Intercom Fin AI Agent: The Complete Guide (2026)

We sell an AI agent ourselves (Macha, more on that near the end), so treat this as a guide written by people who watch this market closely — not a neutral encyclopedia and not a Fin advertisement. Where Fin is strong, we'll say so plainly; where it has real watch-outs, we'll name them.

Update — June 2026: On June 15, 2026, Salesforce signed a definitive agreement to acquire Fin (the company formerly known as Intercom) for ~$3.6 billion. The deal is expected to close around Q4 of Salesforce's FY2027, after which Fin's technology will fold into Salesforce's Agentforce platform. Fin operates independently until the deal closes, so the pricing, features, and trial described below remain current for now — but factor the pending change of ownership into any long-term decision. Sources: Salesforce press release · TechCrunch.

What is Intercom Fin?

Fin is Intercom's AI agent for customer service — software that reads a customer's question, finds the answer in your content, and replies in a human-like way to resolve the issue without a human agent. Intercom positions it as the front line of support: Fin answers first, and only the conversations it can't handle reach your team.

One naming note before we go further: the company rebranded from Intercom to Fin, so "Fin" is now both the product and the company name. We keep saying "Intercom Fin" throughout this guide because that's how most people still search for it, but the current official name is simply Fin.

Under the hood, Fin runs on what Intercom calls the Fin AI Engine — a pipeline of custom models (query refinement, retrieval, reranking, answer generation, and a validation step meant to curb hallucinations) layered on top of frontier large language models. Intercom now brands its latest model as Fin Apex 1.0 and describes a network of models trained specifically on real customer-service conversations (fin.ai). The practical takeaway: Fin isn't a thin wrapper around a chatbot — it's a retrieval-augmented system built to answer from your knowledge rather than the open internet.

Intercom reports Fin has handled 40M+ resolved conversations and cites a trailing-30-day resolution rate around 67% (vendor-reported, as of late 2025), and it backs an enterprise $1M performance guarantee tied to a 65% resolution floor for qualifying accounts (fin.ai). Those are Intercom's own numbers, so weigh them as marketing claims rather than independent benchmarks — but the scale is real, and Fin is one of the most widely deployed AI agents on the market.

How Fin works

Fin's core loop is simple to describe and harder to do well: ingest your content, retrieve the right passage for each question, generate a grounded answer, and escalate when it isn't confident.

It answers from your content. Fin pulls from Help Center articles, internal support docs, PDFs, webpages, and short text "snippets" you write for it, then uses retrieval-augmented generation to ground its replies in those sources (Intercom Help). This is the single biggest determinant of how good Fin is for you: thin or stale documentation produces a thin, stale agent.

It works across channels. Fin isn't chat-only. It runs over live chat (Messenger, WhatsApp, SMS, social), email (with spam/phishing filtering), and phone via Fin Voice, plus internal channels like Slack and Discord. It supports 45+ languages with real-time translation, and Fin Vision lets it interpret images like screenshots and invoices (Intercom Help).

It does more than answer — Procedures and Actions. Beyond Q&A, Fin can run multi-step workflows. Procedures (Intercom's evolution of what it used to call "Fin Tasks") are natural-language instructions that let Fin follow business rules, call APIs/data connectors, and take actions like updating an order or tagging a ticket. Actions connect Fin to your backend systems so it can fetch order status, process a return, or look up account data mid-conversation — the difference between "here's the help article" and "done, your return is processed."

It hands off when it should. Fin is designed to escalate to a human when that's the safest option, and you can configure escalation rules. Notably, if a customer explicitly asks for a human, Fin steps aside (and, importantly for cost, you aren't billed — more below).

Fin beyond Intercom's helpdesk. This is an underrated point: Fin doesn't require you to use Intercom's helpdesk. Through "Fin for platforms," Intercom officially documents Fin running on Salesforce (Service Cloud), HubSpot, and Freshworks (Freshdesk/Freshchat), with other platforms available on request via sales (Intercom Help). Separately, Intercom ships a dedicated Fin for Zendesk integration that works across Zendesk Messenger and Zendesk tickets (fin.ai Help — Fin for Zendesk). If you're on something else (e.g., Zoho Desk, Front, Dixa or Sprinklr), it's worth asking sales rather than assuming support — those aren't named in Intercom's public platform docs. The catch across the board: deploying Fin on a non-Intercom helpdesk is sold through Intercom's sales team, not self-serve, whereas Fin-on-Intercom you can trial yourself.

