Intercom Fin: ‘resolved’ can mean silence; handoffs are a separate outcome

Original definitions. Intercom’s Fin outcomes help page explicitly calls an agent’s answer followed by no request for more help an assumed resolution, including when a frustrated customer leaves without asking for more help; these count as billed outcomes. If the last agent message asks a clarification question, silence means abandonment instead. Its conversation guidance warns that ‘I’m willing to wait for an update’ can be misread as positive feedback, classified as confirmed resolution, and closed; it recommends an escalation rule. Thus status, customer silence, actual issue repair and customer-perceived success are separate observations.

Permission and cost boundaries. In Intercom’s own account of the move from resolution to ‘outcomes’ pricing, an AI completing its configured part before an intentional Procedure handoff also counts as a billable outcome; this is not the same as human-free resolution. Approval-step docs allow a specialist to approve or reject risky actions (refunds, account closure, deletion); a timeout escalates, rather than treating silence as approval. Pricing rules classify routine customer-requested/default/workspace-rule escalations as nonbillable and configured Procedure handoffs as billable. Note the conflict of interest: this is vendor-operational documentation and reported results, not an independent audit.

The denominator matters. Intercom’s June–July 2026 reporting change excluded conversations in which Fin was active but could not answer because configured rules intervened from ‘Fin involved.’ It states the change increases the displayed resolution rate among involved but leaves automation rate (Fin resolutions / all new conversations) unchanged: worked example 250/750→250/500 (33%→50%) while 250/1,000 remains 25%. These figures are a didactic example, not actual customer outcomes. In performance reporting, split total conversations into not active / constrained / involved, then resolved / intentional handoff / escalated / pending; survey-based CX also has selection. Comparisons should use all-incoming requests, verified closure and reopenings alongside short-term resolution rate.

Operator role evidence. Intercom’s self-case report says it created a Knowledge Manager, a Conversation Designer responsible for the end-to-end human handoff, a dedicated AI Support team and specialized technical support roles while claiming 81% of its own support volume resolved by Fin. Its 100 avoided hires / $7.5–9m savings estimate is a modeled counterfactual, not independently established cash savings. In particular, output shifts to keeping the knowledge base, routing and agent evaluation reliable, not zero support work.

Compare Taobao’s experimental retrial and rating measures and Nubank’s agent-variant tests; see closed-loop brief and role map.