"How much does AI cost for a small business?" is really five different questions, and the honest answer is a different one for each. Some of the spend is a monthly fee per person that you can read off a web page. Some is already inside software you pay for. Some is a one-off build with no list price at all. And a good part of the real cost never appears on an invoice. This article is for the owner of a company with ten to two hundred people who typed that question, or "AI cost for small business", or "how much does it cost to use AI in a business", and wants to know where the money goes before signing anything.
I was COO and then President of a public company that closed more than forty acquisitions, and I ran the integration on most of them, which meant I watched many companies with no technical staff adopt, or fail to adopt, new systems. I am a lawyer by training. Now I write fixed-fee AI plans for CEOs of private-equity-backed and owner-led mid-market companies through 013 Labs. Nothing here is legal advice; have your own lawyer review any contract before you sign it.
Why there is no single number
Asking what AI costs is a bit like asking what a vehicle costs. A bicycle courier and a trucking firm are both buying "transport". The first kind of spend below is cheap and easy to compare. The last two are where owners are surprised. I have kept price figures to the ones I could read on a vendor's own pricing page on the day I wrote this, and dated them in the note at the end. Everything else is described by shape, not by an invented number, because prices move and a stale figure is worse than none.
One: a monthly plan for each person on the chat tools
This is the spend most owners mean. The main chat tools sell a business plan: a fee for each user each month, with company-level controls such as a shared workspace and an administrator. The price is listed on the vendor's own pricing page, usually with one figure if you pay for a year up front and a higher one if you pay month to month.
What you are buying is access for the people who will use it, not a result. The arithmetic is simple, since it is the fee multiplied by the number of seats, and so the real decision is how many seats. A common mistake is buying a seat for everyone on day one. A better habit is to start with the handful of people who already use these tools, find out what they use them for, and add seats only when someone can name the task. If you want the difference between the free and paid versions of the same tools, free versus paid explains it, and is ChatGPT safe for business use covers what a business plan does and does not change about your data.
One more point on this kind of spend: some of the chat tools also bill by how much text is processed once they are connected to your own systems. That is a different meter from the monthly seat fee, so ask the vendor to explain it in writing before anything is connected.
Two: AI switched on inside software you already pay for
Your accounting package, your email and office suite, your customer records system and your help desk are all adding AI features. Some are included in the plan you already have. Some are an add-on with its own monthly fee for each user. Some are billed by use.
This is the spend that arrives without a decision. A renewal notice shows a new line, or a feature is turned on by a software update and nobody approved it. The practical step is a one-hour exercise: list the five or six systems your company pays for most, and for each one ask the vendor or look at the admin settings. Is an AI feature on? Is it free, an add-on, or metered? Where does the data go? You may find that part of what you want is already paid for, and that part of what you are paying for is not being used.
The same rule applies here as to everything else: the price on the page is the start of the cost, not the end of it. For example, if the question is whether a tool can help with your books, can AI do bookkeeping for a small business sets out what it can and cannot be trusted to do, which matters more than the fee.
Three: a custom build or an integration
A custom build means someone writes software that connects an AI tool to your documents, your customer records or your order flow, so that it does one job in your way. An integration is a smaller version: connecting something you already have to something else.
There is no list price, and anyone who quotes you one before asking what the job is deserves caution. What drives the cost is not the AI part. It is the work around it: how messy and scattered your data is, how many other systems have to be connected, how much has to be checked before an output is allowed to go anywhere, and who looks after it when something changes. A job that touches a single clean spreadsheet is a very different thing from one that touches three systems that do not agree with each other.
Ask any builder five things before you compare quotes. What exactly will it do and not do? What happens when it is wrong? Who in my company checks the work? What do I own at the end, including the written description of how it works? What does it cost each month to keep running after it is built? A builder who cannot answer the last one in writing is quoting you the cheap part.
Four: outside help, from a written plan to a full build
Outside help comes in a ladder, and it helps to know which rung you are standing on.
A written plan. A short, fixed-price document that says what your people are using, where AI belongs in your work and where it does not, what data may go where, who owns what, and what the first project is. It is the cheapest rung and it makes every later purchase easier to judge.
