Zendesk Verified vs Contained Resolution Report

In May 2026 Zendesk changed how AI agent outcomes are reported. Automated resolution rate stopped being one number and became a mix of two tiers: contained resolution and verified resolution. Most teams saw their AR% go up on the day the change rolled out, without their automation getting any better.

That is the whole problem this report solves. A single blended automated resolution rate now hides the difference between “the AI ended the conversation” and “the AI ended the conversation and something confirmed the customer was actually done.” This guide explains how to split the tiers, how to read the mix, and what to do when contained resolutions grow faster than verified ones.

What changed in Zendesk AI agent reporting

Zendesk unified conversation statuses across messaging, email, and voice, so every AI agent conversation now lands in one of four outcomes:

Outcome	What it means
Unassisted conversation	Only small talk or system replies. No automation performed.
Assisted escalation	The AI contributed, then a human completed the resolution.
Contained resolution	A meaningful request handled by AI with no human involvement or follow-up.
Verified resolution	A meaningful request handled by AI, with additional signals confirming the outcome was complete and satisfactory.

The automated resolution rate calculation changed with it:

Before May 18, 2026: Verified / (Unassisted + Assisted escalations + Contained + Verified) × 100
After May 18, 2026: (Contained + Verified) / (Unassisted + Assisted escalations + Contained + Verified) × 100

Two consequences matter for reporting. First, AR% mechanically increased for most accounts because the numerator got wider. Second, billing did not follow it — Zendesk bills on verified resolutions, and contained resolutions do not consume automated resolutions. So the number that went up is not the number you pay for.

Zendesk also re-mapped historical conversations up to two years back to the new definitions, which means your year-over-year AI trend was rewritten. If you kept manual snapshots of AR% from before May 2026, they no longer reconcile with what Explore shows today.

What this report should answer

A useful verified vs contained report answers:

What share of AI outcomes are verified, contained, assisted escalations, and unassisted?
Is the automated resolution rate rising because more work is confirmed, or because more work is merely contained?
Which use cases, channels, and languages produce verified outcomes, and which only produce contained ones?
Do contained resolutions turn into follow-up tickets, reopens, or repeat contact?

For metric definitions, use the glossary rather than restating them in your dashboard: verified resolution, contained resolution, assisted escalation, bot containment rate, and bot resolution rate. Keep the report beside your support metrics dashboard so AI outcomes are reviewed next to human queue health, not in isolation.

How to build the report in Zendesk

There are two data paths, and serious teams use both.

1. Start in the AI agents reporting dashboard

The AI agents workspace reporting dashboard now shows the tier breakdown natively and updates hourly. Group results by AI agent, channel, and language, then read the Contact reasons and Custom resolutions tabs. This is the fastest way to see the mix, but it stops at the conversation boundary — it will not tell you what happened to the ticket afterwards.

2. Move to Explore for ticket-level truth

Support tickets gained new standard fields alongside the existing Resolution type field:

Resolution tier — Assisted escalation, Contained resolution, or Verified resolution.
Channel group — Digital for email and messaging, Voice for voice channels.

In Explore, use the Support: Tickets dataset and add Resolution tier as an attribute. Once that field is a dimension, you can join AI outcomes to the metrics that actually decide whether automation worked:

reopen rate by resolution tier
repeat contact rate by resolution tier
CSAT by resolution tier
downstream resolution time for tickets that escalated after an AI attempt

3. Build the four core views

Resolution mix over time. A stacked area or column chart of unassisted, assisted escalation, contained, and verified as a share of total AI conversations. This is your headline view. A flat AR% with a shifting internal mix is a change worth investigating.

Verified share of automated resolutions. Verified / (Contained + Verified). This is the single most useful derived number after the change, because it tracks confirmation quality independently of automation volume.

Tier by use case and channel. Use the Use case performance report to find where verification succeeds. Password resets and order status lookups usually verify well. Billing disputes and account changes often contain without verifying.

Post-resolution behaviour by tier. Count of tickets created by the same requester within 72 hours of a contained resolution, compared with the same window after a verified resolution. This is the view that tells you whether containment is real.

