amBrain

Fin (formerly Intercom) alternatives: what the Salesforce deal changes

Published Sep 23, 2026Facts checked: Sep 23, 2026

Short answer

Stay on Fin if you want an AI agent answering on the helpdesk you already run (its pricing page promises "Set up in under an hour") and $0.99 per outcome is small next to your support payroll.[4] Consider an alternative if Fin's billing does not match how you measure success: it counts a resolution when "No further help is requested after Fin's last answer."[4] Consider one too if customer conversations must stay on your own infrastructure, or if Salesforce's ownership, completed on 10 September 2026, changes your vendor review.[3] The alternatives are Zendesk's AI agents, or your own agent on your helpdesk with a self-hosted model.

On this page

What changed in 2026

Fin's about page opens with "Founded as Intercom in 2011, we revolutionized modern customer communication." Its 2026 entries read "The company is renamed from Intercom to Fin." and "Transitions to in-house, purpose-built AI model, Apex 1.0."[1]

On 15 June 2026 Intercom's blog announced that "we just signed an agreement for Salesforce to acquire Fin for ~$3.6B".[2] Salesforce's press release of 10 September 2026 says it "has completed its acquisition of Fin, formerly Intercom", and that "As part of Salesforce's AI Labs, Fin will continue to serve its customers, advance its AI agent capabilities…"[3]

How Fin charges, and what it counts

Fin is sold two ways: on the helpdesk you already run, and together with the Intercom helpdesk. The two are priced differently, and which one you buy decides most of the bill.

On your existing helpdesk, fin.ai/pricing states "$ 0.99 per outcome" with a "50 outcomes per month minimum", and names Intercom, Salesforce, HubSpot, Freshworks, Dixa, Front, Zoho, Sprinklr and Gorgias as the helpdesks it works with.[4]

That option carries no seat charge at all. The pricing FAQ for a non-Intercom helpdesk reads "Unlimited teammates – No additional seat costs." and "There are no additional costs, such as integration fees, setup fees, or platform charges."[4] The seat prices below apply only to the Intercom-helpdesk option, which is where a seat-count argument against Fin applies.

With the Intercom helpdesk you pay for seats on top of outcomes, and the figure depends on how you are billed. On its default "Billed annually" toggle, intercom.com/pricing lists Essential at "$19 per seat/mo", Advanced at "$85 per seat/mo" and Expert at "$132 per seat/mo", each alongside "Fin AI Agent $0.99 per outcome". Billed monthly, the same three plans are $39, $99 and $139: the page's embedded schema.org Offer data gives "price":39 with "unitText":"SEAT_PER_MONTH" for Essential, and its FAQ says "you can move onto our Essential plan starting at $39 per seat per month." Copilot is a separate add-on at "$29 per agent/month billed annually, or $35 per agent/month billed monthly".[5] Budget from the column that matches your own billing term, not from the cheaper one. One caveat on the Essential seat price specifically: it is served from a live pricing experiment. The markup wraps it in data-experiment="team-self-serve-essential-price-reduction", and on the same day the two sites rendered different cohorts to us — variant=treatment on intercom.com/pricing showing "$19 per seat/mo", and variant=control on fin.ai/pricing showing "$ 29 per helpdesk seat per month".[5][4] So the bottom of the seat range is whichever figure your own session is allocated. Take the Essential number from your order form, not from either page.

Four outcome types are billed at the same price, not three: "Resolutions, Procedure handoffs, disqualifications, and self-serve routings are $0.99 each. Qualifications are $9.99 each." A resolution is defined as "Resolution: No further help is requested after Fin's last answer.", and a self-serve routing as Fin matching "a prospect to a routing outcome marked as Self-serve, for example, directing them to sign up for a trial." Billing is capped per conversation: "You're only charged for one outcome per conversation, even if Fin takes multiple actions.", and "You're not charged when a conversation is simply passed to your team without an outcome."[4]

