What is the difference between containment and resolution?
Containment vs resolution is the distinction between a conversation that ended without a human agent and a customer issue that was actually fixed. Containment counts where the conversation stopped. Resolution counts what the customer walked away with. One is a routing fact; the other is an outcome.
Every contained conversation is one of two things: a fixed issue or an abandoned customer. Support teams that report only the first number cannot tell those apart, because both produce the same row in the log: no transfer, no agent, conversation closed.
How containment and resolution are counted
Both numbers come out of the same conversation log, read at different points. Containment is measured at the exit: the share of sessions that closed without a transfer, an escalation, or a queued callback. Chatbot containment rate is the common form of it, and its denominator is every session the automated layer opened. Nothing in that calculation looks at whether the customer’s problem changed state.
Resolution is measured after the exit, and it needs a second signal. A resolution rate counts conversations where the underlying issue stayed closed with no follow-up contact inside a defined window, usually seven or fourteen days. That window is what makes resolution expensive to compute: you cannot score a conversation the moment it ends.
Four signals feed the second number: reopen and repeat-contact detection, post-contact survey responses, downstream ticket creation, and manual review of a sampled slice. First contact resolution applies the same test to human channels, counting an issue closed only when the customer does not come back.
What counts and what does not
Contained: A session that closes inside the automated layer with no transfer, no agent assignment, and no queued callback, whatever the customer felt about it.
Transferred: Any session where a person took over, including silent handoffs, warm transfers, and callbacks the automation scheduled on the customer’s behalf.
Resolved: An issue that survives the repeat-contact window with no reopen, no second channel, and no ticket created downstream by another team.
Contained and unresolved: A session the automation closed while the customer’s problem persisted, usually surfacing days later as a fresh contact.
Resolved after handoff: A person closed the issue, so the interaction counts against containment while still counting fully toward resolution.
Containment vs resolution vs deflection vs escalation rate
Support dashboards stack these four numbers next to each other, and teams read them as if they measured the same event at different zoom levels. Containment records where a conversation ended. Deflection records a contact that never opened a ticket at all. Escalation rate records the share of conversations that reached a person. Resolution records whether the customer’s issue actually closed. Containment and resolution are the pair worth holding together, because the distance between them is where automation quietly fails.
What it counts | What it misses | Typical benchmark | |
|---|---|---|---|
Containment | Sessions closed without a human | Whether the issue was fixed | Your own prior period, split by intent |
Resolution | Issues that stayed closed | Effort and cost spent getting there | Your own baseline, segmented by contact reason |
Deflection | Contacts that never opened a ticket | Whether the customer gave up | Self-service traffic weighed against ticket volume |
Escalation rate | Share handed to a person | Whether the handoff was correct | Tracked per intent, quarter over quarter |
If you are reporting to a finance team, resolution is the number that survives scrutiny, because it attaches to work that ended. Track containment beside it as the diagnostic that shows which intents the automation is only pretending to handle.
Why the gap between containment and resolution matters for customer experience
When only containment is reported, the incentive runs one way: keep the conversation inside the bot. Teams tighten handoff rules, bury the path to a person, and watch the number climb while repeat contacts climb with it. The customer experiences that as a loop, and the second contact usually arrives angrier and more expensive than the first would have been.
The gap also changes commercial conversations. Buyers stopped accepting contained sessions as proof of value, which is why resolution-based pricing exists and why the split between deflection rate and resolution rate gets argued over in procurement.
The tradeoff is real. Raising the resolution bar means handing more conversations to people, which pushes containment down and cost per contact up in the short term. Teams that refuse that trade buy a good containment number with silent churn.
How is the containment-resolution gap calculated?
Compare the number against your own history before comparing it against anyone else’s. Take containment and resolution for the same period, split by contact reason and channel, and subtract: the difference is the share of conversations that ended without a human and without a fix. Segment before you subtract, because a mix shift toward simple order-status questions lifts both numbers without any gain in capability.
Benchmarks do exist for the tasks sitting underneath these metrics, and they measure components. Voice quality is one: the subjective rating method behind Mean Opinion Score is defined in ITU-T Recommendation P.800, and it grades how a call sounded, which says nothing about whether the caller’s problem was solved. Scores from adjacent evaluations travel badly for the same reason: they judge one component, while containment judges a whole conversation.
Standards bodies publish no target figure a support team is expected to hit for either metric, so the honest comparison stays internal: your own segmented baseline, read quarter over quarter.
How AI agents change containment and resolution
Rule-based bots could only contain by exhausting the customer: no path forward, no person available, conversation over. AI agents that call tools change the mechanism, because the agent can look up the order, issue the refund, and write the result back to the CRM, so containment and resolution begin converging on the same conversations.