Key capabilities at a glance

• Grounded answers from Help Center, docs, PDFs, webpages, and snippets (RAG with a validation step).
• Omnichannel: chat, email, phone/voice (Fin Voice), WhatsApp, SMS, social, Slack/Discord.
• Procedures for multi-step, rule-bound workflows (formerly "Tasks").
• Actions / data connectors to read and write to your backend systems.
• Fin Vision for image understanding; 45+ languages with translation.
• Testing tools: simulations, batch testing, and answer inspection before you go live.
• Runs on your existing helpdesk (Intercom natively; Salesforce, HubSpot and Freshworks via "Fin for platforms," plus a dedicated Fin for Zendesk integration).

Intercom Fin pricing: the $0.99-per-resolution model explained

Here's the headline most people come for. Fin charges $0.99 per outcome. As of June 2026, per fin.ai/pricing and Intercom's outcomes documentation, the core support outcomes are:

There's a $49/month base plan. For the standalone "Fin for platforms" deployment (Fin on a non-Intercom helpdesk), Intercom documents a $49/month base fee that includes 50 resolutions, with each additional outcome at $0.99 (Fin pricing: Outcomes). So 100 resolutions in a month works out to roughly $49 (first 50) + $49.50 (next 50) ≈ $99.50. This base plan is the practical floor most people will hit before the per-outcome charges really stack up.

A note on "qualification" and "disqualification." You'll see those listed as outcomes elsewhere — **qualification ($9.99) and disqualification are Fin for Sales outcomes, not standard support resolutions. They apply to Fin scoring sales prospects, and Intercom states Fin for Sales (qualifications/disqualifications) is currently not available with Fin for platforms outside Intercom** (Fin pricing: Outcomes). If you're buying Fin for customer support, the numbers that matter to you are the $0.99 resolution and the $49 base.

There's also a 14-day free trial with unlimited Fin outcomes and no credit card required, so you can pressure-test it on real traffic before committing (fin.ai/pricing).

What's free vs. charged. You're billed once per conversation even if Fin answers multiple questions or runs several Procedures. You are not charged when: a customer explicitly asks for a human, Fin only responds to a greeting (a "hi" back isn't an answer), a Procedure fails to complete, or a conversation transfers to your team without a completed outcome (Intercom Help). That structure is genuinely customer-friendly in places — you don't pay for Fin punting to a human.

If you also use Intercom's helpdesk, Fin's per-outcome charge sits on top of seat fees. Intercom's seats run Essential at $39/seat/mo ($29 annual), Advanced at $99/seat/mo ($85 annual), and Expert at $139/seat/mo ($132 annual), with optional add-ons like Copilot ($35/user/mo) and a Pro analysis add-on ($99 per 1,000 conversations) (Intercom pricing). On a non-Intercom helpdesk via Fin for platforms, you pay the outcome charges without Intercom seats.

The "resolution" definition — and the debate

This is where Fin gets contentious, and it's worth understanding before you sign anything. A resolution is counted two ways (Intercom Help):

1. Confirmed resolution — the customer affirms the answer helped ("yes," "thanks," etc.).
2. Assumed resolution — the customer simply disengages for 24 hours after Fin's last reply.

The assumed-resolution case is the lightning rod. A customer leaving isn't the same as a customer being helped — people abandon chats because they got distracted, gave up, or found the answer elsewhere. Critics (and a fair number of Reddit threads) argue you can be billed for "resolutions" that didn't resolve anything (aimdoc, Gleap). Intercom's counter is that confirmed-plus-assumed is the industry-standard way to measure deflection at scale, and that the carve-outs (no charge for human requests, greetings, failed procedures) prevent the worst over-billing. Both points are fair; just go in with eyes open about what you're paying for.