Fin, Zendesk AI agents and your own agent

Three options, by what each vendor publishes today
FinZendesk AI agentsOwn agent (self-hosted model on your helpdesk)
What you are billed forFour outcome types at $0.99 each (resolution, Procedure handoff, disqualification, self-serve routing) and $9.99 per qualification; "You're only charged for one outcome per conversation, even if Fin takes multiple actions."[4]Automated resolutions, "the unit of measurement used for calculating and billing your account for AI agent usage", grouped into three tiers. "Assisted escalation" and "Contained resolution" each "does not count against your resolution allowance"; only "Verified resolution", defined as "The AI agent successfully resolved the interaction.", draws on it[7]Nothing per conversation; you pay for tokens or for GPU hours
When a conversation counts as over"Resolution: No further help is requested after Fin's last answer."[4]By channel: "Email: The conversation ends 72 hours after the last email in the conversation."; messaging ends 2 hours after the last message by default, "though this can be increased up to 72 hours"; "Voice: The conversation ends immediately upon hangup." An LLM then "evaluates the text of the conversation to confirm that the customer's request was satisfactorily resolved"[7]Your own definition, checked against a sample a person has labelled
Cost model$0.99 per outcome with a 50-outcome monthly minimum on other helpdesks, and no seat charge there.[4] With the Intercom helpdesk, seats are $19–$132 billed annually, or $39–$139 billed monthly, plus Copilot at $29 or $35 per agent a month[5]A pool, not a per-agent quota: "Your account is automatically credited with a resolution allowance, from which the cost of the automated resolutions is drawn." The tiers apply on "Team, Growth, Professional, Enterprise, or Enterprise Plus" plans[7]. The per-agent monthly allotments, the 10,000-a-year cap and overage pricing belonged to the model retired on 18 May 2026[8]Model inference plus engineering and operations; no per-conversation fee
Code and data ownership"Your customer data stays yours"; "Fin can use anonymized customer data for model fine-tuning, and you can opt out at any time with data deleted within 30 days. Third-party AI providers are contractually restricted from using your data for training, with zero data retention policies in place once an output is generated."[6]Governed by your Zendesk agreement; the resolution tiers doc does not cover data terms[7]Conversations, prompts and the model stay on your infrastructure; the code is yours
Time to launch"Set up in under an hour"; "During the 14-day trial, you'll have unlimited access to Fin outcomes, with no limits on usage."[4]Not stated in the resolution tiers doc[7]Scoped per project
Lock-in: term, notice, exit"Minimum commitments apply."[4] Procedures and knowledge sources are configured inside Fin and have to be rebuilt elsewhere; the owner changed to Salesforce on 10 September 2026[3]Not stated in the resolution tiers doc; the annual expiry of an allocation belonged to the model retired on 18 May 2026[8]No vendor contract; the dependency is your own team
Customisation and model choice"Customizable tone and answer length" and "Takes action to update external systems"[4]; the model is Fin's choice — "Fin uses OpenAI, Anthropic, Google, and Intercom models, switching automatically to maintain performance."[6]Configured inside Zendesk; not covered by the resolution tiers doc[7]Any model, prompt, tool, escalation rule and language
Compliance and data residency"Fin holds SOC 2 Type II, ISO 27001, ISO 27701, ISO 27018, ISO 42001, HIPAA, and GDPR/CCPA compliance."; "Data hosted in the US, EU, or Australia"[6]Not covered by the resolution tiers doc[7]Your infrastructure; the certifications are yours to obtain
Scale and availability"Fin resolves over 2 million conversations weekly"[1]; "backed by a 99.8% SLA"[6]Not covered by the resolution tiers doc[7]Your capacity, sized for your peak

Stay on Fin if…

  1. 1.You need an agent live on your current helpdesk now. Fin lists Intercom, Salesforce, HubSpot, Freshworks, Dixa, Front, Zoho, Sprinklr and Gorgias, promises "Set up in under an hour", and offers a trial where "During the 14-day trial, you'll have unlimited access to Fin outcomes, with no limits on usage."[4]
  2. 2.Your volume keeps the bill small. 1,000 outcomes a month is $990 at the published rate, and on a helpdesk other than Intercom there is no seat charge on top.[4] Your own agent does not pay back its engineering at that volume.
  3. 3.You want model choice handled for you. "Fin uses OpenAI, Anthropic, Google, and Intercom models, switching automatically to maintain performance."[6] Model retirements and provider outages are then Fin's problem, not yours; the same trade-off in your own build is laid out in OpenAI API vs a self-hosted LLM.

Consider leaving Fin if…

  1. 1.Symptom: your team disputes what counts as resolved. The invoice counts resolutions that your quality review does not. Fin bills a resolution when "No further help is requested after Fin's last answer",[4] so a customer who gives up and walks away is billed the same as one who confirmed the fix. Zendesk splits the same question three ways and charges only for the verified case.[7]
  2. 2.Symptom: security asks where the conversations go. A customer or a regulator wants transcripts on your own servers. Fin publishes "Data hosted in the US, EU, or Australia" and states that "Fin can use anonymized customer data for model fine-tuning", with an opt-out and data "deleted within 30 days".[6] No hosted agent meets an our-servers-only rule; AI on your own servers covers what replacing it involves.
  3. 3.Symptom: the bill grows in step with your support volume. Every campaign and product launch shows up on the invoice. At the published flat rate, 20,000 outcomes a month is $19,800,[4] before any Intercom seats.