That convergence is uneven. Agents contain best on intents where the action is exposed through an API and the policy is unambiguous, and the residual gap concentrates in cases needing judgment, an exception, or a system the agent has no write access to. Reading containment alone hides that concentration, since one aggregate percentage averages a solved intent together with a stalled one.
The reporting consequence follows: teams increasingly measure containment and resolution quality as a single pipeline, scoring outcomes per intent so the automation roadmap follows the intents where the gap is widest.
What to look for when reporting containment and resolution
Judge a reporting setup on whether it can defend its own numbers.
Coverage comes first: whether the count spans voice, chat, and email, or only the channel the automation launched on. Integration surface decides whether resolution can be verified at all, since the reopen signal usually lives in the ticketing system while the refund confirmation lives in billing. Governance is the ownership question: one team should own the outcome definitions, and every change to them should be dated so a quarter-over-quarter chart stays readable.
Two frameworks bind this work directly. SOC 2 Type II matters because outcome scoring copies transcripts into an analytics store sitting outside the helpdesk’s access controls. GDPR bites harder: retention limits can delete conversations before a fourteen-day reopen window closes, so the scoring job has to run on a shorter clock than the deletion policy. Sampling for customer service quality assurance hits the same deadline.
Containment, resolution, and service level commitments
A service level agreement usually commits to response and resolution times, which quietly makes containment a contractual matter: a conversation held inside automation is still consuming the clock, and a customer who gives up and re-contacts starts a fresh one while the original issue ages unmeasured.
That is why average resolution time belongs beside both numbers. Containment says the conversation stayed; average resolution time says how long closure took; together they show whether automation shortened the path or simply moved where the waiting happened.
What do containment and resolution mean in plain terms?
Think of containment as a measure of whether the call stayed in the room, and resolution as a measure of whether anyone in the room fixed the thing. A locked door produces an excellent first number.
Picture a customer asking to change a delivery address. The automation explains the policy, the customer says thanks, the session closes, and the parcel still goes to the old address. That conversation is contained. Two days later the same person is on the phone with a human about a missed delivery, and the company has paid twice for one problem.
The tradeoff nobody enjoys: the fastest way to raise the first number is to make leaving harder, and the fastest way to raise the second is to let people leave sooner. Any team reporting one number in isolation has picked a side without saying so.
Common containment and resolution mistakes
Four patterns account for most of the damage.
Counting a handoff purely as a containment failure. The mechanism is a missing outcome field: once a conversation leaves the automated layer, most stacks stop attributing the eventual fix to anything, so the agent gets no credit for the diagnosis it completed before transferring.
Scoring resolution at session close. Closure is a state the system controls, while the customer’s return is the only evidence that counts, and it arrives days later. A same-day resolution figure is a containment figure wearing a different label.
Reporting one blended percentage across every intent. Averaging a password reset with a billing dispute produces a number that moves whenever traffic mix moves, and the roadmap then chases noise.
Letting the automation team own both definitions. When the group being measured also decides what counts as resolved, the definition drifts toward whatever the system does well, and nobody notices until repeat contacts surface in someone else’s report.
Frequently Asked Questions
What is the difference between containment rate and resolution rate?
The difference between containment rate and resolution rate is what each one observes. Containment rate observes where a conversation ended: inside the automated layer, with no transfer to a person. Resolution rate observes what happened to the customer’s issue afterwards, judged by whether they came back. A session can pass the first test and fail the second.
Can containment be high while resolution is low?
Containment can be high while resolution is low, and the pattern is common. It appears when the automation offers no handoff path, so customers exhaust the conversation and leave without a fix. The tell is a rise in repeat contacts within a week, often arriving on a different channel and logged as brand-new issues.
Is deflection the same as containment?
Deflection and containment describe different moments. Deflection counts a contact that never reached the support queue at all, usually because a help article or search result answered the question first. Containment counts a conversation that did start with the automated layer and finished there. Deflection is measured against traffic; containment is measured against sessions.
How do you measure resolution after a chatbot conversation ends?
Resolution after a chatbot conversation is measured on a delay. Teams define a reopen window, commonly seven or fourteen days, then check for repeat contacts from the same customer about the same issue across every channel, plus any ticket opened downstream. Survey responses and a manually reviewed sample cover what those signals miss.
What is a good containment rate for a support team?
A good containment rate depends entirely on your contact mix, which makes any single target figure misleading. A queue dominated by order-status questions will contain far more traffic than one handling billing disputes or account recovery. Compare the number against your own prior period, split by contact reason, and always read it beside resolution.
Should containment or resolution be the primary KPI?
Resolution deserves the primary slot, with containment reported beside it as a diagnostic. Resolution attaches to work that actually ended and to cost the business avoided, so it holds up in a finance review. Containment then explains where automation is holding conversations it cannot finish, which tells you which intents to fix next.