Containment rate: what it counts, and what it doesn’t
Containment rate is typically calculated as the percentage of conversations that did not escalate to a human agent. It is easy to measure and easy to report as a headline number, which is exactly why it shows up so often in vendor marketing. But containment rate has no opinion on outcome: a customer who abandons the conversation in frustration without escalating counts identically to a customer whose question was fully answered. A bot that simply fails to offer an escalation path can post an excellent containment rate while producing a worse customer experience than a bot that escalates readily.
Resolution rate: harder to measure, more honest
Resolution rate asks a different question: did the customer’s actual problem get solved, whether or not a human was involved? It requires a definition of "resolved" that goes beyond conversation-ended status — a follow-up signal, an explicit customer confirmation, a lack of repeat contact on the same issue within some window, or a post-interaction survey. It is genuinely harder to instrument than containment rate, which is part of why it is reported less often.
Why containment-only reporting can hide a bad experience
A team optimizing purely for containment rate is incentivized to make the bot harder to escape from, not more useful — fewer visible escalation options, more conversational loops before an agent handoff is offered. That produces a metric that looks like success internally while customer satisfaction and repeat-contact rates move the wrong direction. Any reported containment number should be read alongside a resolution or satisfaction metric, never in isolation.
How to think about measuring Answer Engine and Context Retrieval honestly
Voz360’s Answer Engine is a rule-based conversation system — decision-tree and FAQ bots that resolve repeatable customer questions across chat, voice IVR, and messaging channels. Because it is rule-based rather than generative, a well-scoped deployment should be measurable on resolution, not just containment: did the customer’s question match a defined tree branch and get a correct, complete answer, or did they need a human. Context Retrieval is a different kind of tool entirely — it is Voz360’s vector-embedding-based knowledge search that surfaces the most relevant approved article to an agent in real time; it assists a human rather than replacing one, so the honest metric there is agent time-to-answer and suggestion-acceptance rate, not containment at all. Applying a containment-rate lens to a tool meant to assist agents, not replace them, is a category error worth avoiding in any vendor’s reporting, including Voz360’s.
A practical measurement checklist
Whatever platform a team uses, a defensible AI-metrics report should include: containment rate defined explicitly (what counts as "contained"), a resolution or satisfaction metric measured independently of containment, escalation-path visibility (was the customer offered a human option, and when), and repeat-contact rate on the same issue within a defined window. A number without those definitions attached is not evidence of anything.
Can the vendor tell you — in one sentence — which of their AI capabilities are rule-based, which are generative, and which are still roadmap?