Metrics and terms
The number we decline to produce, and the reason
Containment is the figure quoted in every pitch in this category. It is not computed anywhere in this product, and that is a decision rather than an omission. The short reason is that the widget cannot observe the event the metric is about, and a number produced by guessing at an unobservable event is worse than no number.
not measured here
This product does not compute a containment rate. No tile, list or export contains one, and nothing in the interface uses the word. The nearest figure shown is labelled "Deflection" and hinted "answered, not refused", which is a claim about messages rather than about whether a person was involved.
What it means
Containment is the share of contacts that ended without a person becoming involved. The idea underneath it is reasonable: if an automated layer is doing real work, fewer contacts should need a human, and the share that did not is a rough proxy for that. The structural problem is that the metric records the absence of an event rather than the presence of a good outcome. A visitor who got a perfect answer and a visitor who gave up in disgust both produce the same absence, and from the reporting side the two are indistinguishable. Anything that makes reaching a person harder improves the figure. That property, well covered elsewhere on this site, is why we treat it as a concept to understand rather than a number to display.
How it is actually calculated
How it is calculated where it exists
Conversations that ended without human involvement, divided by all conversations, over a window. Every term in that sentence is a choice, and the choices move the figure more than the underlying performance does.
What counts as a conversation is the first choice: whether a visitor who opened a panel and typed nothing is in the denominator changes the result substantially. What counts as involvement is the second: a completed handover form, any offer of one, or any conversation where somebody asked for a person and did not get one are three different definitions producing three different numbers from identical traffic.
What would have to exist here to compute it
Three things this product does not have. A rule deciding when a conversation ended, since a visitor who closes a tab sends no signal and a conversation has no closure event. A definition of involvement that survives the fact that a handover here is a stored enquiry rather than an assignment to a queue. And a link between a widget conversation and whatever the same person did afterwards on email or the phone.
The third is the one that cannot be fixed by adding a field. A visitor who leaves the widget and telephones you has not been contained, and no widget can know that happened. Every containment figure in this category is silently missing that, including the ones quoted with two decimal places.
The closest honest thing on that page
The deflection figure, which is the share of assistant messages that were not refusals. It is narrower than containment on purpose and it is labelled to say so.
Do not present that figure as containment. It counts whether the assistant spoke, not whether a person was avoided, and the difference between those two is the entire subject of this page.
How the number gets moved without anything improving
Why the number rewards making the product worse
Remove the route to a person and containment reaches a hundred percent, because human involvement was not available to happen. Nobody does that in one step, which is what makes it dangerous: it happens through changes that each look like an improvement, such as offering the handover later, adding a qualifying question, or rewording it so it sounds like a last resort.
Abandonment counts as a success. A visitor who received a poor answer and closed the tab was contained, and in many implementations that population is the largest single contributor to a high score.
A confidently wrong answer scores identically to a correct one. The visitor believed it, left satisfied, and will be in touch later about something more expensive. This is the failure that makes the metric actively misleading rather than merely uninformative.
What to look at instead, or alongside
- The unanswered questions list, which names specific gaps rather than implying an outcome.
- The count of stored enquiries, read as workload, since those are handovers that actually happened.
- Total contacts across every channel over time, which is the question containment is usually a proxy for and cannot be gamed by hiding the exit.
- A hand checked sample of answers, because containment cannot tell a correct answer from a confident wrong one and a person can.
Questions
- A vendor quoted us a containment figure. What should we ask?
- Three things: what counts as a conversation, what counts as involvement, and at what point in the conversation a person is first offered. The third sets the ceiling on the number by itself. A vendor who answers all three specifically is describing a measurement; one who cannot is quoting a definition chosen to produce a large number.
- Could you add it as an optional tile?
- We could compute something and call it that, and it would be wrong in the ways described above regardless of how carefully it was labelled. A figure on a dashboard gets quoted without its caption, so the honest choice is not to produce one.
- Our management asks for it every quarter. What do we send?
- Send total contacts across all channels over time, which is the thing they actually want to know, plus the unanswered list as evidence of what is being worked on. If the word has to appear, write one sentence saying it is not measured and why, and expect that sentence to be more useful than the number would have been.
Keep reading
- Conversation volume: what a conversation count can tell youHow many people opened the widget and said something. A useful denominator, a poor headline, and a number that moves with your marketing.
- First response time has no meaning without a queueThis metric measures how long somebody waited for a human to pick up. There is no queue in a widget, and no first response time is computed.
- Escalation rate: what a handover is here, and why it is not a rateHandover enquiries are stored and listed. No rate is computed against them, and dividing by conversations by hand needs three caveats.
- Every metric and termWhat each one means, and which of them this product actually computes.
Try it on your own material
Upload a document or point it at your site, paste one line of HTML, then ask it something only your business could answer.