Sunday, 9:15 PM
Three price quotes are open on the screen. The first has a single line: "AI agent for your business, everything included." The second has four pages of technical terms. And the third is half the price, but in small print it says "does not include third-party costs." The business owner has been trying to compare them for an hour.
The problem isn't the price. The problem is that you can't tell what you're buying.
"How much does an AI agent cost?" is a fair question, but the real answer starts with another one: what exactly are you paying for. An AI agent isn't an off-the-shelf product with a price tag. It's a one-time setup, a few services you pay for every month, and sometimes ongoing support. Once you understand the parts, you can compare proposals, spot an inflated one, and see what will change when your usage grows.
In short
- How much an AI agent costs depends on the tasks, the number of systems you connect and the scope of support. There is no single price that fits every business.
- The cost has three parts: a one-time setup, monthly external services (an AI model, a server, tools), and support and maintenance, which are usually optional.
- There are also costs that don't appear in the proposal: your time for approvals, updating the instructions and knowledge, and swings in usage.
- A good proposal breaks down every part. A proposal with no breakdown, or one that asks for your personal passwords, is a red flag.
Why there is no single price for an AI agent
Think of a kitchen renovation. No serious contractor will give you a price over the phone before seeing the kitchen. There are cabinets, plumbing, electrical work, and whatever turns up only when you open the wall. And even after the kitchen is done, someone has to pay for the water and electricity.
An AI agent works the same way. The setup is the renovation: connecting systems, defining permissions, teaching the agent your business. The external services are the running bills: the model that does the thinking, the server the agent runs on, and additional tools as needed. Support is the person who fixes it when something leaks, and improves it when you want to add something.
And just as with the kitchen, the difference between proposals isn't only in the bottom line. It's in what's included, what isn't, and what will arrive on the bill in three months.
What the cost of an AI agent for business is made of
Here is the full map. Not every line applies to every business, but it's worth knowing they all exist.
| Component | What it includes | Payment type | Who you pay |
|---|---|---|---|
| Setup | Scoping tasks, connecting systems, permissions, learning the business, a trial period | One-time | Whoever sets up the agent |
| AI model | A fixed subscription or pay-per-use (tokens) | Monthly | The model provider, for example OpenAI |
| Server | The cloud computer the agent runs on 24/7, including backups | Monthly | The cloud provider |
| Additional tools | For example WhatsApp Business API, an email tool, file storage | Monthly or per use | Each provider separately |
| Support and maintenance | Adjustments, fixes, updates, adding tasks | Monthly, usually optional | Whoever sets up the agent |
| Your time | Approvals, feedback, updating instructions | Not in money, but real | Yourself |
What the cost looks like over a year
Most people look at the big number in the proposal, the setup. But over time the picture changes: the setup is paid once, and the monthly components add up. The chart below is an illustration of the structure, not of prices.
How the cost is divided over the first year (an illustration)
An illustration only. The heights show the ratio between components, not real prices. Every business divides it differently.
Two conclusions from the chart. The first: when you compare proposals, compare a year's cost, not just the setup. A proposal with a cheap setup and expensive monthly support can end up costing more by the end of the year. The second: the first months are an investment. After the trial period, most of the cost is ongoing and more predictable.
Setup: what you pay for once
Setup is the hardest part to compare, because it is mostly work. Here is what should be inside:
- Scoping. Which tasks the agent does, what counts as "good," and what needs your approval.
- Connections. A WordPress site, a store, Google Ads, Meta, Analytics. Each system is a separate connection, with its own permissions. The more systems, the more work.
- Permissions and security. A dedicated user for the agent in each system, with the minimum permission it needs. This takes time, and it's exactly the time you want spent.
- Learning the business. Tone of voice, examples, rules, products. This is what turns a generic agent into an agent that knows you.
- A trial period. A few weeks of close checking and fixes until the tasks work the way they should.
