Fin by Intercom

Fin can be live on an existing helpdesk (Salesforce, HubSpot, Freshdesk, Zendesk via API/MCP) in under an hour without a migration project, and it works across chat, email, voice, WhatsApp, SMS and Slack from one configuration. The clearest independent evidence of real-world performance comes from Anthropic's own published case study: after roughly two years of tuning, Fin is involved in 80% of Anthropic's incoming support queries, resolves 79% of the conversations it touches, and now closes 63% of Anthropic's total support volume end-to-end (about 560,000 resolutions/month), up from 'nearly half' of conversations in its first weeks live. Anthropic picked Fin after evaluating build-vs-buy and has publicly credited the underlying Claude models for answer quality. Intercom was also an early mover on usage-based AI pricing (charging per resolved outcome rather than per seat) starting in 2023, a model most competitors have since copied.

Fin's own marketing claims a 76% average resolution rate across 12,000+ AI Agent customers (30,000+ companies use the broader Fin/Intercom product line). VentureBeat reported in March 2026, based on benchmarks Intercom shared with the outlet, that Fin's average resolution rate had grown from 23% at 2023 launch to 67%, rising roughly 1 percentage point per month since — a trajectory consistent with today's higher vendor figure. VentureBeat also noted Intercom declined to name the base model behind its 'Apex' model, despite describing itself as transparent, and could not independently verify the vendor-supplied benchmark comparisons against GPT-5.4 and Claude. Separately, a cluster of blog posts (Builts AI, CloneDesk and similar sites) claim independently-tested resolution rates in the 38–53% range; these sites sell competing AI-support or automation services and don't disclose that conflict prominently, so treat their specific numbers as competitor marketing rather than neutral benchmarks. The one number we could trace to a specific, named customer with public attribution is Anthropic's 79%.

As of today, Fin costs $0.99 per 'outcome' on every plan. Used with the Intercom helpdesk, that's layered on top of per-seat pricing: Essential from $29/seat/month, Advanced from $85/seat/month, Expert from $132/seat/month (annual billing; roughly 25-30% more if billed monthly). Used standalone with another helpdesk, it's $0.99/outcome with no seats required but a 50-outcome/month minimum and custom quoting via sales. Per Intercom's own help documentation, an outcome is billed once per conversation and includes a 'confirmed resolution' (the customer says it helped) as well as an 'assumed resolution' — the customer simply doesn't reply again after Fin's last answer, whether or not they were actually satisfied. Reported G2 reviews cite this outcome-based model as the most common source of pricing complaints, with costs described as unpredictable and rising quickly at higher volume once seats and outcomes are both scaling.

In May 2026 Intercom renamed its corporate entity to Fin, and on June 15, 2026 Salesforce announced a definitive agreement to acquire Fin for approximately $3.6 billion — Salesforce's largest AI-agent acquisition to date. The deal is expected to close in Salesforce's fiscal Q4 2027, subject to regulatory clearance, and had not closed as of this listing. Fin was founded as Intercom in 2011, raised roughly $292M in venture funding, and was last valued around $1.3B in a 2018 private round — the acquisition price represents a large step up from that figure. Fin currently markets itself as helpdesk-agnostic (works with Salesforce, HubSpot, Freshdesk and others), but Salesforce is also the maker of the competing Agentforce product line; buyers who are not already on Salesforce and who value long-term platform independence should treat the roadmap and continued multi-helpdesk support as an open question until the deal closes and Salesforce states its integration plans.

Intercom/Fin publishes a real security page (not just marketing copy) listing SOC 2 Type II, ISO 27001, ISO 42001 (AI governance), ISO 27701 (privacy) and AIUC-1 (an AI-agent-specific standard) certifications, plus GDPR/CCPA compliance and a self-serve trust center for audit reports. On April 30, 2026, Intercom's official 'intercom-client' npm package (7.0.4) was compromised for about two hours as part of the 'Shai-Hulud' supply-chain worm campaign: attackers used a stolen CI/CD credential (not a stolen maintainer account) to publish a version containing a credential-harvesting payload, affecting a package with roughly 361,000 weekly downloads. GitHub's security advisory rates it CVSS 9.3 (critical); Intercom pulled the package and the community coordinated a fix within the exposure window. This is a CI/CD pipeline compromise on a client SDK, not a breach of Fin's hosted infrastructure or customer data — but it's a data point on their software-supply-chain security posture worth knowing.