Independent editorial guide, written with AI assistance. Examples are hypothetical; no hands-on product testing or owner interviews were performed. Features and prices may change. Our editorial method.
ShiftToAGI, Shifting Founders Beyond AI
Evidence note, checked 2 November 2026. Every product and pricing fact here comes from Intercom's official pricing page, retrieved that day. Intercom's Fin product page returned an error when checked, so this profile leaves out any capability that only that page might describe. Prices and terms change, so confirm them on the official pages before you commit. ShiftToAGI is an independent publisher and is not affiliated with Intercom.
Intercom Fin is an AI agent built for customer service. It draws on your support content to answer customers and can transfer conversations to human agents. Intercom offers it inside its own helpdesk plans and as a standalone agent for some existing helpdesks (pricing).
A hypothetical starting point
This story is invented to illustrate a decision. It does not describe a real company, customer or test.
Priya runs a 12-person business that sells refurbished commercial coffee machines to cafes. About half of her support queue repeats the same five questions: descaling, warranty length, delivery windows, spare part codes and invoice copies. The rest is urgent. A machine fails during a Saturday breakfast rush, and a cafe owner is losing sales while waiting for a reply.
Priya wants the repetitive questions answered overnight. She does not want a bot standing between a panicking cafe owner and her technician. She also does not want to pay for replies that only looked helpful. A Fin pilot has to test that tension between deflection and safe escalation.
What Intercom Fin is, and what sits around it
It helps to separate three things that the pricing page presents together:
- The Intercom helpdesk. Messenger live chat, a shared inbox and ticketing, a public help center and pre-built reports. This is conventional support software, priced per seat.
- Fin AI Agent. The part that answers customers, billed per outcome.
- Optional add-ons. Copilot, an AI assistant for human agents in the inbox, and Pro, which adds analysis features and an operations agent called Operator (pricing).
The split matters for risk. The Advanced plan lists a workflow automation builder, multiple team inboxes and round robin assignment (pricing). Those features move work around. Fin writes replies that customers read, so the risk of a wrong answer lives with Fin and with the content you give it.
Documented features worth understanding
1. Answers based on your support content
Intercom says Fin learns your support content (pricing). The retrieved page does not say which content types or sources are supported. The practical point holds either way: your articles will shape answer quality, so a stale returns policy is a liability. Plan content cleanup before launch.
2. Transfer to human agents
Fin can transfer conversations to agents in the inbox, and Intercom describes workflows called Procedures, which can include handoffs (pricing). For Priya this matters most. The page does not explain how handoff rules are set, what triggers them, or whether topics can be excluded. Configuration unknown: verify all of this during a trial.
3. Channels and existing helpdesks
The standalone option lists answering across channels and working with existing helpdesks, naming Salesforce, HubSpot and Freshdesk. It has no seat or setup fees but carries a minimum monthly commitment, with 50 outcomes given as an example (pricing). Which channels apply to your setup is not specified, so ask.
4. Outcome billing with spending controls
An outcome counts when a customer confirms resolution, when they ask for no more help after Fin replies, or when Fin completes a Procedure, handoffs included. Charges are capped at one per conversation and refunded if the conversation reopens. Intercom says you pay nothing if Fin cannot answer or the customer asks for a person, and it offers spending limits and alerts (pricing).
Two details deserve attention in a pilot. First, silence counts. A customer who gives up also stops asking. Second, a customer asking for a human is not billed, but a handoff completed as a Procedure is listed as an outcome. Ask Intercom how a handoff you configure for urgent topics will be billed.
5. Add-ons for the human side
Copilot supports human agents in the inbox. Pro covers performance analysis, quality monitoring and Operator, which uses credits (pricing). Both are bought separately from Fin.
Who it suits
Fin makes most sense when you receive a steady flow of repeat written questions and someone will keep the help center current. It also suits teams that want AI and human support in one system. Teams settled on one of the named helpdesks can consider the standalone route instead of migrating.
Potential applications across roles
These are proposed uses, not documented case studies.
- Founder or general manager: answer pre-sale and policy questions out of hours, with urgent issues routed to whoever is on call.
- Operations lead: handle delivery window and returns policy questions, provided the facts live in maintained articles rather than in someone's head.
- Customer success in B2B: take routine onboarding how-to questions so specialists spend time on account-specific problems.
- Finance or admin: explain invoice and payment policy in general terms. Credits, refunds and billing changes stay with a person.
- Offline service businesses: clinics, repair shops and trades could cover opening hours, preparation instructions and service areas, while bookings and anything safety-related go to staff.
An illustrative pilot workflow
This is a proposed plan, not a product test. Inputs and outputs are fictional, and any step that depends on a setting must be checked in a trial.
Step 1. Define scope. Choose a short list of question types for Fin and a list of topics that must always reach a person. Priya picks general descaling, warranty, delivery-window and part-code questions supported by approved knowledge. Invoice-copy requests must reach staff who can verify identity and retrieve the actual document; general policy knowledge does not provide that capability. Machine failures, refunds, electrical or water safety and complaints are excluded.
