Per resolution, per message, per reply: AI chat pricing compared
Intercom Fin, Zendesk, Gorgias, Tidio, Chatbase, SiteGPT, CustomGPT and Crisp bill for different events. Here is what each one charges for, what that does to your invoice, and when each model is the right one.
Eight tools, eight pricing pages, and no two of them billing the same event. This is not an accident of the market — the unit a vendor picks tells you what they want you to do more of. Here is what each one charges for, and what that does to an invoice at your volume.
Five units, one conversation
A visitor asks a question, gets an answer, asks two follow-ups and leaves happy. Nobody disputes what happened. What happens on the invoice depends entirely on which word the vendor chose:
- Outcome — one charge for the whole conversation. Intercom's definition includes a handoff to a human, so Fin bills even when it could not finish the job.
- Resolution — one charge, but only when the AI finished without a person. Zendesk states it that way. Cleaner in principle; the arguments are about the edges.
- Conversation — Tidio's Lyro counts any exchange it replied to at least once.
- Message or credit — every answer is billed. Four follow-ups cost four times one.
- Reply — ours. Every answer draws from a monthly allowance that is bought in advance, so a busy month costs nothing extra until the allowance runs out.
- Billed events in this conversation
- 1
- Price of one event
- $0.99
- This conversation costs
- $0.99
- 1,000 like it
- $990
Per-resolution pricing bills you more precisely as the assistant works better. A monthly allowance bills you the same either way. Neither is dishonest; they are bets on different futures.
The tools, with their own numbers
Two filters, because they are the two things that most often disqualify a tool outright: whether a person can take the conversation over, and whether it understands a Shopify order. Several tools have both, but only two or three plans up.
| Tool | Cheapest way in | Charges by | Hands off |
|---|---|---|---|
| Zinx Chat | $29 / moStarter plan, 25 free replies first | reply | Yes |
| Tidio Lyro | $24 / moStarter, 50 Lyro conversations | conversation | Yes |
| Chatbase | $40 / moHobby, 700 message credits | message | Yes |
| Crisp | $45 / moMini, 4 seats and $5 of AI credit | conversation | Yes |
| Intercom Fin | $50 / mo50 outcomes a month, on your own helpdesk | outcome$0.99 each | Yes |
| Gorgias AI Agent | $50 / moBasic, 300 tickets a month | resolution$0.90 each | Yes |
| Zendesk AI agents | $55 / moSuite Team, one agent | resolution$1.50 each | Yes |
| SiteGPT | $59 / moStarter, 4,000 messages | message | Higher plan |
| CustomGPT.ai | $99 / moStandard, 1,000 queries | query | Higher plan |
Their own published prices, read on 2026-10-07. Entry prices are the cheapest paid way in, which for several of them is billed annually. Follow the link on each name for what it says today.
The second price list
The headline number is rarely the whole bill. Three things reliably sit underneath it:
- Branding. Chatbase charges $99 a month to remove its badge from your widget — more than the widget itself on the entry plan. CustomGPT puts it on a $499 tier. Ours is on every paid plan, including the $29 one.
- Handoff. SiteGPT starts handoff on its second plan. Chatbase starts it on its third. If the assistant cannot reach a person on the plan you are buying, you are buying a dead end.
- The helpdesk underneath. Fin, Zendesk and Gorgias price AI as an addition to a seat-based product. If you do not already run that helpdesk, the AI price is the smaller half of the decision.
When each model is the right one
Per resolution suits a large support operation with predictable volume and an existing helpdesk, where the comparison is a salaried agent's hour and the finance team thinks in cost-per-ticket. It is the model most exposed to Gartner's prediction (2026) that a generative-AI resolution passes $3 by 2030.
Per message or credit suits low, spiky volume — a docs site where most visitors ask once. It punishes exactly the conversations you want: the long ones where somebody is deciding whether to buy.
A monthly allowance suits anyone who needs to know the number before the month starts. That is most small businesses, and nearly every one-person company. You pay for headroom you may not use; in exchange nothing about a good month costs you money.
What to test, rather than argue about
Every tool here will answer a delivery question well. The differences show up on the awkward questions, and you can find them in an afternoon:
- Ask it something your site genuinely does not answer. A good assistant declines and offers a person; a bad one invents a policy.
- Ask it to do something only a human can — a refund, an address change — and watch what happens next.
- Ask the same question three ways. Retrieval quality is more variable than model quality.
- Read a week of real conversations before you decide. The transcripts settle arguments that feature tables cannot.
If you want the measurement to be rigorous rather than impressionistic, that is a split test, and the discipline is the same as testing anything else on a storefront — our sister product Zinx Signal exists for that, and its write-up on measuring revenue rather than clicks applies here almost word for word: a chat widget that lifts engagement and lowers orders is not a win.
Ours is $29 a month for 600 replies, with handoff, branding removal and source citations on every paid plan, and a calculator for your own volume if you would rather see the arithmetic than take our word for it.
Common questions
- How much does Intercom Fin cost?
- Intercom Fin charges $0.99 per outcome, billed once per conversation, with a 50-outcome monthly minimum when run on another helpdesk. Inside Intercom it also needs a seat, from $29 a month.
- What is a good alternative to per-resolution chatbot pricing?
- Look for a published monthly allowance rather than a per-resolution rate, a free tier you can test with, handoff on the entry plan rather than two plans up, and no separate charge to remove the vendor's badge. Zinx Chat starts at $29 a month for 600 replies with all four.
- Is per-resolution pricing better than a monthly allowance?
- Per-resolution pricing is predictable per conversation but scales directly with success: the better the assistant works, the larger the bill. A monthly allowance caps the bill and makes a busy month free at the margin, but you pay for the allowance whether you use it or not.