Per-Resolution, Credits or Flat Fee? How AI Chatbot Pricing Really Adds Up
Three ways AI support tools charge you, worked through at 300, 1,000 and 3,000 conversations a month, so you can see which one fits your volume.
Published Updated 6 min read
AI support tools charge in three main ways: per resolved conversation, per message credit, or a flat monthly fee with a message cap. Per-resolution pricing is easy to justify but grows in a straight line with your volume. Flat plans are cheap and predictable until you outgrow the cap. Credits sit in between and are the hardest to forecast.
I build ChatBento, which uses the flat-fee model, so I have a side in this. To keep it honest, this post does the math for all three with real published prices, and says where the per-resolution model is the better deal.
The three pricing models
1. Per resolution (or "per outcome"). You pay each time the AI settles a conversation. The best-known example is Intercom's Fin: $0.99 per outcome when you run it on your existing helpdesk, with a 50-outcome monthly minimum (checked October 22, 2026). Fin counts one outcome per conversation, and its billable outcomes include resolutions and some handoffs, not only fully solved chats. Conversations passed to your team without an outcome aren't charged.
2. Message credits. You buy a monthly allowance of credits, and each AI reply uses some of them. Chatbase works this way: its pricing page lists Hobby at $40 a month with 700 message credits, Standard at $150 with 4,000 and Pro at $500 with 15,000, plus auto-recharge at $40 per 1,000 extra credits (checked October 22, 2026). On credit systems in general, read how many credits one reply uses. Some tools charge more credits for replies from bigger models, and Chatbase's pricing page doesn't list per-model rates, so ask before you assume one credit equals one reply.
3. Flat fee with a message cap. You pay a fixed monthly price for a set number of AI replies. ChatBento is Free with 50 AI messages a month, $8 for 2,000, $30 for 10,000 and $100 for 40,000. When you hit the cap, the widget switches to a contact form instead of charging you overage.
First, count the right unit
Before comparing prices, convert everything to the same unit. The three models count different things:
- Per resolution counts conversations that end in an outcome.
- Credits count replies, sometimes weighted by model.
- Flat caps count replies (at ChatBento, one message is one AI reply).
A conversation usually has more than one reply. In the examples below I assume an average of 3 AI replies per conversation and that 60% of conversations end in a billable outcome. Your numbers will differ, so swap in your own. If you don't know them yet, a week of transcripts will tell you.
Worked example: 300, 1,000 and 3,000 conversations a month
| Conversations | Replies (×3) | $0.99/outcome, 60% billable | $0.99/outcome, all billable | Flat plan (ChatBento) |
|---|---|---|---|---|
| 300 | 900 | $178 | $297 | Hobby, $8 |
| 1,000 | 3,000 | $594 | $990 | Standard, $30 |
| 3,000 | 9,000 | $1,782 | $2,970 | Standard, $30 |
Hobby includes 2,000 replies a month and Standard 10,000. The credit model depends on how many credits a reply costs. If one reply is one credit, 900 replies a month would need Chatbase Standard ($150, 4,000 credits) because Hobby's 700 credits run out, and 9,000 replies would need Pro ($500, 15,000 credits).
Two things stand out. Per-outcome cost scales in a straight line: ten times the conversations is ten times the bill. Flat and credit plans move in steps, so you pay the same whether you use 3,000 or 9,000 replies on a given tier.
Why per-resolution pricing can still be the right call
Comparing cost per answer makes per-resolution pricing look expensive, but that isn't the whole story:
- You're buying a different product. Fin works with a full helpdesk (Intercom's own or one you already use) and can run configured procedures before handing off to your team. A website chat widget that answers from your pages is a smaller tool. If your support runs on Intercom and you need the AI to take actions, the per-outcome price buys things a $30 widget doesn't do.
- You only pay when it works, at least by the vendor's definition of an outcome. If your AI handles few conversations well, your bill stays small.
- It maps to labor cost. If a human answer costs you, say, $5 in staff time, $0.99 per outcome is still a saving. Big support teams think in cost per ticket, and this model speaks that language.
Where it hurts is predictability. A busy month, a launch or a pricing change that floods your chat with questions raises the bill at the same rate. Read the definition of an "outcome" closely too: an assumed resolution, where the visitor simply stops replying, may count.
Why flat fees suit most small websites
For a small business site, the questions are usually simple and repetitive: pricing, delivery, refunds, booking. That suits a flat plan:
- Predictable cost. You know the bill before the month starts.
- No penalty for success. If your chatbot gets popular, your price doesn't follow it up until you change tier.
- A safe ceiling. At ChatBento, when you hit the cap the widget stops writing AI answers and shows the contact form, so visitors can still reach you and nothing extra is charged.
The weakness is the cliff at the cap. If you regularly run past it, you need the next tier, and a plan that's always maxed out by the 20th isn't the bargain it looks like. Size the plan to your busiest month, not your average one.
How to pick in four steps
- Measure. Count chat conversations or contact-form submissions per month, and estimate replies per conversation. If you have no chatbot yet, use your support inbox volume as a starting point.
- Convert to one unit. Turn every vendor's price into cost per month at your volume, using its own unit: outcomes, credits or replies.
- Price your worst month. Run the numbers at double your normal volume. Per-outcome and credit plans can grow sharply here; flat plans don't change until you cross a cap.
- Match features to the price. If you need the bot to take actions in other systems or work inside a helpdesk, compare tools that do that, even if they cost more. If you need answers on your website plus lead capture and handoff, don't pay for a helpdesk platform you won't use.
Questions to ask any vendor
- What exactly is one billable unit: a reply, a conversation or an outcome?
- Does a reply from a premium model cost more?
- What happens at the limit: overage charges, auto-recharge, or the bot stops?
- Are test chats in the dashboard billed?
- Is there a monthly minimum, and are seats extra?
If a pricing page can't answer these, ask sales before you sign up. Surprise invoices come from the questions nobody asked.
Run your own numbers
The chatbot ROI calculator works out what an AI chatbot saves you in support hours and what it costs, using your own volume. ChatBento's plans, from the free tier to Pro, are on the pricing page.