Resolution vs. containment vs. deflection

Deflection counts customers who used self-service and did not open a ticket; containment counts AI conversations that were not handed to a human; resolution counts conversations the vendor's rule says were resolved. Each is a step closer to the customer's problem being solved, and only an audited resolution checks that it stayed solved: no repeat contact within 7 days, no handoff, and a read confirmation.

DeflectionContainmentResolution (vendor's rule)Audited resolution
CountsSelf-service sessions with no ticket afterAI conversations with no transfer to a humanAI conversations the vendor's rule counts as resolvedClaimed resolutions that held
Typical ruleA ratio of sessions to tickets; Zendesk divides help center sessions by users who opened tickets (Zendesk)No escalationNo escalation, plus a quiet period (72 hours or less where a vendor states one), an answer from content, or the vendor's own model (definitions compared)No repeat contact in 7 days, no handoff of any kind, judge confirmation on a sample (standard)
Can count a customer who gave upYesYesYes, under a silence ruleOnly if they never came back and the answer read as resolving the request
Can count a customer who came back the next dayYesYesYes, in a new conversation or after the window closesNo
Who decidesThe tool's analyticsThe platform's escalation recordThe vendorRules on the customer's records, and a judge validated against people
GlossaryDeflection rateContainment rateResolution rateAudited resolution

How they nest#

self-service sessions
 └─ deflected (no ticket opened)            ← deflection
AI conversations
 └─ not handed to a human                  ← containment
     └─ counted resolved by the vendor     ← resolution rate (claimed)
         └─ held: no repeat contact, no
            handoff, read as resolved      ← audited resolution

Deflection sits outside the conversation count: a deflected customer has no conversation to check, so Attify's standards do not measure it.

Synthetic example#

Of 1,000 AI conversations: 850 not handed to a human (containment 85.0%); the vendor counts 700 resolved (70.0%); 441 pass the 7-day, handoff and judge checks (audited 44.1%). Three different rates, each correct under its own rule.

Questions

Our bot shows 80% deflection but ticket volume hasn't dropped. Why?

Deflection counts customers who did not open a ticket after self-service, not customers whose problem was solved. Some came back another way, and some would never have opened a ticket. Look at repeat contacts and tickets per customer instead.

Which one should I report to finance?

Audited resolutions, because they check that the outcome held; the other three do not. Show the vendor's resolution rate next to it, with the rule each used.

Sources

  1. Zendesk. Getting started with self-service, Part 6: Tracking essential self-service metrics. Read 8 Nov 2026.
  2. Attify. Attify Audited Resolution Standard v0.11: Relation to established metrics. Published 9 Nov 2026. Read 9 Nov 2026.