How much does an AI agent cost? For a small business that wants one answering customers on its website and messaging channels, a hosted AI agent costs roughly $20 to $150 a month, and the price is decided far more by the pricing model than by the sticker. Platforms built for support teams charge per agent seat or per resolved conversation, which gets expensive fast for a business with no support desk. Building your own costs less in licence fees and far more in time, model bills and maintenance. This guide puts real numbers on every route, down to the cost per hour, per reply and per resolution, and gives you a way to check within a month whether the money came back.
Ultimo Bots is an AI agent for any website with a flat monthly price: from $19 a month on yearly billing, it answers from your own content around the clock on your site, Messenger, Instagram, Telegram and Slack, captures leads, hands over to a person, and connects to HubSpot, Mailchimp or a Zapier, Make or n8n workflow, with every channel and every action included in every plan.
See what an AI agent costs you
How much does an AI agent cost? The short answer
The phrase covers four very different purchases, and most of what ranks for it answers only one of them. Whether the product is sold as an AI agent, an AI receptionist or AI customer service software, the price ranges and the pricing models below are the same.
| Route | What you pay | What you also pay | Who it suits |
|---|---|---|---|
| Hosted AI agent | A flat monthly plan, roughly $20 to $150 for a small business, and quickly more once a support suite charges per seat and per resolution | An afternoon of setup, your own content | Anyone without a developer on staff |
| AI inside a support suite | A fee for every agent seat plus a fee each time the AI resolves a conversation | A bill that grows with your volume, so a busy month costs more than a quiet one | Teams that already run a help desk |
| Build it yourself with no-code tools | Tool subscriptions plus the AI model's usage fees, typically tens of dollars a month and rising with traffic | Your evenings, the upkeep, being on call when it answers wrong | People who enjoy the building |
| Custom development | Development agencies publish quotes from about $10,000 for a simple agent to six figures for larger systems | Hosting, model fees and a retainer for changes | Companies with unusual requirements and a budget line for them |
If you run a small business and the agent's job is to answer customers, capture enquiries and pass the hard ones to you, the first row is where the question ends. The rest of this guide is about reading that row correctly.
The four pricing models, and which one punishes a small business
Two tools at the same monthly price can cost you very differently after three months. What AI agents cost per month is decided by the model, and the model is what you are actually buying. These are the four AI agent pricing models you will meet:
- Per seat. You pay for each person who can log in. A business with one and a half people pays for a structure it does not have, and the price has nothing to do with how much work the AI does.
- Per resolution. You pay each time the agent settles a conversation, often around a dollar. It sounds fair, and it means a good month costs more than a bad one, so budgeting is guesswork until you have history.
- Per token or per call. The do-it-yourself model. You pay the AI provider for every word read and written, at the per-token rates the model vendors publish. Cheap at ten conversations a week, unpredictable at two hundred, and the bill arrives from a company that does not know your customers exist.
- Per message volume. A flat price for a capacity band, for example 500 or 2,000 AI replies a month. Predictable, and it does not punish the agent for being useful.
For a business without a support team, only the last one prices what you are buying. You want an agent, not seats, and you want to know in January what December costs.
What an AI agent costs per hour, per reply and per resolution
Price pages quote AI agents in different units, and each unit hides an assumption. Converting between them is plain division, so here it is with the inputs shown.
Per hour. A flat-plan agent is on around the clock, so the AI agent cost per hour is the monthly price over the 720 hours of a 30-day month: about 4 cents for Smart at $29, 8 cents for Boost at $59 and 21 cents for Ultimo at $149. Whatever an hour of the person answering your messages today is worth, nobody needs to quote a wage to see it is more than that.
Per reply. Divide the plan by its allowance: $29 for 500 AI replies is about 6 cents a reply, $59 for 2,000 about 3 cents, $149 for 10,000 about 1.5 cents. That assumes the allowance is fully used; use half of it and the cost per reply doubles.
