AI Customer Service Cost per Resolution: Fin vs Zendesk vs Agentforce vs Copilot Studio

Every vendor selling an AI support agent now says you "only pay for results". Read the pricing pages and the billing docs side by side, and they mean six different things. Some bill a resolution, some a conversation, some every answer and action inside it. And the vendors that do bill resolutions don't agree on what a resolution is.

So we read each vendor's pricing page and billing docs on 27 September 2026 and priced one support queue on all of them. We haven't run a support bot on any of these platforms, and none of these vendors pay us. The shorter version, one 3,000-conversation example next to Zapier and n8n, is on our AI automation hub. This is the deep version.

What one resolved conversation costs

A resolved conversation costs $0.99 on Intercom Fin and about $1.50 on Zendesk at any resolution rate. On platforms that bill every conversation, the price per resolution depends on how often the bot succeeds: Agentforce's $2 conversation costs $2.50 per resolution at 80% resolved and $5.00 at 40%.

The formula is the monthly AI bill divided by the conversations the bot resolved. Our assumptions: three AI turns a conversation, each with 4,000 input and 300 output tokens; three Agentforce actions a conversation; one generative answer (2 credits) and one agent action (5 credits) a turn on Copilot Studio; Zendesk's Suite Team plan with five agents, which includes 25 automated resolutions a month; no Fin handoffs you've configured to bill. Zendesk sits just under $1.50 only because of those 25 included resolutions.

Two rows need a warning. The cheapest, the self-built agent, is model tokens only: you still build it, host it and fix it. And Agentforce Flex and Copilot Studio are cheap here because we assumed short conversations. A bot that looks things up, checks an order and updates a record runs more actions and credits per chat, and their bill grows with it.

The monthly bill at 1,000, 5,000 and 20,000 conversations

At 60% resolved, the AI part of the bill for 5,000 conversations a month ranges from $1,000 on Copilot Studio to $2,970 on Fin, $4,463 on Zendesk and $10,000 on Agentforce Conversations. Seats and helpdesk licenses come on top.

How each row is counted:

• Fin: resolved conversations times $0.99. On a helpdesk other than Intercom there's a minimum of 50 outcomes a month, so $49.50.
• Zendesk: resolutions above the 25 included, times $1.50 when you buy them ahead ("committed", sold in blocks of 100 or more). Pay-as-you-go overage is $2.00, which adds about a third.
• Agentforce Conversations: every conversation times $2.
• Freshdesk Freddy: every session times $0.49, from $49 per 100 sessions. Each paid plan gets 500 free sessions once per account, which we left out.
• Agentforce Flex Credits: three actions a conversation at 20 credits each, from $500 per 100,000 credits, so $0.10 an action.
• Copilot Studio: 21 credits a conversation, bought in whole packs of 25,000 credits for $200. At 20,000 conversations that's 420,000 credits, or 17 packs.
• Self-built: gpt-6-sol at $2 per million input tokens and $10 per million output is $0.033 a conversation. n8n Cloud Pro, €50 a month billed annually ($57.02), includes 10,000 executions, and n8n's own FAQ says a chatbot uses one execution per message. That covers about 3,333 three-turn conversations. Above that you'd self-host n8n and pay for a server we didn't price.

Why the Copilot Studio row is cheap per chat: Microsoft bills a generative answer at 2 credits and an agent action at 5, and one credit in a pack costs $0.008. Switch on tenant graph grounding and each message adds 10 credits, which more than doubles our figure. A reasoning model adds a premium rate per 1,000 tokens on top.

What each vendor counts as a resolution

The vendors that bill per resolution use different tests. Intercom counts a resolution when the customer confirms or simply leaves after Fin's answer. Zendesk counts one only when the conversation wasn't escalated and an LLM reviewing the transcript confirms the request was resolved.

In the vendors' own words:

