Per-resolution AI pricing sounds fair, and often is. But it ties your bill to your busiest month. Here's what the big vendors charge and what to check.
It's the end of a good month. A sale went well, the phones were busy, and the chat on your website handled more questions than ever. Then the invoice for the AI arrives, and it's two or three times what you budgeted.
Nothing went wrong. That's the uncomfortable part. Under per-resolution pricing, a busier month is supposed to cost more, because you pay each time the AI handles a conversation for you. If you run a small business and you're comparing AI chatbot pricing, it's worth understanding exactly how that bill is built before you sign.
I build AI chat agents for a living, and we chose not to price this way, so read this with that in mind. I'll try to give the other side its due, because per-resolution pricing makes real sense for some businesses.
What the big vendors charge right now
Here's what the published pricing pages say at the time of writing (October 2026).
Intercom Fin charges from $0.99 per "Fin outcome", on top of help desk seats from $29 per seat a month, or with no seats if you use Fin with another help desk. Fin's own pricing page says you're charged for one outcome per conversation, and a resolution counts when "no further help is requested after Fin's last answer." Handoffs to a human through a configured procedure also count as an outcome. Sales qualifications are $9.99 each, and there's a minimum of 50 outcomes a month.
Zendesk bills AI agent usage in automated resolutions. Plans include a small allowance per human agent: 5 a month on Support Team and Suite Team, 10 on Suite Professional. Beyond that, the pricing page lists $1.50 per resolution on a commitment, or $2.00 pay as you go. Zendesk's help article explains that an LLM checks each conversation to confirm the request was really resolved, and that a messaging conversation ends 2 hours after the first message by default.
Salesforce Agentforce lists $2 per conversation for external customer agents, or Flex Credits at $500 per 100,000 credits, which are used up per action.
Why vendors price this way (and when it's fair)
The argument for per-resolution pricing is simple: you only pay when the AI does the job. If it can't help and a person has to step in, Zendesk doesn't charge for that conversation. That's a fair deal, and it puts pressure on the vendor to make the AI actually good.
It also fits businesses with steady, predictable volume. If you get roughly the same number of questions every month, the bill is roughly the same too, and you can compare it directly against the cost of a person answering those questions.
And to be fair, the vendors know bills can jump. Fin says you can set alerts and hard limits to control spend. Zendesk lets admins choose between letting the AI keep going and paying pay-as-you-go rates, or pausing the AI so requests go to live agents instead. Those controls exist. The question is whether you've set them.
Why per-resolution bills surprise small businesses
The surprise usually comes from three places.
Your busy months are your expensive months. A product launch, a holiday rush, a shipping delay or a website outage all bring more questions. Under per-resolution pricing, the AI handling that rush is exactly what drives the bill up, in the month you're already stretched.
Here's a simple illustration, not a real customer. Say your AI normally resolves 500 chats a month. A busy month brings 1,500. At Zendesk's pay-as-you-go rate of $2.00, that's the difference between $1,000 and $3,000. The price per chat didn't change. Your month did.
"Resolved" is the vendor's definition. Each vendor counts it differently: no further help requested, an LLM's judgment, a time window closing. None of these are unreasonable, but a customer who gave up and left can look a lot like a customer who got their answer. It's worth reading the definition before you agree to pay for it.
The default isn't always the one you'd pick. If you never open the usage settings, you may find out about overage the way most people find out: on the invoice.
How we handle it: no overage bills
At AtomicX we went the other way. Our AI plans run on a monthly allowance of AI credits. When the credits run out, the AI stops and your team gets the chats. No overage charges.
You get alerts at 80% and 90% of your allowance, so running out never sneaks up on you. If you'd rather keep the AI running, you can turn on optional auto-buy with a monthly cap you set. Without that, your bill is your plan.
There's a trade-off, and I want to be straight about it. When the allowance is used up, your team answers those chats themselves until the next month. For most small businesses I talk to, that's the right trade: a known cost, and a busy month that lands on your team's desk rather than your credit card. When a person picks up a chat, they get the transcript and an AI summary, so the customer doesn't have to start over.
The bigger pattern
Per-resolution pricing isn't a trick. It's one honest way to price AI, and for large teams with steady volume it can be the best one. The trouble is that it moves the risk of a busy month from the vendor to you, and small businesses usually have less room to absorb it.
Whichever tool you choose, ask three questions before you sign: what counts as a billable conversation, what happens when you go over, and which setting is on by default. If the answers are clear, the invoice should hold no surprises.
Want to see how our AI credits and alerts work on your own site? Book a 20-minute demo and we'll walk you through it. You can also read how credits work in our help centre.