Intercom Fin

Editorial verdict

Fin is Intercom’s customer-facing AI agent for answering support questions, completing configured procedures and handing conversations to humans. Its strongest advantage is a mature support workflow around the model: controlled knowledge sources, audience targeting, omnichannel deployment, reporting and escalation. Outcome-based pricing aligns some cost with successful work, but buyers must understand what counts as an outcome and model volume carefully.

Best for

Fin fits support organizations with a substantial, maintained knowledge base and enough repetitive demand to justify automation. It can run inside Intercom’s helpdesk or, through the standalone offer, alongside an existing helpdesk such as Salesforce. It is less suitable when answers rely on undocumented judgment, rapidly changing policy or high-risk decisions that require a trained person.

How Fin works

Fin retrieves information from approved sources such as public and private help content, PDFs, webpages and internal material, then generates a grounded answer. Teams can configure tone, audience, region, channel, escalation and procedures that interact with external systems. Fin is distinct from Copilot: Fin talks to customers, while Copilot assists human teammates in the inbox.

Current pricing

Official pricing reviewed in September 2026 listed Fin at $0.99 per outcome. An outcome is counted when a customer confirms resolution, does not request more help after Fin responds, or Fin completes a configured procedure, including some handoffs. Intercom says a conversation is charged only once even when Fin answers multiple questions.

Using Fin with Intercom also involves seat pricing: Essential was $29 per seat per month, Advanced $85 and Expert $132 on the displayed annual option. Standalone Fin required no Intercom seats but applied a minimum monthly commitment, with 50 outcomes given as an example. Additional charges can include WhatsApp, SMS, phone, campaigns, Copilot and Pro operational tooling.

Plan and add-on context

Essential includes Messenger, shared inbox, ticketing, reports and a public help center. Advanced adds multiple team inboxes, workflow automation, round-robin assignment and private or multilingual help centers. Expert adds SSO, identity management, SLAs, multibrand capabilities and HIPAA support. Copilot was listed at $29 per agent monthly on annual billing; Pro began at $99 monthly for analysis and operational AI features. Confirm packaging and regional availability in the live quote.

Channels and controls

Intercom documents Fin support across Messenger, email, WhatsApp, SMS, Facebook and Instagram, with additional deployment possible through integrations. Teams can target customers by plan, location or brand; adjust content by audience; define handoffs; and stop answers at a resolution limit. API-created conversations require a Workflow to trigger Fin. Third-party channel integrations may have different troubleshooting support.

Where it falls short

Fin’s accuracy cannot exceed the quality, scope and freshness of its sources and procedures. Outcome pricing can be economical for high-value resolutions but unpredictable when volume grows or when the definition includes procedure handoffs. Reported resolution rate is not the same as verified customer success, retained revenue or lower total support cost. Unsupported edge cases, authentication, regulated advice and emotionally sensitive conversations need deliberate escalation.

A practical evaluation test

1. Select at least 100 representative historical conversations, including known failure and escalation cases.
2. Clean the relevant knowledge sources, remove conflicts and assign content owners.
3. Run Fin in preview or controlled traffic, scoring factual accuracy, policy compliance, citation grounding and tone.
4. Test multilingual, ambiguous, adversarial and account-specific questions.
5. Exercise identity checks, procedures, handoffs, outages and human takeover.
6. Calculate cost per verified resolution using outcomes, seats, channels, add-ons and the human work required to maintain content.

Security, privacy and AI governance

Intercom’s additional product terms state that inputs can include help content, conversation data, procedures and personal data, and may be processed by third-party AI providers acting as subprocessors. Intercom contractually restricts those providers from using Customer Data to train their services. Customers remain responsible for lawful inputs, ongoing suitability, transparency and independent review of outputs.

Inform users they are interacting with AI, mark generated outputs as required, restrict sensitive fields, apply retention controls and test that audience segmentation cannot leak private content. Regulated buyers should validate hosting region, DPA, subprocessors, HIPAA configuration where applicable, audit logs, incident handling and the 99.8% target availability described in the SLA.

Alternatives

Zendesk AI suits Zendesk-centered teams, Ada focuses on automated customer service, Salesforce Agentforce fits Salesforce estates, Forethought targets support automation, and custom retrieval systems offer control for organizations with strong AI engineering and governance.

Final recommendation

Choose Fin when the support knowledge and escalation system are already disciplined. Pilot against real conversations, measure verified outcomes rather than vendor resolution labels, and set hard spend limits. The standalone option is attractive for preserving an existing helpdesk; the integrated Intercom route is stronger when inbox, ticketing, knowledge and automation should share one platform.

Research note: pricing, outcome definitions, channels, capabilities and AI terms were checked against Intercom’s official pricing, help and legal documentation on September 1, 2026.