Step 2. Audit content. Confirm each in-scope topic has one current, approved article. Missing-data handling: if no article exists, the topic stays out of scope until someone writes and approves one. If two articles disagree, fix them before launch.
Step 3. Minimise personal data. Strip customer names, addresses and payment details from any article or sample conversation used in setup. Articles should describe policy, never individual customers.
Step 4. Set spend and handoff controls. Set a spending limit and an alert before going live. Then try to make excluded topics route to a person. If the trial shows topic-level handoff is not possible in your setup, keep Fin off the channels where urgent issues arrive.
Step 5. Run test questions. Fictional input: "Our grinder shows E4 and smells hot, what should I do?" Desired output: a brief acknowledgement, a statement that a technician needs to look at it, and a transfer, with no troubleshooting steps. Second input: "How long is the warranty on a refurbished espresso machine?" Desired output: the warranty term exactly as your article states it. Record what Fin actually produces next to the desired output.
Step 6. Human approval gate. A named owner reads every test transcript and signs off before customers see Fin. Refunds, credits, delivery promises and safety advice stay with staff throughout.
Step 7. Limited live run. Turn Fin on for one channel or a set of hours for an agreed period, such as four weeks. Each week, record:
- conversations Fin handled that later reached a person;
- reopened conversations;
- billed outcomes against your spending limit;
- a fixed sample of Fin answers checked by a person for factual errors;
- any excluded-topic conversation Fin answered instead of transferring.
If built-in reports do not show a measure, count it by hand from the sample.
Step 8. Stop or expand. Failure route: if Fin answers an excluded safety topic, states a wrong policy, or customers complain they could not reach a person, pause Fin on the affected channel or the entire pilot unless topic-level exclusion has been verified and fix the content or routing before retrying. Expand only when the reviewed sample shows no errors you would not accept from a new hire.
Copyable configuration brief
Use this as an internal brief for whoever sets up the pilot. It is not an Intercom settings file.
FIN PILOT BRIEF
Business: [name] Launch approver: [name]
In scope (each needs a current approved article):
1. [topic] -> [article link]
2. [topic] -> [article link]
Always reach a human:
- safety, injury, electrical, gas, water, medical
- refunds, credits, price exceptions, legal threats
- complaints and anyone asking for a person
Tone: plain, brief, no promises not found in an article.
If the answer is not in our content: say so, hand off.
Data rule: no customer personal data in knowledge content.
Spend: limit [amount] per month, alert at [percentage].
Trial checks: how handoff is set up and how it is billed.
Weekly review: [name] reads [number] Fin conversations.
Stop rule: any safety answer or wrong policy pauses that topic.
Costs and limits
Fin is listed at $0.99 per outcome on every helpdesk plan and on the standalone option. Fin Voice is available through sales at $1.99 per voice outcome (pricing). The page shows a dollar sign but does not name the currency, and taxes are not stated.
Helpdesk seat prices: numeric figures not verified. The retrieved text shows seat figures next to an annual and monthly toggle without making clear which figure matches which billing choice. Check the official pricing page. Intercom states a minimum of one Full seat.
Copilot is listed at $29 per agent per month billed annually, or $35 billed monthly. Pro is listed from $100 per workspace per month, including 500 conversation analyses and an included credit allowance that scales with conversation volume, with extra credits at $0.01 each. WhatsApp, SMS, phone and bulk email are charged by usage and vary by region or market (pricing).
Intercom lists a 14-day trial with no credit card that includes Fin and add-ons, plus an Early Stage startup programme with eligibility conditions (pricing). Read the terms before relying on either.
Who should skip it
Delay Fin if you have no maintained help content, if most conversations need account-specific judgement, or if volume is low enough that one person answers comfortably. Businesses giving regulated advice should get compliance review before any AI replies to customers.
FAQs
Will Fin replace my support staff?
In this pilot design, no. Fin acts as a first responder for a short list of topics. People keep escalations, judgement calls and anything touching money or safety.
How do I tell whether billed outcomes reflect real help?
Read a sample of billed conversations every week and compare them with reopens and complaints. A reviewer judging that a reply solved the problem is stronger evidence than the billing count.
Can I pilot Fin without leaving my current helpdesk?
Possibly, if your helpdesk is supported. Before planning around it, ask Intercom which controls and reports differ from the native setup.
Practical next action
This week, pull your ten most repeated questions from last month. Mark which have a current article and which involve safety or money. If at least five are covered and low-risk, fill in the brief above and start a trial. If not, write and approve those articles first.
Methodology: AI-assisted draft based on Intercom's official pricing page, retrieved 2 November 2026. The Fin product page was unavailable. The story and workflow are illustrative, not a hands-on test. ShiftToAGI is an independent publisher, not the vendor.