A migration path in four steps

  1. 1.Export and label. Take out conversations, help-centre content and your procedures. Have a person mark a sample as resolved or not: that labelled set, not a vendor's counter, is how you judge the replacement.
  2. 2.Run the new agent in assisted mode. Let it draft replies that human agents approve inside your helpdesk — through its API, or on an open-source helpdesk such as Chatwoot or Zammad.[9][10] The shadow, assisted, automated sequence is described in the AI pilot that never reached production.
  3. 3.Automate one channel at a time. Start with the channel and language where the labelled set shows the highest confirmed-resolution rate, and keep the human handoff one click away.
  4. 4.Switch Fin off at the end of the commitment. The pricing page says "Minimum commitments apply."[4] Check yours, archive the transcripts, and stop outcome billing once the last channel has moved.

The two open-source helpdesks are licensed differently. Chatwoot's LICENSE separates the tree: "All content that resides under the "enterprise/" directory" carries its own terms, while "Content outside of the above mentioned directories … is available under the "MIT Expat" license".[9] Zammad is published under the GNU AGPL v3.[10] Read the boundary before you build on either.

What drives the cost of running your own agent

The model is the small part. Assume a support conversation of 3,000 input and 500 output tokens.

On OpenAI's gpt-5.6-luna, at $0.20 per million input tokens and $1.20 per million output tokens,[11] that conversation costs about $0.0012 — $0.0006 of input plus $0.0006 of output. Self-hosted it is a share of GPU time instead: Azure's retail price for one Standard_NC40ads_H100_v5 (H100 NVL) in East US is $6.98 an hour.[12] Either way the model call is three orders of magnitude below the $0.99 outcome price. The difference pays for everything around the model: retrieval over your help content, the helpdesk integration, evaluation, actions in other systems and operations, which is the work an own agent consists of.

Cost drivers, not a price
Cost driverWhat drives itPublic reference point
Model inferenceConversations per hour at peak, languages, context lengthgpt-5.6-luna at $0.20 and $1.20 per million tokens[11]; or self-hosted with vLLM, Apache-2.0,[13] on GPUs from $6.98 an hour[12]
Retrieval over help contentArticles, policies and product data kept currentNo public reference; scoped per project
Helpdesk integrationReading tickets, writing replies, handing off to a humanYour helpdesk's API, or Chatwoot[9] or Zammad[10]
Evaluation and quality reviewLabelled conversations, rerun on every changeNo public reference; scoped per project
Actions in other systemsRefunds, order changes, account updates behind permissionsNo public reference; each integration is scoped
OperationsMonitoring, incident response, model upgradesNo public reference; scoped per project

Frequently asked questions

Yes. Fin's about page says "Founded as Intercom in 2011, we revolutionized modern customer communication." and records for 2026 that "The company is renamed from Intercom to Fin." The same page notes the move to an in-house model: "Transitions to in-house, purpose-built AI model, Apex 1.0."[1]

Yes. Intercom's blog announced on 15 June 2026 that "we just signed an agreement for Salesforce to acquire Fin for ~$3.6B".[2] Salesforce's press release of 10 September 2026 says it "has completed its acquisition of Fin, formerly Intercom", and that "As part of Salesforce's AI Labs, Fin will continue to serve its customers, advance its AI agent capabilities…"[3]

One outcome per conversation, in one of five kinds. "Resolutions, Procedure handoffs, disqualifications, and self-serve routings are $0.99 each. Qualifications are $9.99 each." A resolution means "No further help is requested after Fin's last answer", and a self-serve routing means Fin matched "a prospect to a routing outcome marked as Self-serve, for example, directing them to sign up for a trial." "You're only charged for one outcome per conversation, even if Fin takes multiple actions", and "You're not charged when a conversation is simply passed to your team without an outcome."[4]

