AI & ML Quick Path

Company case

Intercom: Agents / customer support

Customer supportIntermediateRebuild effort 30–40 h (estimate)

What the published sources confirm

Intercom's Fin is an AI agent that answers customer questions from a company's help content and hands off to human agents. Intercom publishes an average "resolution rate" across customers and charges per resolution.

Business problem and AI concept

Problem
Support teams receive many questions already answered in help articles.
Concept
Grounded answering over help content, escalation, and outcome-based measurement.

Reported results

  • Intercom's published average resolution rate has changed over time: an earlier Intercom blog post cited 41%; an Intercom comparison page checked on 2026-10-30 cites 76% across 8,000+ customers.

    Intercom pages, checked 2026-10-30. Caveat: Vendor-defined metric that changes with product versions and customer mix; compare definitions before comparing numbers.

Suggested workflow

Written for learning, not the company's system. This flow shows how you could build something similar. It is not a description of Intercom's private architecture.

  1. Step 1: Ingest help-centre articles and keep them in sync

  2. Step 2: Retrieve relevant passages for each question

  3. Step 3: Answer with links to sources

  4. Step 4: Detect low confidence or requests for a person

  5. Step 5: Hand off with a conversation summary

  6. Step 6: Measure resolution with a clear definition and sample reviews

Suggested technologies and alternatives

LayerSuggestedAlternatives
RetrievalHybrid search over help articlesManaged search service
AnsweringModel API with grounding promptLangChain retriever chain
HandoffHelpdesk APIWebhook to ticket queue
QualityWeekly human review sampleAutomated faithfulness checks

Small example

Input
"How do I change the email on my account?"
Output
Three-step answer citing the "Update your email" article; offers a person if the steps do not work.

Rebuild it yourself

Build a help-centre answer bot over 30 synthetic articles with escalation and a resolution-rate definition.

Definition of done

  • Written definition of "resolved"
  • Every answer links an article
  • Escalation works for 10 scripted cases
  • Weekly review sample procedure documented

Security and privacy risks

  • Counting abandoned conversations as resolved
  • Stale articles
  • Hallucinated steps

Sources

Listed by title and publisher. This guide contains no external links; search for the title if you want to read the original.

  • Fin, the AI agent for customer service, keeps getting better. Intercom blog. Vendor blog.
  • Intercom vs LiveAgent comparison (states current average resolution rate). Intercom. Vendor page.