Why a call failed

The failure label on every finished call, what each one means, and who acts on it.

Every finished call in your workspace can carry one failure label: why the call went wrong, and which side caused it. The label shows as a chip on the Calls page, in full in the call's drawer, as a breakdown on a campaign, and for the last 30 days on each AI agent's overview.

A call with no label chip had no failure observed. That is not the same as proof that the call went well: some failures can only be detected when the call's full evidence (its transcript and its timing telemetry) has arrived.

The label is not the reason the call ended. Every call also reports why it ended (the caller hung up, the AI agent finished, a safety rule ended it); see Post-call results. The failure label reads that reason, the line's own result and the call's evidence, and says whether anything went wrong.

Three places a call can fail

Layer	What happened	Who acts
Telephony	The call never became a conversation.	You: check the contact list, the calling times or the number.
Audio and media	A conversation started and Skysay's voice pipeline broke.	Skysay: the failure is ours.
AI agent behaviour	The pipeline worked; the AI agent or one of your own systems failed.	You: check the AI agent's prompt, your webhook or tool, or the transfer number.

When a call shows more than one problem, the label names the root cause: a call that never connected is a telephony failure whatever else happened, and a broken pipeline explains anything the AI agent did wrong on that call. The other problems observed stay visible in the call's drawer.

The labels

Telephony

Label	Meaning
Inbound call rejected	Skysay's admission refused the incoming call before it was answered, for example because the workspace could not take it.
Invalid number	The number does not exist or cannot be dialled.
Rejected by the carrier	The carrier refused the call.
Line busy	The line was busy.
No answer	Nobody answered before the ring timed out.
Answered by a machine	The line was answered, but only by a machine, music or tones, never a person.
Telephony technical failure	The call failed before it could connect, for a technical reason on the line.

Audio and media

Label	Meaning
Media session lost	The audio session went away during the call.
Voice engine lost	The conversation engine closed underneath the call.
Reply stalled	The AI agent stopped replying while the caller was talking to it, and the call was ended.
Caller speech missed	The call was ended for no caller speech although the caller did speak.
AI agent went silent	The caller spoke and the AI agent never replied.
Speech recognition stalled	Recognising what the caller said failed or timed out.
Language model failed	The language model failed or timed out on a reply.
Speech synthesis failed	Turning the reply into speech failed or timed out.
Goodbye cut off	The AI agent's closing line was cut before it finished playing.

AI agent behaviour

Label	Meaning
Transfer went wrong	A transfer to a person did not land: it failed, was rejected, nobody answered, the line was busy, or its outcome never came back.
Text message not sent	An SMS the AI agent tried to send during the call failed.
Tool call failed	A tool the AI agent called (your webhook or integration) failed or timed out.

Limited evidence

Some labels rest on the call's transcript and timing telemetry: Caller speech missed, AI agent went silent, Speech recognition stalled, Language model failed, Speech synthesis failed, Goodbye cut off, Tool call failed and Text message not sent. Skysay only uses them when that evidence is complete, because on partial evidence a missing turn may be a gap in the evidence rather than something that happened.

When a call connected but its evidence is incomplete, the call's drawer says so: Limited evidence: only carrier and session problems can be detected for this call. Labels are decided again as evidence arrives, so a call can gain a more specific label a few moments after it ends.

Finding failed calls

On the Calls page, the failure filter narrows the list to one layer or one label; the count and "Load more" follow the filter. A campaign's detail page breaks its calls down by layer and label, and each AI agent's overview shows the same breakdown for its calls from the last 30 days. When there is nothing to show, the breakdown says No failures detected rather than drawing zeroes.