Intercom Fin — AI Customer Support at Scale
Intercom Fin — AI Customer Support at Scale
TL;DR: Intercom integrated Claude into their Fin AI agent, serving 25,000+ customers and handling millions of support queries. Result: 86% resolution rate with human-quality responses, 40% fewer escalations, and response times dropping from 30 minutes to seconds.
Company Context
Intercom: Leading customer service platform Scale: 25,000+ customers globally Challenge: Help customers resolve support queries at scale without sacrificing quality
The Integration
Intercom built Fin with four Claude-powered capabilities:
1. Knowledge
Fin learns everything about a company’s products and services from their knowledge base.
Before: Customers spend 30 minutes searching help articles After: Instant answers generated from the full knowledge base
2. Behavior
Fin adapts communication style to match each business — tone, formality, and response length customized per company.
3. Actions
Fin takes concrete actions on behalf of customers:
- Processing refunds
- Managing account changes
- Updating settings
- Routing to specialists when needed
4. Personalization
Responses tailored to customer context, history, and the specific nature of their query.
Results
| Metric | Achievement |
|---|---|
| Resolution rate | 86% |
| Out-of-box baseline | 51% |
| Response time | 30 minutes → seconds |
| Language support | 45+ languages |
| Human escalation reduction | 40% |
Customer Success Stories
Synthesia
Timeline: 6 months with Fin
| Metric | Result |
|---|---|
| Conversations resolved by AI | 6,000+ |
| Hours saved | 1,300+ |
| Self-serve support rate | Up to 87% |
Fundrise
Timeline: 3 months with Fin
| Metric | Result |
|---|---|
| Support volume automated | 50%+ |
| Response accuracy | 95% maintained |
Implementation Philosophy
Intercom designed Fin to genuinely resolve customer issues, not deflect them.
“Truly resolving customer questions is the better approach to customer support in the long run.”
This philosophy drives:
- Quality over speed optimization
- Resolution rate as primary metric
- Human escalation as feature, not failure
Technical Approach
Rigorous Testing
Intercom’s ML team conducts exhaustive evaluations:
- All updates compared against production baselines
- Performance validated before deployment
- Consistency maintained at scale
Partnership Value
Close work with Anthropic’s solutions engineers helped:
- Optimize implementation
- Unlock personalization features
- Develop policy-aware responses
- Build conversation analysis capabilities
Key Design Decisions
- Resolution focus — Metric that matters is problems solved, not conversations deflected
- Multilingual native — 45+ languages built-in, not bolted on
- Customizable behavior — Each business gets their own Fin personality
- Action capability — Not just answering questions, but taking actions
- Human handoff — Seamless escalation for complex issues
Implications for Other Businesses
When This Pattern Works
- High-volume support operations
- Well-documented knowledge bases
- Definable success criteria (resolution)
- Need for multilingual support
Key Success Factors
- Quality knowledge base as foundation
- Clear escalation triggers
- Regular evaluation against baselines
- Focus on resolution, not deflection
Key Takeaways
- 86% resolution rate achievable with proper implementation
- 51% baseline out-of-box — improvement comes from customization
- Action capability (refunds, account changes) differentiates from simple chatbots
- Resolution focus beats deflection focus
- Rigorous testing against production baselines essential
Related
- automation/departmental-ai-guide — Customer support implementation guide
- glossary/ai-agent — What AI agents are
- tools/mcp — Integration protocol used
- glossary/llm-evals — Evaluation approach used by Intercom
Sources
- Intercom Customer Story — Anthropic
- Fin 2: Powered by Claude — Intercom Blog