• Intercom Fin. "A resolution is a type of outcome that is counted when, following Fin's last answer in a conversation, the customer either confirms the answer was satisfactory (confirmed resolution), or exits the conversation without requesting further assistance (assumed resolution)." Intercom's help article is candid about the edge case: "A customer leaving after receiving Fin's answer without requesting further help is counted as an Assumed Resolution and will be charged." Fin also bills a "Procedure handoff", when Fin "successfully executes a Procedure that you've configured to end in a handoff to a human or a workflow", at the same $0.99. It doesn't bill escalations from Fin's default behavior or your workspace escalation rules, and if a customer comes back to the same conversation for more help, "that resolution will be deducted and not charged."
• Zendesk. "Paying per automated resolution means that you pay only for customer requests that were successfully resolved by the AI agent, without any escalation to a human agent." On its current resolution tiers, a "Verified resolution" is one where "the customer did not request further assistance" and "a verification process is performed by an LLM that evaluates the text of the conversation to confirm that the customer's request was satisfactorily resolved." A conversation that fails that check is a "Contained resolution", which "does not count against your resolution allowance". A messaging conversation ends two hours after the last message by default, an email conversation 72 hours after the first email.
• Freshdesk Freddy. Bills sessions, not resolutions: "A session is a unique interaction between an end-user and an AI Agent." For email, a session "is defined as a 72-hour window starting from the customer's first email."
• Salesforce Agentforce. Offers two meters and you pick one per org: "Flex Credits and Conversations will not be supported in the same org." Each action is "a specific function that an AI agent executes on the platform, such as updating a record, summarizing a complex case, answering a product inquiry". The pricing page doesn't define what ends one billable conversation, and the help docs only load in a browser, so we can't quote a definition.
• Microsoft Copilot Studio. No resolution concept. Every answer and every action is billed in credits: "Agent actions refer to steps such as triggers, deep reasoning, and topic transitions".
• Self-built with n8n. No resolution concept either. You pay per execution and per token, and deciding what counts as solved is your own reporting job.

So "$0.99 per resolution" and "$1.50 per resolution" aren't the same unit. Fin's definition counts customers who give up. Zendesk's filters them through a second model. Which one costs less for you depends on how many of your customers leave quietly after a wrong answer, a number only your own transcripts can tell you.

Paying for failures

Platforms that bill per conversation, session, action or credit charge the same for a chat the bot fails as for one it solves. At 1,000 conversations a month and 40% resolved, $1,200 of Agentforce's $2,000 bill and $294 of Freshdesk's $490 pay for the 600 conversations a human still had to finish.

The Flex and Copilot Studio figures assume a failed chat uses as many steps as a solved one. A bot that fails fast uses fewer; one that loops before escalating uses more.

The flip side is where per-resolution pricing turns expensive: a good bot. Per conversation, Fin costs $0.99 times your resolution rate. That's the same as Freshdesk's $0.49 session at a 49.5% resolution rate, Agentforce Flex's three actions at 30%, and our Copilot Studio conversation at 17%. Above those rates, the per-step meters are cheaper per chat. Against Agentforce's $2 conversation, Fin is cheaper at any rate.

What you pay before the AI: seats and licenses

Most AI agents sit inside a helpdesk you pay for per agent seat. For a team of five agents on annual billing, the base ranges from $0 for Fin on a helpdesk you already have to $275 a month on Zendesk Suite Team.

Salesforce is the hard one to total. Agentforce runs inside a Salesforce org, and Service Cloud lists Starter Suite at $25 a user a month, Pro Suite at $100, Core at $195, Advanced at $395 and Max at $550, all billed annually. The Agentforce pricing page doesn't say which edition a customer-facing agent needs, so ask sales before you budget. Salesforce also notes its examples "do not include other costs like Data 360 credits or other consumption services."

Put together for five agents, 5,000 chats and 60% resolved: Freshdesk Growth with Freddy comes to about $2,545 a month, Intercom Essential with Fin about $3,115, and Zendesk Suite Team about $4,738. Copilot Studio's $1,000 and Fin-on-your-helpdesk's $2,970 still need a helpdesk underneath.

Zendesk has a moving part. Since 18 May 2026 it has been moving accounts to "resolution tiers" funded by a dollar allowance: $2 per Suite Team seat a month, $5 on Growth and Professional, $10 on Enterprise. Its public pricing page still shows $1.50 committed and $2.00 pay-as-you-go per automated resolution, and the tier docs say their example prices "are placeholders only, not the actual prices." If your account is on tiers, get the per-tier price from sales.

The self-built route

A self-built agent on n8n with a model API costs the least per resolution by far: about $0.033 a conversation in gpt-6-sol tokens, plus n8n's execution fee. What it doesn't include is the product. Fin, Zendesk, Agentforce and Copilot Studio come with the chat widget, the handoff to a human, the reporting and the guardrails. On n8n you build those.

Three things to know before you choose it:

• Executions add up per message. n8n bills one execution per workflow run, and for a chatbot its FAQ says to multiply conversations "by the average number of messages per conversation."
• Self-hosting is free only for your own use. n8n's Sustainable Use License allows use "for your own internal business purposes". You can't resell it as a hosted service.
• The model is the other lever. A small open model through an EU or US API costs a fraction of gpt-6-sol, and if customer data must stay in the EU, our private LLM cost guide prices every EU-resident route.

How to wire one up, with the agent nodes, is in our n8n AI agents guide, and the wider builder lane is on the AI agents hub.

How to choose

aliteq sells none of these products and earns nothing from these links. Every price here was read on the vendor's own page or docs on 27 September 2026.