The Problem
Containment rate is one of the most-watched self-service metrics, but on its own it's shallow. It tells you the percentage of calls that stayed in self-service — not why the rest escalated, and crucially not whether they needed to. Some escalations are completely valid: the request genuinely needs a human. Others are misses — the caller wanted something self-service could have handled, but a gap in understanding, a missing option, or a clumsy flow pushed them to an agent anyway.
Those misses are exactly where containment can be improved, and they're invisible in the headline number. Finding them by hand means analysts listening to escalated calls one by one — so it rarely happens at any real scale.
The AI Use Case
Have AI review each self-service conversation that escalated and classify whether it was a genuine candidate for containment — then write that marker back into your analytics so it becomes a filterable, reportable dimension.
For every escalated interaction, the AI reads what happened in self-service and judges:
- Was this containable? Could self-service reasonably have resolved what the caller wanted?
- If yes, why didn't it? Intent not understood, no matching option, flow dead-end, caller confusion, or the caller just asked for a human.
- How confident is the judgement? So you can focus on the clear-cut cases first.
That verdict is attached to the contact record as a marker in analytics — "containment candidate: yes/no", plus a reason category — alongside everything else you already track.
What the Marker Unlocks
A true picture of containment
You can now separate valid escalations from missed ones, and report a "missed containment rate" — the calls that leaked to an agent unnecessarily. That's a far more actionable number than raw containment.
A prioritised fix list
Group the missed-containment markers by reason and by topic, and the biggest opportunities jump out: the intent that self-service keeps failing to catch, the common request with no self-service path, the flow where everyone drops out. You fix the causes with the most volume behind them first.
A feedback loop for improvement
Re-run the analysis after each change and watch the missed-containment markers for that topic fall. It's a direct, measurable signal for the kind of continuous self-service improvement that keeps a bot getting better over time.
"Candidate" is the important word. The marker flags calls worth reviewing, not a final verdict that a call definitely should have been contained. Treat it as a lens that points analysts and designers at the right conversations — the judgement about whether and how to change a flow still belongs to a human.
The Value
- Missed containment becomes visible — the gap between actual and achievable containment is measured, not guessed.
- Improvement is prioritised — effort goes where the volume and the opportunity are.
- Analysis at full scale — every escalated call reviewed, not a hand-picked sample.
- A measurable loop — changes can be proven to move the number.
- Lower cost — every recovered containment is an interaction that never needs an agent.
What You Need to Build It
- Self-service transcripts/events for escalated contacts, tied to the analytics record.
- An LLM review step prompted to judge containability and categorise the reason, with a confidence score.
- A write-back path to attach the marker as a contact attribute or analytics dimension.
- Reporting that slices missed containment by reason and topic.
- Clear criteria for what "containable" means for your business — this is worth agreeing up front, in the spirit of defining success criteria for AI self-service.
Explore More AI Use Cases
Part of a growing series of practical AI use cases for contact centres. Browse the full set on The AI Use Cases hub, and see the wider picture in 15 ways AI can transform your contact centre. Got one you'd like covered? Get in touch.