Salesforce launched Agentforce at $2 a conversation. It priced out smaller organizations and felt expensive for a quick question. So it added credits at $0.10 per action. Then per-user add-ons. Then an all-you-can-eat contract.
Three pricing models in about a year. If Salesforce needed that many attempts, you deserve a calculator.
This one compares the four models that matter for an AI agent, using your own numbers.
How to use the AI agent pricing calculator
- Describe the work: conversations per month, the share your agent resolves without a human, and what each conversation costs you in model tokens, search and hosting.
- Enter the customer's side: what a human-handled ticket costs them, and how many human seats the agent replaces. Ask your customer for real figures; the pre-fill is illustrative.
- Set a price for each model: per AI seat, per conversation, per resolution, and a hybrid of a platform fee plus a smaller per-resolution charge.
- Pick a stress-test rate, the resolution rate in a bad month. The table shows your margin at both rates.
How the maths works
Resolutions = Conversations × resolution rate Your cost = Conversations × cost per conversation (you pay for failures too) Customer value = Resolutions × human cost per ticket Per seat revenue = Seats replaced × seat price Per conversation revenue = Conversations × price Per resolution revenue = Resolutions × price Hybrid revenue = Platform fee + Resolutions × price Your margin = (Revenue − Your cost) ÷ Revenue Customer saves = Customer value − Revenue
The chart redraws your margin for each model across resolution rates from 10% to 100%, so you can see which one breaks first.
A worked example
The pre-fill follows the worked example in our AI agent pricing guide: 5,000 conversations a month, 65% resolved, at about $0.15 a conversation on a $2 / $10 model.
At Fin's $0.99 per resolution, the customer pays about $3,217 for 3,250 resolutions. Your cost is $750, so your margin is 77%. Now drop the resolution rate to 40%. Revenue falls to $1,980 but the cost doesn't move, and your margin falls to 62%.
The hybrid ($1,000 platform fee plus $0.60 a resolution) starts lower at 75% but only slides to about 66% in the bad month. That's the platform fee doing its job.
How to read the result
Two tests, one for each side of the table.
Does the customer win? "Customer saves" should be clearly positive. If the agent costs more than the people it replaces, nobody signs. Aim for the customer keeping several times what they pay you.
Do you win in a bad month? Look at the stress column. Per-resolution pricing is the easiest to sell (Fin's pitch works because a support manager can compare $0.99 with a human ticket), but it puts all the risk on you. Per conversation puts it on the customer, which is why Decagon says most of its customers choose per-conversation pricing.
And watch the market. In April 2026 HubSpot cut its Customer Agent from $1 per conversation to $0.50 per resolved conversation. Type $0.50 into the per-resolution box and see whether your margin survives a competitor's announcement.
My own preference is the oldest trick in the book (Model #8, minimum commitment): a platform fee that covers acquisition and onboarding, with outcomes on top where they can be counted cleanly.
What to do next
- Get your real cost per conversation with the AI Token Cost Calculator.
- Read all six agent pricing models with real 2026 prices in AI agent pricing.
- See how Intercom bet the company on 99 cents per resolution.
- Define your outcome properly before you charge for it: outcome-based pricing.
- Building agents for clients? See how much to charge for building AI agents.
Frequently asked questions
What is a good price per resolution for an AI agent?
Published 2026 prices give a range: Fin charges $0.99 per outcome, HubSpot $0.50 per resolved conversation and Zendesk $1.50 committed or $2.00 pay-as-you-go per automated resolution. The right number sits well below your customer's cost of a human-handled ticket and well above your cost per conversation divided by your resolution rate.
Why does my resolution rate matter so much?
With per-resolution pricing you pay model costs on every conversation, but you're only paid for the ones the agent resolves. If your resolution rate falls from 65% to 40%, your revenue drops by more than a third while your costs stay the same. That's why the calculator shows a stress-test column next to your expected margin.
Should I charge per seat for an AI agent?
Seat pricing is easy to buy and predictable for both sides, but it has a flaw for agents: the better your agent works, the fewer human seats your customer needs. Charging per AI seat replaced can work for staff-facing agents, but revenue won't grow with volume. Many vendors pair a seat with usage instead.
What counts as a resolution?
Define it in your contract before you charge for it. Fin counts an outcome when the customer confirms the answer helped, doesn't ask for more help after the response, or Fin completes a defined workflow. Whatever you choose, it must be measurable, attributable to your agent, and something your customer already tracks.
How do I protect my margin if model prices rise?
Route simple questions to a small, cheap model, cache the instructions you resend on every call, and keep a platform fee or minimum commitment in your contract. Then rerun this calculator with a higher cost per conversation. If the stress-test margin is still above about 60%, your pricing can absorb a vendor price change.
Built on The Subscription Playbook
This tool turns frameworks from Robert Coorey's book into something you can use in minutes. Robert is the author of The Subscription Playbook and a Director at BPAG Accountants. More about Robert.