What makes setup more expensive? More systems, old or non-standard systems, tasks that need complex judgment, and a requirement to work across several channels at once. What makes it cheaper? Starting with 2–3 tasks and expanding later. We detailed the steps themselves in the guide to building an AI agent for your business.
Subscription or pay-per-use: the two models for the AI model
The model is the agent's "brain," and there are two main ways to pay for it. The difference is like the difference between a monthly gym membership and paying for each visit.
A fixed subscription
Like a monthly gym membership
- A fixed price per month, usually per user
- Easy to plan a budget
- Usage limits are set by the plan
- Suits predictable, even use
- Example: paid ChatGPT plans
Pay-per-use (API)
Like paying for each visit
- You pay by the amount of text going in and out (tokens)
- A quiet month costs less, a busy month costs more
- You need a spending cap and alerts
- Lets you choose a different model for each task
- Example: OpenAI's API
How does pay-per-use work? As of September 2026, OpenAI's API is priced per million tokens. A token is a small piece of text, roughly part of a word. There is a separate price for text going into the model (the instructions, the documents, the question) and for text coming out of it (the answer), and the text coming out costs more. Text that repeats and is kept in the cache (cached input) is significantly cheaper, and work that isn't urgent can be run in Batch mode at a discount. Using tools such as web search is billed separately. The prices themselves vary by model, so we won't quote them here: they update often, and it's best to check the official pricing page.
What does this mean in practice? An agent that writes short product descriptions uses little. An agent that reads long reports every day, summarizes documents or answers customers at high volume uses much more. So there's no point asking "how much does the API cost" without first asking "what will the agent do, and how many times."
Server and additional tools: the quiet bills
The server. An agent that works for you even when your computer is off needs to run on a server in the cloud. That's a monthly bill from the cloud provider, depending on the size of the server and the backups. Usually it isn't the big component, but it's fixed, and it's worth making sure it has a backup and is registered in the business's name.
WhatsApp Business API. If the agent is supposed to send messages to customers on WhatsApp, that has its own pricing model. As of September 2026, Meta charges per message: marketing template messages are always charged, service messages that answer a customer are free, and update messages (like an order confirmation) are free within the 24-hour service window after the customer writes. The price depends on the country of the person receiving the message. In addition, there's sometimes a cost for the provider that connects you to the platform. The implication: an agent that replies to customers who contacted you costs differently from an agent that sends a marketing campaign to a list.
Other tools. An email service, file storage, a design tool, a CRM. Some you already have, some will need a subscription. Ask for a full list up front, so you don't discover them on your credit card statement.
The costs missing from the price quote
These are the costs almost nobody writes down, and they're exactly the ones that create disappointment in the third month.
The hidden costs, and what to do about them
Your time
- Approving posts, products and changes
- Feedback in the first weeks
- More at the start, less over time
- What to do: define in advance what needs approval and what doesn't
Knowledge and instructions
- Prices, products and promotions change
- The agent knows only what it was updated on
- Old instructions produce mistakes
- What to do: decide who updates, and how often
Usage spikes
- A busy month, a season, a promotion
- A task that gets stuck in a loop
- Long documents read again and again
- What to do: a spending cap and an alert at the provider
Your time is a real cost. An agent that prepares everything but waits for your approval on every comma will give you back less time than you thought. On the other hand, an agent that publishes on its own with no check at all is a risk. The middle path: at the start you approve a lot, and once you see that a certain task works well, you decide together whether to ease it.
Knowledge that goes stale costs money. If you changed a price and the agent doesn't know, it will write the old price. That's not a bug in the agent, it's maintenance that someone has to do. Ask in advance: who's responsible for it, you or the provider?
A usage spike shouldn't be a surprise. With pay-per-use, an unusual month will cost more. That's normal. What isn't normal is to find out only on the bill. With most API providers you can set a budget and alerts, and it's worth requiring that this be set up during setup.