Per resolution. Some platforms bill each time the AI settles a conversation: Intercom lists its Fin agent at $0.99 per outcome (checked on fin.ai, 27 September 2026). A flat plan has no such fee, so its cost per resolution is the plan price divided by the conversations it handled.
| Unit | Flat plan (Boost, $59 a month) | Per-resolution pricing ($0.99 per outcome) |
|---|---|---|
| Per hour, always on | About 8 cents ($59 / 720 hours) | Nothing while idle, then $0.99 for each resolved conversation |
| Per AI reply | About 3 cents ($59 / 2,000 replies) | Not billed per reply |
| Per resolved conversation of three replies, allowance fully used | About 9 cents ($59 / 666 conversations) | $0.99 |
A conversation is usually several replies, not one, so count the back-and-forth of a typical enquiry in your own inbox and multiply it by your monthly enquiries before you choose a plan.
One example: 200 conversations a month, four pricing models
Three numbers replace any AI agent cost calculator: conversations a month, AI replies per conversation, and people who need a login. Take a small business with 200 conversations a month at about three AI replies each (600 replies), 150 of them settled without a person, and two people who need a login. Competitor list prices were checked on the vendors' pricing pages on 27 September 2026.
| Model | What it bills | This month |
|---|---|---|
| Support suite, per seat plus per resolution | 2 helpdesk seats at $29 (Intercom Essential, billed annually) plus 150 Fin outcomes at $0.99 | $58 + $148.50 = $206.50, rising with every busy week |
| Hosted agent, flat capacity (Ultimo Bots) | 600 replies are more than Smart's 500 and fit inside Boost's 2,000 | $59, or $39 a month billed yearly |
| Build it yourself, per token | Model fees at the vendor's published rates plus tool plans | Usually the smallest line; the real cost is your hours |
| Custom development | A build fee plus hosting and a retainer | The running costs alone usually exceed every plan above |
Two things are worth reading out of that table. Intercom's $29 is its per-seat price on annual billing, so the like-for-like comparison is $206.50 against $39. And the lines move differently when business picks up: double the traffic to 400 conversations with the same share settled (300 outcomes) and the suite line becomes $58 + $297 = $355, while 1,200 replies still fit inside Boost at $59.
A support suite prices the work of a support team and earns its seats when you have one. When the support team is you, the flat line is the one that stays put; the details are in how Ultimo Bots compares with Intercom.
What an AI agent costs with Ultimo Bots
Most AI agent pricing sits behind a demo request. Ours does not, so here it is:
| Plan | Monthly | Yearly (per month) | AI replies per month | Knowledge sources |
|---|---|---|---|---|
| Smart | $29 | $19 | 500 | 200 |
| Boost | $59 | $39 | 2,000 | 1,000 |
| Ultimo | $149 | $99 | 10,000 | 10,000 |
Every plan includes every channel, every app action, live chat handoff, lead capture and analytics. There are no per-seat charges because you are buying the agent, not logins, and no per-resolution fee because a busy month should not be a penalty. Boost and Ultimo also re-read your website on a schedule (monthly and daily) and let you remove the Ultimo branding from the widget.
When a plan's monthly allowance is used up, nothing is metered quietly onto your invoice. You are warned before you get there, and you can either move up a plan or buy a one-time top-up pack, starting at $19 for 500 extra replies. Every plan starts with a 7-day free trial and can be cancelled any time. The full comparison, including skills and behaviours per plan, is on the pricing page.
The costs that are not on any price list
The subscription is the visible part. Four other costs decide whether an AI agent is cheap or expensive for you, and none of them appears in a pricing table.
- Your time to set it up. With a hosted agent this is an afternoon: point it at your website, add the documents customers ask about, ask it twenty real questions and fix the thin answers. With a self-built agent it is weekends, and it never quite ends.
- The quality of what it reads. An agent is only as good as your content. The first month of conversations will show you which page is unclear and which question you never answered on your site. Fixing those costs ten minutes each and is the best-value work in the whole project.
