Fin vs. Zendesk vs. Gorgias vs. Freshdesk: The AI Support Agent Is Now Priced Four Different Ways

The decision rule

Before you compare any two of these prices, ask one question of each vendor: does the meter run when the AI tries, or only when it succeeds?

Zendesk bills only when the AI closes the ticket with no human involved; HubSpot's unit is a "resolved conversation," which its announcement does not define further. Freshdesk bills a session the moment the AI engages, whether or not it finishes. Fin bills one "outcome" per conversation, and a handoff completed through a configured Procedure counts as an outcome; the page does not say whether an ordinary escalation does, so ask. Gorgias bills any automation that closes a ticket, AI or not.

Pick the vendor whose unit matches your containment rate. If your AI will resolve most of what it touches, a per-attempt price is fine and usually cheaper. If it will hand off half of everything for the first six months, a per-attempt price makes you pay for the failures.

What each option costs

Prices and plan names are as published on each vendor's own pricing page; last verified 3 October 2026. This category re-prices often; verify before you sign.

Copilots for your human agents are a separate line on every one of these: Fin Copilot $35 per user/month, Zendesk Copilot $50 per agent/month yearly, Freshdesk Freddy Copilot $29 per agent/month annual (Pro and Enterprise only). Do not let a demo blur the agent price into the copilot price.

Five units, worked on the same month

Assume a 5-agent team handling 3,000 conversations a month. Assume the AI fully resolves 60% of them (1,800) and hands off 40% (1,200). Seats are on annual terms where a price is published.

Three things jump out. Freshdesk is cheapest, and at its worst case it ties HubSpot; that worst case is the realistic one, because a session starts when the AI engages, not when it wins. HubSpot's 3,000 included credits are per subscription, not per seat, and reset monthly, so they cover 60 resolutions a month and the rest is metered. And Fin's spread is entirely a definitional question you must ask before signing: does a conversation the AI hands to a human bill as an outcome? The pricing page lists "procedure handoff" as a billable outcome, so budget as if the answer is yes until a contract says otherwise.

The ownership changes you will be asked about

Two of these products changed hands in 2026.

Salesforce completed its acquisition of Fin on 10 September 2026. Fin sits inside Salesforce AI Labs and, per the press release, "will continue to serve its customers." The release describes a vendor-reported base of "more than 30,000 companies." Salesforce's 10-Q for the quarter ending 31 July 2026 put the deal at approximately $3.6 billion in cash. Update, 3 October 2026: the fin.ai pricing page still read $0.99 per outcome with the 50-outcome minimum, and said nothing about the Salesforce acquisition (Salesforce appears only as a supported helpdesk). Our tool page remains at /tools/intercom for now; the product you are buying is Fin.

Zendesk completed its acquisition of Forethought on 26 March 2026 (announced 11 March). The product is now "Forethought AI Agents by Zendesk" across chat, email and voice; no pricing was disclosed in the release, and it is not on the public Zendesk pricing page. If you are quoted a Forethought-branded agent, ask whether it bills on the same AR meter as the native one.

What ownership changes mean for a buyer is simple: get the per-unit price, the unit definition and the minimum in the contract, not the pricing page, because the page is now maintained by a different company than the one that wrote it.

When not to deflect

Update, 28 September 2026: California's governor approved AB 1609. It applies to businesses with more than $500,000,000 in gross annual revenue nationally and is operative 1 January 2027. A covered business's customer-service chatbot must carry "a clear and conspicuous disclosure that the customer service chatbot is artificially generated and not human" where a customer might reasonably think otherwise, and the business must make a good-faith effort to connect a customer who asks to a human agent within 15 minutes, or schedule one within a business day. Civil penalties are up to $5,000 for an initial violation and up to $10,000 for each subsequent one.

Most readers of this site are far below that revenue line. The design lesson still applies: an agent that cannot be escaped is a complaint generator, and every one of these vendors bills you for the resolution the customer did not want. Build the "talk to a person" path first, then measure containment.

What breaks in month three

The containment number stops meaning anything. Week one you read the transcripts of every "resolved" conversation. By month three nobody does, and "resolved" quietly drifts toward "customer gave up." The tell is a rising AI-resolution count alongside a rising reopen or repeat-contact rate. Zendesk and HubSpot bill on that resolution count, so drift costs money directly.

The knowledge base goes stale and the AI does not know. You change a return window or a pricing tier; the articles the agent reasons over still say the old thing. Every vendor here grounds on your help center. None of them tell you an article is contradicting your current policy.

The minimum bites. Fin's 50-outcome minimum is small for a 3,000-conversation team and large for a 150-conversation one. Gorgias's pricing page does not say whether included interactions roll over; ask. Read the minimums against your slowest month, not your average.

What to ask on the demo

• Show me, in the billing export, one conversation the AI handed to a human. Was it charged?
• What exactly resets monthly, what rolls over, and what is the minimum commitment on the AI line?
• If I turn the AI off for a week, what do I pay that week?
• Which plan tier is the AI gated to, and does that tier price change if I add an agent seat?
• Who owns the transcripts if we leave, and in what format?

Next step

Start with the ticket mix, not the vendor. Our support inbox triage workflow is the classification step that tells you what your containment rate can realistically be, and the RAG support agent workflow covers the grounding problem that decides whether "resolved" means resolved. For the full buying checklist, the free AI Support Agent Buyer's Guide walks each unit definition; ecommerce teams should also read the ecommerce support hub and our earlier helpdesk comparison. Head-to-heads: Freshdesk vs. Zendesk, Gorgias vs. Zendesk. Tool pages: Fin, Zendesk, Gorgias, Freshdesk, HubSpot.