Guided setup. Someone helps you turn the plan into working habits: configuring the plan you bought, writing the data rule, training the first group and setting up the checking.
A build. Software written for you, as described above, usually done in stages with a stop point after each.
Ongoing support. Someone keeps it working, watches for changes and answers staff questions.
Pricing differs: some advisers charge a fixed fee, some bill by the hour, some charge a monthly retainer. The question to ask is not "what is your rate" but "what will I hold in my hand at the end, and what does it cost me if I stop here?" If you are choosing between advisers, how to choose an AI strategy consultant lists the questions that separate a useful one from a salesperson. And if you do not yet know what to point the money at, picking your first AI project is the place to start before you buy anything.
Five: the costs that never appear on an invoice
For most small companies this is the largest line, and it is the one almost nobody budgets.
Staff time to check outputs. AI tools write fluent, confident text that is sometimes wrong. Somebody has to look at every output before it is used, and that time is real. If checking an output takes as long as doing the task, there is no saving. Estimate it honestly in week one, from a person who does the task.
Training. People do not pick this up from an invitation email. Budget a few hours each, plus time for the questions that come up in the first month. The cost shows up as slower work for a short while before it shows up as faster work.
Data clean-up. A tool is only as good as what you give it. If your customer list has three spellings of every name, or your price list lives in five spreadsheets, the first job is tidying that up. Owners often discover that this is the whole project.
Someone who owns it. Somebody has to decide which tools are allowed, keep the list of who has a seat, write and repeat the data rule, and answer when a customer or a lender asks how the company uses AI. In a company with no IT department that job falls on a manager's already full week. Name the person, and give them the hours, or it will not happen.
Switching and waste. Seats nobody uses, a second tool bought because the first was hard to learn, and a plan that renews automatically after the enthusiasm has gone. A calendar entry sixty days before each renewal costs nothing and catches most of it.
If you want the full picture, AI for a small business with no IT department is a thirty-day plan that gives you the real numbers from your own company, rather than anyone's averages.
Once the tools are in, the question becomes whether they pay for themselves. How to check, from your own records: is AI actually saving my business money?
Picture the spend going wrong
Picture a regional services company of sixty people. The owner reads an article about AI and buys business seats for the whole office on an annual plan. Nobody was asked what they already use, so a third of the seats are never opened, and the people who do use the tool keep their own personal accounts as well. Separately, the company's accounting software quietly adds an AI add-on at the next renewal, which nobody notices. A consultant quotes for a custom build, but nobody has cleaned the customer records, so the first two months go on tidying data. No one is named to own it. Six months in, the owner can say what the company spends on subscriptions but not what any of it achieved, and cannot tell a lender what the company does with AI. The repair is not a bigger budget. It is a smaller one, spent in the right order: find out what is used, write the data rule, pick one task, then buy what that task needs. This is a hypothesis, a composite and not a client.
Mistakes that repeat
Buying seats for everyone at once. Counting only the fee on the pricing page. Missing the AI add-ons already on your renewals. Asking for a custom build before a written plan. Not asking what the monthly cost is after a build. Forgetting staff time to check outputs. No named owner. No renewal reminder.
Frequently asked questions
How much does AI cost for a small business?
It depends on which of five kinds of spend you mean: a monthly fee for each user on a chat tool, AI features inside software you already pay for, a custom build, outside help, and the hidden costs of staff time, training, data clean-up and ownership. The first is listed on the vendor's own pricing page. The others you should ask about in writing.
Is a business plan for a chat tool worth it?
When a handful of people already use these tools for real work and you want company controls over the data, usually yes. Start with those people, not the whole office, and add a seat only when someone can name the task.
What is the biggest cost owners miss?
Staff time. Checking every output, training people and tidying data cost more, in most small companies, than the subscription.
Do I need a custom build?
Probably not at first. Most owners get further with a plan they already pay for and one well-chosen task. A build makes sense when a job is repeated often, the data is in decent shape, and you can say what a good result looks like.
How do I stop the cost creeping up?
Keep a one-page list of every tool, who has a seat, what it costs and who owns it, and set a reminder before each renewal.
Not sure where you stand? Take the 3-minute AI readiness scorecard.