Cost predictability at scale

The deeper issue isn't the $0.99 — it's that the bill scales with how much customers use Fin, not with anything you directly control. That's great when volume is low and brutal when it spikes. Third-party roundups and community posts cite migrations where bills jumped from roughly $119 to $854/month, one team reporting $4,000 climbing to $9,000/month, and another projecting $1,200 to $10,000/month at scale (myaskai, Gleap). Treat those as anecdotal community reports, not audited figures — but the pattern is real and intuitive: the more successful your deployment, the bigger the invoice. Do the math at your projected volume. At 2,000 resolutions/month, Fin's resolution charges alone are roughly $1,980, before seats.

What real users say

We aren't an Intercom customer, so rather than pretend otherwise, here's the verifiable third-party signal. On G2, Fin by Intercom holds a 4.5/5 rating across 2,900+ reviews and is listed as G2's #1-ranked AI Agent by review volume (G2); Intercom as a whole carries a comparable 4.5/5 on Capterra (Capterra). The sentiment is genuinely strong on capability and setup, with the recurring complaint being — predictably — pricing.

A representative example: a Verified User in Financial Services at a mid-market company (51–1,000 employees), reviewing on G2 (May 2025), praised Fin's "seamless setup within the Intercom inbox and its ability to surface accurate, conversational answers from our existing help center — dramatically reducing average first-response times and agent load," while flagging that "higher-tier pricing may be a barrier for smaller teams" (G2). That split — love the product, watch the bill — is the most common pattern in Fin's public reviews, and it lines up with the cost dynamics below.

Strengths

• Genuinely capable. Fin is one of the most mature AI agents available — strong RAG, multi-channel (including voice), real workflow automation via Procedures and Actions, and Vision. It's not a toy.
• Aligned-incentive pricing (mostly). Pay-per-outcome means if Fin doesn't resolve, you broadly don't pay — and you're not billed when it routes to a human. For low-to-moderate volume, that's a clean, low-risk way to start.
• Helpdesk-agnostic. Between Fin for platforms (Salesforce, HubSpot, Freshworks) and the dedicated Fin for Zendesk integration, you don't have to rip out your existing helpdesk to use it.
• Fast to stand up. Intercom claims under an hour to deploy, and the 14-day unlimited trial lets you validate on live traffic with no card.
• Scale and proof points. 40M+ resolved conversations and an enterprise performance guarantee signal a battle-tested product (vendor-reported, but meaningful).

Honest limitations

• Cost unpredictability at scale. The flip side of usage-based pricing: success raises your bill, and forecasting at high volume is hard. Heavy-volume teams often find a flat or per-action model cheaper and steadier.
• The resolution-definition dispute. Assumed resolutions mean you can pay for conversations that may not have been truly resolved. Audit your assumed-vs-confirmed mix.
• Only as good as your content. Fin is RAG-grounded — sparse, outdated, or contradictory documentation produces weak answers. Budget real effort for knowledge upkeep.
• Non-Intercom deployment is sales-led. Want Fin on Zendesk or Salesforce? That's a sales conversation and likely an annual commitment, not a self-serve toggle.
• Stacked costs on Intercom. If you run Intercom's helpdesk too, seats + outcomes + add-ons (Copilot, Pro) add up quickly.
• Roadmap and ownership uncertainty. With Salesforce's pending acquisition (above), Fin's long-term roadmap — and a likely migration into Salesforce's Agentforce platform — is not yet settled. For multi-year decisions, weigh that uncertainty alongside today's product.

Who Fin is best for

Fin is a strong fit if you want a mature, multi-channel AI agent with public, outcome-based pricing and you have (or will build) solid help content. It's especially attractive for teams already on Intercom, for those who value paying only when the agent resolves, and for low-to-moderate ticket volumes where the per-resolution math stays comfortable. It's a tougher fit if you have very high volume (predictability suffers), if your content is thin, or if you want a self-serve deployment on a non-Intercom helpdesk.

Alternatives to consider

Fin is the category benchmark, but it isn't the only option. Worth evaluating honestly:

• Zendesk AI agents (Advanced AI / Ultimate) — native if you're on Zendesk; tightly integrated but priced and architected differently. See our Zendesk vs. Intercom breakdown.
• Decagon, Ada, Sierra — enterprise-grade AI agents competing directly with Fin, generally quote-only.
• Macha — an AI agent layer that runs on top of your existing Zendesk or Freshdesk (more below).
• For a broader, hands-on roundup of options that sit alongside Zendesk, see our guide to the best AI Zendesk alternatives.

