AI agent pricing: per seat, per resolution, per conversation

TL;DR

• AI agent vendors charge for very different units: seats, resolutions or "outcomes," conversations, messages or credits, or a monthly plan with an allowance. The headline price means little until you know the unit.
• Per-seat pricing tracks your team; per-resolution tracks success; per-conversation and per-credit track usage. Each one shifts a different risk onto you.
• Model your own volume at two or three success rates before comparing vendors, and read the definition of the billable unit word by word.

Why AI agent pricing is hard to compare

Traditional software priced by seat: one person, one license. An AI agent breaks that model, because the agent does work that no seat represents, and every conversation has a real cost for the vendor in model usage, measured in tokens. Vendors have answered with different units, and as of October 2026 the market has not settled on one.

All prices below are list prices published by each vendor, checked on October 20, 2026. They change often, and several pages say so. Treat them as examples of each model, not as a ranking: these products differ in what they include, from a full help desk to a single website widget.

Model 1: per seat

You pay for each person on your team who uses the product, and AI features are included or sold as an add-on per user.

• Intercom lists its Essential, Advanced and Expert plans at $29, $85 and $132 per seat per month, with Fin, its AI agent, billed separately per outcome on top (Intercom pricing).
• Salesforce lists Agentforce add-ons for Sales, Service and other clouds at $125 per user per month (Salesforce).

Good: predictable; easy to budget. Watch out: seats count your staff, not your visitors. An agent that handles more conversations doesn't change the seat count, so per-seat pricing alone rarely captures what a customer-facing agent costs. That is why most vendors pair seats with a usage charge.

Model 2: per resolution or outcome

You pay only when the agent resolves a conversation without a person stepping in.

• Intercom charges $0.99 per Fin outcome, at most once per conversation. An outcome is counted when the customer confirms the issue is resolved, when they "don't ask for more help after Fin responds," or when Fin completes a workflow, including handoffs (Intercom pricing). Used with another help desk, Fin has a minimum monthly commitment.
• Zendesk bills AI agents in automated resolutions: requests "successfully resolved by the AI agent, without any escalation to a human agent." Plans include a few per agent per month (5 on Suite Team, 10 on Suite Professional); additional ones are listed at $1.50 committed or $2.00 pay-as-you-go (Zendesk pricing).

Good: you pay for success, not for attempts. The vendor has a reason to make the agent work. Watch out: the definition decides the bill. Intercom's help center explains that a customer who leaves after Fin's answer without asking for more help is counted as an "assumed resolution," even if they were unhappy, while explicit requests for a person, greetings and unanswered clarifying questions are not billed (Intercom Help). Costs also rise as the agent improves, and that is harder to forecast.

Model 3: per conversation

You pay for each conversation the agent takes part in, whatever happens in it.

• Salesforce lists Agentforce at $2 per conversation as one of its options (Salesforce).
• Tidio prices its Lyro AI agent by conversations, starting at $32.50 a month for 50, with the first 50 conversations free once. A conversation counts when it has at least one reply from the AI agent, and a customer returning to the same conversation still counts once (Tidio pricing).

Good: simple to understand and to count yourself. Watch out: you pay for conversations that end badly, including ones that go straight to a person. The definition of "one conversation" (time windows, returning visitors, channels) matters as much as the price.

Model 4: per message, action or credit

You buy a pool of credits, and each message, generated answer or action consumes some of them.

• Chatbase sells plans with monthly message credits, from $32 a month (billed yearly) for 700 credits to $400 a month for 15,000, with auto-recharge at $40 per 1,000 extra credits (Chatbase pricing).
• Salesforce sells Flex Credits at $500 per 100,000; a standard Agentforce action uses 20 credits and a voice action 30, so about $0.10 and $0.15 per action. Flex Credits and per-conversation pricing are not supported in the same org (Salesforce).
• Microsoft Copilot Studio sells capacity packs of 25,000 Copilot Credits for $200 per pack per month, with pay-as-you-go billing through Azure as an alternative (Microsoft).

Good: closest to the vendor's real costs; works for agents that do many different things. Watch out: forecasting. Different features burn credits at different rates, longer conversations cost more, and you need the vendor's usage reports to know where credits went.

Model 5: plan with a monthly allowance

A flat monthly fee that includes a set amount of usage, with extra usage sold in packs. Many of the vendors above combine this with another model: Zendesk includes automated resolutions in each plan, Chatbase and Tidio sell plans sized by credits or conversations.

Good: the most predictable bill; easy for small teams. Watch out: unused allowance usually doesn't roll over, and you need to know whether usage above the allowance is billed automatically or the agent stops answering.

For transparency: ASDK, which is in early access, uses this model. Plans include a monthly allowance of answers, and extra answers are sold in packs of 1,000. The current numbers are on the pricing page.

The models side by side

A worked example: how success changes the bill

Imagine a website agent handling 2,000 conversations a month, with an average of three billable actions each. Here is the usage cost under three models at the list rates above, at three resolution rates. It is illustrative only: it ignores seats, minimums, allowances and everything the products include beyond the agent.

Now divide by the number of resolved conversations, which is what you actually want to buy:

Three lessons fall out of this:

• Per-outcome pricing fixes the cost per success and lets the total grow with it.
• Per-conversation and per-action pricing reward a better agent: the same bill buys more resolutions as quality improves.
• Conversation length is the hidden variable for credit and action pricing. If the average rises from three actions to eight, that column more than doubles.

Run the same tables with your own volumes before you talk to vendors. The guide to measuring an AI agent explains how to estimate resolution and handoff rates honestly.

Questions to ask every vendor

Matching the model to your situation

• Low volume, small team: a plan with an allowance or a free tier keeps the bill flat while you learn what visitors ask.
• High-volume support with well-organized help content: per-resolution pricing can work well, if the definition is strict and you can audit it.
• Sales and lead generation: look for pricing tied to usage or plans rather than "resolutions," and judge the agent by meetings and qualified leads, as covered in AI sales agent for B2B websites.
• Voice: expect per-minute pricing on top of everything else; voice agents on websites covers those costs.

Every pricing model is reasonable for some business and expensive for another. Start from your own numbers: conversations per month, how many the agent can realistically resolve, and how long they run. Then price each vendor's unit against that, and buy the model whose risks you are comfortable carrying.