Containment Rate is a Vanity Metric
High containment means the AI handled the call. It doesn't mean the customer's problem was solved. Stop optimizing for containment and start measuring resolution.
Containment rate is the industry's favorite metric: what percentage of calls were handled by the AI without a human? It's easy to measure, easy to inflate, and fundamentally misleading. A caller who hangs up in frustration after three minutes of unhelpful conversation counts as 'contained.' The problem wasn't solved. The customer is now angrier. But the dashboard looks great.
How containment gets gamed
Make the escalation path harder to reach — containment goes up. Add more conversation loops before offering a human — containment goes up. Set the escalation threshold so high that only the most persistent callers get through — containment goes up. None of these improve customer experience. All of them inflate the metric that vendors report.
Resolution is the metric that matters
Resolution rate with confirmation: did the customer's issue actually get solved, verified by post-call survey, absence of repeat contact within 48 hours, or successful action completion (appointment booked, refund processed, information delivered)? This is harder to measure. It requires connecting call outcomes to downstream systems. But it's the only metric that correlates with customer satisfaction and cost savings.
The right escalation rate is not zero
An agent with a 0% escalation rate isn't impressive — it's concerning. It means the agent is either handling interactions it shouldn't be or making it too hard for callers to reach a human. The ideal escalation rate is whatever percentage represents genuinely complex interactions that require human judgment. For most deployments, that's 15–30%. Design your canvas with clear, easy escalation paths. Your CSAT will thank you.
Ready to build?
See how Mazed's multimodal AI agents work for your use case.