Containment is not resolution, and optimising the first destroys the second

Three metrics get used interchangeably in this category and mean different things.

Deflection: the contact never reached a queue. Containment: the interaction ended without escalation to a human. Resolution: the customer's problem was actually solved.

A system can score very well on the first two while failing the third, and the failure is invisible in the dashboard that reports them.

Containment is measured at the end of an interaction. Resolution can only be measured afterwards — did they come back, did the underlying thing get fixed, did they complete what they were trying to do.

Anything measured at the end of an interaction can be improved by making the interaction harder to escape. That is not a hypothetical failure mode; it is the predictable consequence of putting a containment target on a team.

Containment counts an ending. Resolution counts an outcome. A customer who abandons the chat in frustration and phones instead is contained by every definition the dashboard uses, and has cost you two contacts rather than one.

This is not a flaw in a particular vendor's measurement. It is what happens when the easiest thing to instrument — did this session reach a human — becomes the thing that gets reported. Nobody chose it; it accumulated.

The correction is not to distrust containment. It is to refuse to report it alone.

Reported industry figures vary widely, which itself tells you the definitions are not stable. Vendor and analyst write-ups in 2026 commonly place well-configured deployments in the 70–80% containment range and average ones considerably lower, while separately reporting true resolution rates far below containment — often by tens of percentage points.

Treat any specific number with caution; the durable finding is the gap, and that it is large.

Resolution rate: the share of contained interactions where the customer did not return about the same issue within a defined window.

Seven days is a reasonable default. It requires linking contacts to a customer identity over time — which brings you back to the identity problem, as almost everything in this field eventually does.

Make escalation easy and visible. An agent that offers a human early gets better resolution and, counter-intuitively, often better satisfaction than one that resists.

Instrument the handover. The most damaging pattern is not the bot that fails; it is the bot that fails and then hands over without context, so the customer repeats everything. That converts a mildly annoying interaction into a memorable one.

Report the pair, never the single number. Containment alone is not reportable. Containment with resolution is a real measurement, and the ratio between them is a better indicator of quality than either.

If the objective is genuinely cost reduction, say so and measure cost per resolved contact — which keeps the incentive attached to resolution. Cost per contained contact optimises for a number that can rise while the business gets worse.

If the goal is cost reduction, say so, and measure cost per resolved issue rather than containment. That is a legitimate objective and it survives scrutiny.

The failure mode is claiming a service improvement while measuring a cost one. It works for two quarters, and then somebody compares the containment chart with the complaint volume and the credibility of every other number in the deck goes with it.

Deflection, containment and resolution get used interchangeably in vendor material. They are not synonyms, and the gap between the first and the last is where the money is.

Report the pair. Containment without resolution is not a measurement, it is a hope. Define the return window, count issues rather than sessions, look across channels, and design the agent to escalate early — accepting that this lowers the number you used to report and raises the one that matters.