How to compare AI agent price quotes
To compare apples to apples, ask each provider to break their proposal down into the same components: setup, external services (each one separately), support, and what isn't included. Then calculate a year's cost for each proposal. Only then do the numbers really speak.
Here is the checklist we'd like every business owner to go through before signing. Mark each question you got a clear answer to.
Questions to ask before signing
Red flags in an AI agent price quote
Some proposals are worth stopping at to ask one more question. Not because whoever wrote them is necessarily unfair, but because they don't let you know what you're paying for.
- "Everything included, unlimited." AI models cost money by use. Someone who promises "unlimited" at a fixed price will either limit you somewhere else, or will have to change the terms when your usage grows. Ask where the limit is.
- One line with no breakdown. A single monthly price without detailing what's inside doesn't let you compare, and doesn't let you know what will change.
- The agent runs on your personal passwords. An agent needs a dedicated user in each system, with the minimum permission it needs. Access with your password means it can do everything you can, including things you didn't intend.
- Not a word about approvals. If it doesn't say what the agent does on its own and what waits for you, ask before it happens in practice.
- Accounts in the provider's name. If the model subscription, the server and the data are registered with the provider, it's hard to leave.
- A promise of precise savings. "You'll save 20 hours a week" sounds good, but nobody can know that before seeing your business.
Want to check a proposal you already have in hand? Answer the questions according to what it says.
Is it worth the money? Calculate against today's hours
The question "how much does an AI agent cost" doesn't stand alone. It's always "how much compared with what." And the alternative isn't zero: it's the hours that you, or an employee, spend today on the same work. Or the work that isn't getting done at all.
This is how we'd do the calculation:
- How many hours a month go today to the tasks you want to hand off? The calculator in the guide AI agent for business helps you reach that number in a minute.
- How much is an hour worth to you? If it's your own hour, the question is what you'd do with it instead. If it's an hour of an employee or a freelancer, it has a clear price.
- How much time will you still spend even with an agent? Approvals, feedback and updates. An agent doesn't make the time disappear, it moves you from "doing" to "approving."
- What is a year's cost? Setup plus twelve months of services and support. That's the number you compare against the hours.
And there's a side that doesn't fit in the calculator: tasks that today simply don't happen. Products that wait two weeks to go up, a report nobody reads, a post that keeps being postponed. It's hard to price, but worth writing down next to the numbers.
How Platin Digital prices
We don't publish a price list, because the price really does depend on the tasks, the connections and the scope of support. What we do: before you start, you get a breakdown of the three components, setup, external services and maintenance, so you know exactly what you're paying for. Platin Digital is an OpenAI Select Partner, and our agents are built with whichever models and tools fit each business best, from OpenAI, Anthropic, Google and others.
Frequently asked questions
How much does an AI agent for business cost?
The cost of an AI agent for business is made up of a one-time setup, monthly external services such as the AI model, a server and additional tools, and support and maintenance, which are usually optional. The total price depends on the number of tasks, the number of systems you connect and the scope of support, so it's worth asking for a breakdown of every component and calculating a year's cost.
What is the difference between a ChatGPT subscription and paying for the API?
A ChatGPT subscription is a fixed monthly price, usually per user, with usage limits set by the plan. The API is billed by use, at a price per million tokens, where text coming out of the model costs more than text going into it. An agent that runs on a server and works in the business's systems usually uses the API.
Can I know in advance how much the model usage will cost?
You can estimate it from the tasks and how often they repeat, but not know exactly. That's why it's important to set a spending cap and alerts with the API provider, and to check actual usage after the first month.
What happens if I stop working with the provider?
It depends on what was agreed. It's best for the accounts with the service providers to be in the business's name, and for it to be written down which data, instructions and files you receive and how access is disconnected. Some connections will need adjustment if you move to another system.
Why doesn't Platin Digital publish a price?
Because every agent is built from different components. Before you start, you get a breakdown of the setup, the external services and the maintenance, so you can compare and decide.