Outcomes stay at $0.99, and seats are added. On the default "Billed annually" toggle intercom.com/pricing shows "$19 per seat/mo" for Essential, "$85 per seat/mo" for Advanced and "$132 per seat/mo" for Expert; billed monthly the same plans are $39, $99 and $139 a seat, and the page's FAQ puts it as "you can move onto our Essential plan starting at $39 per seat per month." Copilot is an extra "$29 per agent/month billed annually, or $35 per agent/month billed monthly".[5] The Essential figure is inside a live pricing experiment (data-experiment="team-self-serve-essential-price-reduction"), and on the same day intercom.com/pricing served us "$19 per seat/mo" while fin.ai/pricing served "$ 29 per helpdesk seat per month", so confirm that one on your order form.[5][4] On any other helpdesk there is no seat charge: "Unlimited teammates – No additional seat costs."[4]

Since 18 May 2026, from an allowance rather than a per-agent quota: "Your account is automatically credited with a resolution allowance, from which the cost of the automated resolutions is drawn." Resolutions fall into three tiers — "Assisted escalation" and "Contained resolution", each of which "does not count against your resolution allowance", and "Verified resolution", where an LLM confirms the request was satisfactorily resolved.[7] The older per-agent allotments with a 10,000-a-year cap describe the model retired on that date.[8]

Its trust page says "Fin can use anonymized customer data for model fine-tuning, and you can opt out at any time with data deleted within 30 days. Third-party AI providers are contractually restricted from using your data for training, with zero data retention policies in place once an output is generated."[6]

Yes: a self-hosted model, retrieval over your help content, and your helpdesk's API, with a person approving replies until the labelled set says otherwise. AI on your own servers covers what to decide first, and vLLM vs Ollama covers what to serve it with.

About amBrain

  • We have taken an LLM integration to production inside a client's FinTech perimeter: extracting and normalising unstructured broker and venue notices — corporate actions, instrument and margin changes — into structured records the trading system consumes.
  • The client keeps full ownership of the product and the code, except our reusable components.

Disclosure: this page is published by amBrain. The last column of the comparison table is the kind of work amBrain does. Every vendor fact on this page links to its source and was read on the date shown there; prices quoted from Intercom are the ones its pricing page renders on its default annual-billing toggle. Fin, Intercom, Salesforce, Zendesk, Chatwoot, Zammad, OpenAI, Anthropic, Google, Microsoft Azure and vLLM are trademarks of their owners and are used only to identify the products discussed. No vendor named here reviewed or endorsed this page.

Sources

  1. [1]Fin, About (founding, the rename, the move to Apex 1.0, and weekly conversation volume). Vendor's own page · checked Sep 23, 2026
  2. [2]Intercom, Salesforce signs definitive agreement to acquire Fin, Jun 15, 2026. Press · checked Sep 23, 2026
  3. [3]Salesforce, Salesforce Completes Acquisition of Fin, Sep 10, 2026. Press · checked Sep 23, 2026
  4. [4]Fin, Fin AI Agent Pricing (outcome types and prices, monthly minimum, supported helpdesks, trial terms, and the pricing FAQ). Vendor's own page · checked Sep 23, 2026
  5. [5]Intercom, Pricing (plan cards on the default annual toggle, the embedded schema.org Offer data for monthly billing, and the Copilot FAQ; the Essential seat price is served from the experiment wrapper data-experiment="team-self-serve-essential-price-reduction", so it is cohort-dependent — read raw, not through a summariser). Vendor's own page · checked Sep 23, 2026
  6. [6]Fin, Trust and Reliability (model providers, data hosting, the fine-tuning opt-out, the SLA and the certification list). Vendor's own page · checked Sep 23, 2026
  7. [7]Zendesk, About automated resolution tiers (the current model, introduced 18 May 2026: the resolution allowance and the three tiers). Official documentation · checked Sep 23, 2026
  8. [8]Zendesk, About the automated resolutions platform (prior to resolution tiers) (the model retired on 18 May 2026: per-agent allotments, the 10,000-a-year cap and annual expiry). Official documentation · checked Sep 23, 2026
  9. [9]Chatwoot, Source repository and LICENSE. Vendor's own page · checked Sep 23, 2026
  10. [10]Zammad, Source repository and README. Vendor's own page · checked Sep 23, 2026
  11. [11]OpenAI, API pricing. Official documentation · checked Sep 23, 2026
  12. [12]Microsoft Azure, Retail Prices API: Virtual Machines, East US, H100 sizes (JSON response; Standard_NC40ads_H100_v5 (H100 NVL) at $6.98 per hour). Vendor's own page · checked Sep 23, 2026
  13. [13]vLLM, Source repository and README (Apache-2.0). Vendor's own page · checked Sep 23, 2026

Related

Free project intro call, 30 minutes.

We reply within 24 hours.

Book a Call