How Much Does an AI Chatbot Cost? Per-Resolution Tools vs Custom
What an AI chatbot costs, how per-resolution and per-conversation pricing really compare, and the volume where a custom chatbot starts to pay back.
By Sameer Rahman- Updated
- 8 min read
On this page
Most AI chatbot cost comparisons stop at the price list. Fin charges $0.99 a resolution, HubSpot charges $0.50, Botpress charges $750 a month. But those prices count different things, and they change in different ways as your chatbot gets better. A cheaper-looking option can end up costing you more.
This guide shows how to compare them properly. You'll see what each pricing model really costs at your resolution rate, how to work out the most your chatbot could ever resolve before you buy anything, and the volume where a custom build starts to make sense.
Key takeaways
- Per-resolution tools list about $0.50 to $2.00 per resolved conversation. Custom builds run about $15,000 to $60,000 in engineering time.
- Per-resolution bills rise as your chatbot improves. Per-conversation plans do the opposite, and cross over at around a 50% resolution rate in our example.
- Your ticket mix sets the ceiling. Sorting 200 tickets tells you more than any vendor's resolution rate.
- Connecting account data is usually worth more than better answers, because it opens up a whole group of tickets that documents can't answer.
- A $20,000 custom build breaks even around 1,550 resolutions a month against a $0.99 tool, in our example.
How much does an AI chatbot cost?
An AI chatbot costs about $0.50 to $2.00 per resolved conversation on a per-resolution tool, up to about $750 a month on a builder plan, or about $15,000 to $60,000 for a custom build plus running costs. Which one is cheapest depends mostly on your volume and on what kinds of questions you get.
An AI chatbot here means a chat assistant that answers from your own content and can take actions, like opening a ticket or checking an order. These are list prices as of November 7, 2026.
| Option | Listed price | What it counts |
|---|---|---|
| HubSpot customer agent (opens in a new tab) | $0.50, needs a Professional or Enterprise edition | Each resolution |
| Gorgias (opens in a new tab) | $0.90 on annual plans, $1.00 on monthly plans | Each resolved interaction |
| Fin (opens in a new tab) | $0.99, with a 50-outcome monthly minimum | Each outcome |
| Zendesk (opens in a new tab) | $1.50 committed, $2.00 pay-as-you-go, on top of a per-agent plan | Each automated resolution |
| Botpress (opens in a new tab) | $150 or $750 a month, billed annually | Each conversation, with 250 or 1,500 included |
| Dify (opens in a new tab) | $590 or $1,590 a year | Message credits, or your own model key |
| Custom build that answers from your help content | About $15,000 to $22,000, or 4 to 6 weeks of engineering | One-time, plus running costs |
| Custom build connected to orders, tickets, and your CRM | About $45,000 to $60,000, or 3 to 4 months of engineering | One-time, plus running costs |
The custom build rows are engineering time at about $93 an hour. That's the median US software developer pay (opens in a new tab) with benefits added (opens in a new tab).
Per-resolution and per-conversation pricing move in opposite directions
This is the difference most comparisons miss. A per-resolution tool charges you more as your chatbot gets better. A per-conversation plan charges the same either way, so each resolution gets cheaper as your chatbot improves.
Take 2,000 conversations a month. Here's Fin at $0.99 against the Botpress Team plan, which costs $750 and adds $50 for each extra 100 conversations, so $1,000 at this volume.
| Resolution rate | Resolutions | Fin | Botpress Team | Botpress per resolution |
|---|---|---|---|---|
| 40% | 800 | $792 | $1,000 | $1.25 |
| 50% | 1,000 | $990 | $1,000 | $1.00 |
| 60% | 1,200 | $1,188 | $1,000 | $0.83 |
| 70% | 1,400 | $1,386 | $1,000 | $0.71 |
So the plans cross at about a 50% resolution rate. Below that, paying per resolution protects you, because you don't pay for the conversations it can't solve. Above that, you're paying more for each gain. That's why per-resolution is a good way to start, when you don't know your rate yet, and worth checking again once your rate settles.
What's actually being counted
Fin also counts an assumed resolution (opens in a new tab), which is when a customer leaves without asking for more help. Some of those people got their answer, and some gave up. That changes what you're paying. If 30% of your resolutions came from people who gave up, your real price is about $1.41 for each customer helped, not $0.99.
You can check this in an afternoon. Pull 100 assumed resolutions and see how many of those customers contacted you again about the same thing within a week. It also gives you a truer resolution rate to use when you compare options.
Your ticket mix sets the ceiling
Vendors publish resolution rates, but what you can reach depends on what your customers ask. You can find your own ceiling before you buy anything. Take your last 200 tickets and sort them into three groups.
| Group | Example | What it takes |
|---|---|---|
| Answerable from your docs | "What's your refund policy?" | Clean help content |
| Needs account data | "Where's my order?" | A connection to your systems, with identity checks |
| Needs judgment | "I want an exception to the policy" | A person, with the conversation passed along |
Say that gives you 90, 70, and 40 tickets. A chatbot that only reads documents tops out at 45% no matter how good the model is, and a real one will land somewhat below that. Connect your account data and the ceiling goes to 80%.
At 2,000 tickets a month, that's 700 more tickets the chatbot could take. If a ticket costs you $4 in staff time, say eight minutes at $30 an hour, that's up to $2,800 a month. This is usually a bigger gain than switching models or tools. So ask this about any option you're considering, whether it can connect to your systems, what that takes, and who maintains it.