4. Set the comparison window carefully

Because Zendesk re-mapped historical data, comparisons across the May 18, 2026 boundary are only valid if you use the new definitions on both sides. Compare re-mapped history to current data, never an old exported AR% to a current one. If a leadership deck cites AI performance from before the change, restate it before reusing the number.

How to interpret the patterns

This is the default outcome of the definitional change, not an improvement. More conversations now count in the numerator. Report the verified share alongside AR% so the jump is not mistaken for progress.

Contained is growing while verified is shrinking

The AI is ending more conversations without confirmation signals. Sometimes that is genuine — a customer got their answer and left. Often it means conversations are timing out or customers are giving up. Check reopen rate and repeat contact rate for those requesters before deciding.

High assisted escalation, low contained

The AI is doing real work but rarely finishing. This is not a failure state. It usually means your use cases are complex enough that partial handling is the right ceiling. The question becomes handoff quality: check escalation quality and whether customers repeat themselves to the human agent.

Conversations are reaching the AI agent that it was never designed to handle, or routing is sending small talk into an automation flow. This is a routing and intake problem, not a model problem. Compare with Zendesk Ticket Deflection Report.

Verified is strong on one channel only

Voice and digital verify differently because the confirmation signals differ. Use the Channel group field to keep them separate before drawing conclusions about the AI agent itself.

Common mistakes

Treating the AR% increase as a performance win. The formula changed. Restate the baseline before claiming improvement.
Reporting AR% as a cost metric. You are billed on verified resolutions. Contained resolutions do not consume automated resolutions, so AR% and your invoice diverge by design.
Comparing pre-change exports to post-change dashboards. Historical data was re-mapped in Zendesk but not in your spreadsheets.
Ignoring assisted escalations. They are the clearest evidence of AI value in complex queues, and teams that only track full automation undercount them.
Reading tiers without downstream context. A contained resolution followed by a new ticket two hours later is not a resolution. Pair the tier with reopen rate and repeat contact rate.
Forgetting the intelligent triage change. Intelligent triage auto-replies are no longer charged as automated resolutions, which shifts historical cost comparisons too.

What to do when contained outruns verified

If the contained tier grows while verified stays flat:

Isolate the use cases where the gap is widest. The Use case performance report makes this quick.
Pull a sample of contained conversations and read the last three turns. You are looking for whether the customer confirmed, went quiet, or restated the problem.
Check the same requesters for a follow-up ticket in the next 72 hours. Silent containment plus a new ticket is a failed resolution wearing a success label.
Decide whether the fix is knowledge coverage, an explicit confirmation step at the end of the flow, or a narrower use case scope.
Re-measure verified share, not AR%, four weeks later.

The goal is not to maximise the automated resolution rate. It is to make sure the share of that rate you can actually trust is growing.

Where this report fits in your dashboard

Keep this report beside:

Together those views show how much work automation absorbed, how much of it was confirmed, and how much came back anyway.

FAQ

Does a higher automated resolution rate mean my AI improved? Not by itself, and not across the May 18, 2026 boundary. The numerator changed to include contained resolutions. Track verified share of automated resolutions to see real movement.

Which tier am I billed for? Verified resolutions. Contained resolutions count toward AR% but do not consume automated resolutions. That is why AR% is a performance metric and not a cost forecast.

Can I report resolution tier in Explore, or only in the AI agents dashboard? Both. The AI agents dashboard gives the fastest conversation-level view; the Resolution tier and Channel group ticket fields let you build ticket-level reports in Explore and join AI outcomes to reopen rate, CSAT, and resolution time.

My historical AI numbers changed. Is that a bug? No. Zendesk re-mapped conversations up to two years back so historical data uses the new definitions consistently. Your old exports are the thing that is now out of date.

How often should I review this report? Weekly for the tier mix, monthly for use case level verification quality. Automation drifts faster than human queue performance because knowledge sources and flows change underneath it.

See verified and contained AI resolutions next to reopens and CSAT - start free