Where Macha fits — and the honest pricing contrast

Full disclosure, this is our product. Macha is an AI agent layer that runs on top of your existing Zendesk or Freshdesk — it isn't a helpdesk and it isn't trying to replace yours. Like Fin, it reads the customer's question, answers from your connected knowledge and past tickets, takes actions, and escalates to a human with full context when it isn't confident. You can see how Macha works on Zendesk.

The cleanest, fairest contrast is the pricing model, not the feature list — Fin is a genuinely strong product. Fin bills per resolution/outcome: you pay when a conversation is deemed resolved (including assumed resolutions). Macha bills per AI action — each automated step it takes, whether drafting a reply, tagging, routing, or resolving — because most automation isn't a tidy "resolution," it's work done along the way. Neither model is universally better: per-resolution can be cheaper at low volume and only-pay-on-success is reassuring; per-action tends to be more predictable as volume grows and rewards automation that isn't a clean deflection. The right answer depends on your volume and how much of your work is "answer-and-done" versus multi-step. If you want to compare on your own traffic, you can 7-day free trial, no credit card required.

Frequently asked questions

What is Intercom Fin? Fin is Intercom's AI agent for customer service. It reads customer questions, retrieves answers from your help content using RAG, and resolves issues across chat, email, and voice — escalating to a human when it can't help. It's built on Intercom's proprietary Fin AI Engine layered over frontier LLMs.

How much does Intercom Fin cost? $0.99 per resolution, with a 50-resolution monthly minimum. For the standalone "Fin for platforms" deployment there's a $49/month base plan that includes 50 resolutions, then $0.99 each after. There's a 14-day free trial with unlimited outcomes and no credit card. If you also use Intercom's helpdesk, seats start at $39/seat/mo ($29 annual) on top. (Note: "qualification" at $9.99 and "disqualification" are Fin for Sales outcomes, not standard support resolutions.)

What counts as a "resolution" in Fin? Either a confirmed resolution (the customer affirms Fin's answer helped) or an assumed resolution (the customer disengages for 24 hours after Fin's last reply). You're billed once per conversation. You're not charged when a customer asks for a human, when Fin only replies to a greeting, or when a Procedure fails.

Can Fin work on Zendesk or Salesforce? Yes. Intercom documents "Fin for platforms" on Salesforce, HubSpot and Freshworks (with other platforms on request), and ships a dedicated Fin for Zendesk integration covering Zendesk Messenger and tickets. Note that non-Intercom deployments are handled through Intercom's sales team rather than self-serve.

Is Fin's pricing predictable? At low-to-moderate volume, reasonably so. At high volume it's less predictable because the bill scales with customer usage — more successful deflection means a bigger invoice. Model the cost at your real volume before committing, and watch your assumed-vs-confirmed resolution mix.

What's the difference between Fin and Macha? Both are AI agents that answer from your knowledge and escalate to humans. Macha runs as a layer on top of Zendesk or Freshdesk and bills per AI action; Fin bills per resolution/outcome and can run on Intercom or other helpdesks. Choose based on volume and pricing-model fit.

The bottom line

Intercom Fin is one of the most capable and widely deployed AI agents on the market — strong retrieval, true multi-channel reach (including voice), real workflow automation, and a public price you can actually model. Its $0.99-per-outcome structure is refreshingly transparent and, in places, genuinely fair (you don't pay when it routes to a human). The real questions are about scale and definitions: the bill grows with success, "assumed resolutions" can blur what you're paying for, and Fin is only as good as the content you feed it.

If you're on Intercom or want a mature agent with pay-on-outcome pricing and moderate volume, Fin is an easy shortlist entry. If you're on Zendesk or Freshdesk and want a per-action model that runs on top of your existing stack, weigh an alternative like Macha — and run the numbers on your own traffic before you decide either way.

Pricing and capabilities verified against fin.ai and Intercom's documentation, June 2026. Intercom updates Fin and its pricing frequently — confirm current figures on Intercom's site before relying on them.