The mix also tells you what to fix first. If the first group is small, a chatbot won't do much until the docs improve or your account data is connected. If the first group is large but the bot still misses, your content is the problem, and buying a different tool won't solve it.
What a custom build changes
A custom build swaps a per-resolution fee for a fixed monthly cost. The model itself is a tiny part of that. At Claude's list prices (opens in a new tab), a four-question conversation that reads 3,000 tokens of your content per answer and writes 200 costs about $0.002 on Haiku 5.5 and about $0.03 on Sonnet 5.5. A token is a small chunk of text, roughly part of a word.
So almost everything you pay a per-resolution tool goes toward the product and the work around it. A custom build means you take on that work, or pay someone to. That's only worth it once the fees are big enough.
The volume where custom pays back
Here's a way to find that point. Take a $20,000 build spread over 24 months, $500 a month for upkeep, and a Dify Team plan at about $133 a month, which comes to about $1,466 a month in fixed costs. Say each resolution takes 1.5 conversations on average, at Sonnet 5.5 prices, so about $0.05 per resolution in model fees.
| Replacing a tool at | Break-even volume |
|---|---|
| $0.50 per resolution | About 3,250 resolutions a month |
| $0.90 per resolution | About 1,700 resolutions a month |
| $0.99 per resolution | About 1,550 resolutions a month |
| $1.50 per resolution | About 1,000 resolutions a month |
| $2.00 per resolution | About 750 resolutions a month |
The formula is your monthly fixed cost divided by the fee you'd avoid, minus your model cost. Below your break-even number, a tool is cheaper on cost alone. Above it, each resolution saves you close to the full fee, so the savings add up quickly. If you need a help desk plan either way, this comparison stays the same.
Plug in your own build quote. Double the build and upkeep and the break-even roughly doubles as well, so a $50,000 project needs much more volume to make sense on cost alone.
A per-resolution tool fits when
- You're under your break-even volume
- Most tickets can be answered from your docs
- Your systems are already supported by the tool
- You don't yet know your resolution rate
A custom build fits when
- You're well past your break-even volume
- Much of your volume needs account data the tool can't reach
- You need your own rules for what it can say and do
- You want to own the accounts, content, and logs
Common mistakes when budgeting for a chatbot
- Comparing the price per resolution without the rate. A $0.50 tool that resolves 30% can be worth less than a $0.99 tool that resolves 60%, because every missed ticket still costs you staff time.
- Treating a failed conversation as free. A customer who fought with the bot arrives annoyed, and your agent has to read the whole thread. Make sure handoffs carry the full conversation so that the agent doesn't start over.
- Testing on your top 10 questions. Those are the ones any bot gets right. Test on a random sample of real tickets, since that's what your resolution rate will come from.
- Skipping the content work. Old and conflicting docs lower your ceiling in any tool. Fix these first and every option gets better.
- Forgetting to check again. The right pricing model changes as your rate and volume grow. Put a date on the calendar to redo the math in six months.
Conclusion
The right chatbot depends on more than the listed price. Sort your tickets to find your ceiling, compare options at your real resolution rate, and work out where a custom build breaks even. Most teams under that volume should start with a tool, and most teams well over it should at least price a custom build.
We build AI chatbots on tools like Botpress, Langflow, and Dify, in your own accounts, connected to your CRM and help desk. Every project is scoped and quoted after a discovery call. If you're comparing builders, see the best AI chatbot development agencies.
Frequently asked questions
Per-resolution tools list about $0.50 to $2.00 for each resolved conversation, so 1,000 resolutions costs $500 to $2,000 plus any plan you need. Builder plans run from free to $750 a month or more, and a custom build adds a one-time fee on top.
Most of the price is engineering time. At about $93 an hour, a chatbot that answers from your help content is about $15,000 to $22,000, and one connected to orders, tickets, and your CRM is about $45,000 to $60,000. How messy your content is and how many systems it connects to matter most.
In our example, a $20,000 build breaks even at about 1,550 resolutions a month against a $0.99 tool, about 3,250 against a $0.50 tool, and about 1,000 against a $1.50 tool.
It depends on your ticket mix. Sort a sample of your tickets by whether they can be answered from your docs, need account data, or need judgment. The first group is your ceiling for a chatbot that only reads documents.
Usually not. A four-question conversation costs about $0.002 to $0.03 in Claude model fees at list price, against about $1 per resolution on many tools. Most of what you pay for is the product and the work around it.
Sources
- Fin, Pricing (opens in a new tab), read November 7, 2026.
- Intercom, Fin AI Agent outcomes (opens in a new tab), read November 7, 2026.
- HubSpot, Customer agent (opens in a new tab), read November 7, 2026.
- Gorgias, Pricing (opens in a new tab), read November 7, 2026.
- Zendesk, Pricing (opens in a new tab), read November 7, 2026.
- Botpress, Pricing (opens in a new tab), read November 7, 2026.
- Dify, Pricing (opens in a new tab), read November 7, 2026.
- Anthropic, Claude pricing (opens in a new tab), read November 7, 2026.
- U.S. Bureau of Labor Statistics, Software developers, quality assurance analysts, and testers (opens in a new tab), Occupational Outlook Handbook, read November 10, 2026.
- U.S. Bureau of Labor Statistics, Employer Costs for Employee Compensation, June 2026 (opens in a new tab), released September 9, 2026, read November 10, 2026.