LivePerson names a Human skill as its handoff switch

The conversation

LivePerson sells Conversational Cloud, a platform businesses use to build automated conversation flows with a path to a human agent. As retrieved on 16 September 2026, its own product page states the platform 'powers nearly 1 billion conversational interactions each month,' and the company's own developer documentation sets out, step by step, how a bot hands a conversation to a person.

What the documents show

The product page lists headline outcome figures — among them '90% automation containment rates' and a '50% decrease in agent attrition rates' — without stating the industries, accounts, or measurement window behind them; these are LivePerson's own promotional claims, not independently verified results. The page also notes the company 'has been acquired by SoundHound AI,' a change in ownership that sits alongside, but is not explained by, those performance statistics.

The system boundary

The developer documentation describes the actual mechanism: a bot dialog is set to trigger on patterns such as 'agent,' 'human,' or 'representative,' and the handoff itself runs through what LivePerson calls an 'Agent Transfer' interaction, tied to a named skill — the documentation's own example creates a skill called 'Human' — that must have at least one available agent assigned and marked 'Online.' If no skill ID is configured, the platform surfaces an explicit error rather than failing silently. Separately, the product page states that in at least some configurations, human agents 'review, edit, and approve AI responses before they are sent,' naming a second, distinct human-in-the-loop step.

Where it fails

The documentation explains how to configure a transfer but not what happens if no agent holding the relevant skill is online when a customer requests one — whether the conversation queues, shows a wait message, or times out is not described in the sourced material. The promotional statistics, meanwhile, carry no stated denominator or comparison group, unlike Fin's more precisely defined resolution metrics published elsewhere in this category.

• What happens to a transfer request when no agent holding the target skill is online?
• Which accounts or industries produced the 90% containment and 50% attrition figures LivePerson highlights?
• In which of its configurations does a human review AI responses before they reach a customer, and in which does the AI reply directly?

LivePerson's own documentation is unusually mechanical about the bot-to-agent handoff itself, even where its marketing statistics are not; the two kinds of document should be read differently.