- What happens when it cannot answer. Automation with no exit turns a mildly annoyed customer into a former one, and that is the most expensive outcome of all. A handoff to a person, in the same window, with the history intact, has to be part of the price. With Ultimo Bots it is included in every plan through the built-in live chat, not sold as an add-on.
- Who is on call. If you build it, the answer is you, including the night it confidently invents your prices because a model update changed its behaviour. If you buy it, the answer is the provider. That difference is the real reason hosted agents exist.
Add one more line if your agent triggers work elsewhere: a Zapier, Make or n8n account has its own plan. A single workflow usually fits in their free or entry tiers, and it is worth knowing before you budget.
How much does it cost to build an AI agent?
Half of what you will read about AI agent cost is written by people who build them for a living, so the question deserves a fair hearing. A custom build pays for more than the code:
- Scoping and conversation design: what it answers, what it collects, when it hands over.
- Integration work per system: each CRM, calendar or shop is its own job, and its own upkeep.
- Model usage: tokens for every question, growing with your traffic.
- Hosting and retrieval: a server and a vector database to search your documents.
- Monitoring: logs and test conversations, because a model update can change answers overnight.
- The retainer: every change after launch, from a new price list to a new channel.
Agency cost guides published this year quote around $10,000 for a simple custom agent and five to six figures for larger systems, before monthly running costs. A no-code build swaps the build fee for tool subscriptions, model fees and your own evenings.
With a hosted agent, the build is a description: in Ultimo Bots you set up a skill by describing it in plain words to the build agent in the AI Studio, with no flowchart to draw and no code to maintain.
| Hosted agent | No-code build | Custom build | |
|---|---|---|---|
| Time to first answer | An afternoon | A weekend or two | Weeks to months |
| Monthly cost at small-business traffic | $19 to $149 | Tool plans plus model fees | Hosting, model fees, retainer |
| Handoff to a human | Included | You wire it | You specify it |
| Channels beyond the website | Included | One at a time, each its own project | Each scoped and billed |
| Who fixes it at midnight | The provider | You | The agency, on their terms |
Building makes sense when your requirements are genuinely unusual, when you want a system deeply woven into your own software, or when building is the point. For answering customers, capturing enquiries and handing over to a person, you are paying to reinvent something that costs less than a phone line.
What the agent does for that money
A price only means something next to the work. At $19 a month on yearly billing, the job Ultimo Bots does looks like this:
- Answers the questions you answer every week, in the customer's language, from your own website, documents and connected sources such as Google Drive, OneDrive or Notion.
- Captures the name, contact and request of anyone who wants something you cannot give in a chat, and notifies you immediately.
- Shares your Calendly link or books straight into Cal.com when someone asks for a time, and looks up an order in a connected Wix store after verifying the customer's email. For any other shop you can have it pass the order number to a Make or n8n workflow that answers back in the chat.
- Passes what it collected to HubSpot, Mailchimp or a Zapier, Make or n8n workflow, so the follow-up runs without you.
- Does all of it on your website, through a shareable chat link, and in Facebook Messenger, Instagram, Telegram and Slack, from one knowledge base.
How to know within a month whether it paid for itself
The arithmetic is simpler than any ROI calculator. Take the plan price and hold it against three numbers you can read yourself after thirty days.
Enquiries captured outside your hours. Each one sits in your leads inbox with the transcript attached, and each is a conversation you would not have had. If a single one of them is worth more than the plan, the agent is paid for, and for most service businesses one is.
Questions answered without you. Count the conversations that ended without anyone stepping in. Multiply by the minutes each would have cost you. That is the time the subscription bought back.
Where it handed over. Handoffs clustering on one question point at a missing page, not a failing agent. Ten minutes fixes it, permanently, and the next month's numbers improve on their own.
If after a month you see leads with timestamps from the evening, a list of questions you no longer answer by hand, and two pages you want to sharpen, the AI agent cost you a coffee a week and earned rather more than that. That is a realistic outcome for an afternoon of setup, and it is the only cost comparison that matters.