Containment is not resolution
Most AI support tools report containment. It is the easiest number to move and the least useful one to trust. Here is the measurement stack operators should ask for instead.
Whizdom AI · Product team · 22 July 2026 · 6 min read
Containment tells you a conversation ended without a human. It does not tell you whether the player got what they came for. A player who gives up is contained. A player who is told to email support is contained. The number goes up either way, which is precisely why it is the number most vendors lead with.
The two numbers that actually differ
In live iGaming deployments measured over 30 days, Whizdom AI Agent recorded 57.2% containment across 89,000+ conversations. Over the same window, the AI-resolved outcome rate — outcomes classified as resolved under Whizdom's outcome methodology — was 19.4%. In a second deployment of 18,680 conversations, containment reached 72.6% and the AI-resolved outcome rate 35.2%, with a 27.4% human escalation rate.
The gap between those two figures is the honest part of the report. Resolution is a strict subset of containment. If a vendor cannot show you both, they are showing you the flattering one.
What to ask a vendor for
- Containment rate, defined precisely: conversations completed without human escalation.
- Resolved outcome rate, with the classification methodology written down.
- Human escalation rate, broken down by journey.
- Conversation pace per turn — a slow answer is a queue by another name.
- FTE-equivalent capacity freed, with the assumptions behind the estimate stated.
Why the distinction is commercial, not academic
In iGaming, the conversations sitting in a queue are the highest-stakes ones a player will ever have with you: a payment blocked, a withdrawal delayed, an account locked. A contained-but-unresolved conversation in that context is not a saved ticket. It is a retention event going the wrong way.
Every conversation sitting in a human queue is a player waiting at their most frustrated moment.
That is the reason Whizdom AI Agent is built around a performance loop rather than a reporting dashboard: Live Rescue surfaces stuck or failed conversations for real-time intervention, Agent Health scoring tracks containment, resolution, quality and efficiency against configurable targets, and playbook updates close the gap over time.