent’s channels include this conversation’s channel. There is no fallback to “the only agent you have”. Fix in Routing.
3
The agent does not answer automatically
Answer inbound messages automatically is off — the agent drafts on request only. Fix on Behaviour.
4
The workspace is out of credits
On built-in AI with an empty wallet and no provider key of your own, the conversation is left for a person rather than the customer being told a business is out of AI credits. Fix by topping up.
5
A teammate owns the thread
A conversation assigned to a member is that member’s. Clear the owner in the thread header to let the agent reply.
6
AI replies are paused on this conversation
Set automatically by a handoff, and by the bot button in the thread header. The same button resumes it.
7
The conversation's reply budget is used up
Max replies per conversation — three by default — counted over a rolling 24-hour window that opens with the first reply after a quiet period. A customer who comes back the next day gets a full budget; a runaway loop inside one window is stopped at the cap.
8
Test mode excludes this customer
With test mode on, only the listed phone numbers are answered — and Instagram and website conversations have no phone number, so they are never answered while it is on. See Test mode.
9
An automation already answers this message
An active automation on a message trigger, scoped to this channel, whose keywords match this text, owns the reply. One message gets one reply, and the automation wins. Check Automations.
10
A handoff phrase matched
One of the Phrases that always hand off appeared in the message. The agent stands down without asking the model, and your handoff message is sent if you set one.
11
The model handed off, or came back empty
The agent decided it could not help safely. AI replies are paused on that conversation and it waits for your team.
12
The provider call failed
A bad key, a rate limit or a timeout. Your Fallback message is sent if you wrote one; with the field blank the agent stays silent rather than apologising in a voice you did not choose.
When it replies but seems to have forgotten a skill
The reply arrives, and reads as though the agent never knew it could book a meeting, log a row or send a basket link. That is a withheld tool: the agent was never shown it, so it could not mention it.
The Skills tab lists the tools each enabled skill can call, and the readiness panel names the two connection gaps it can see.
When the answer is wrong rather than missing
Open the test drawer and ask the same question. The evidence line under the reply names the documents the answer was built from and every tool that ran.- The right document was not retrieved — split a long document into one topic per document, or check its status is Ready rather than Keyword only or Failed.
- The document was retrieved and ignored — a ground rule or an instruction is contradicting it. Ground rules outrank everything below them, including your documents.
- It invented something — fill in What the business does, and add the fact as a document. The agent is told not to guess, but with nothing to go on it has only the customer’s question.
- It sounds wrong rather than being wrong — tone, reply length and Extra voice notes on Persona.