What Does AI Actually Cost a Small Business?

Last updated June 2026. First published with current per-seat pricing from Microsoft and OpenAI, Statistics Canada AI-adoption data, and the full hidden-cost breakdown.

"How much does AI cost?" is the first question almost every business owner asks, and the honest answer is that the price tag on an AI tool is the part that matters least. The subscription is easy to find. What the vendor page never shows you is the cost of getting the tool to actually work in your business, and that is where most of the money, and most of the risk, lives.

This is a plain-English breakdown of what AI really costs a small or mid-size business in 2026: the per-seat tools, the custom builds, and the hidden costs that decide whether your spend pays off. The figures below are current published list prices, each linked to its source, and we have not invented a single number.

The Short Answer

For a typical small business, AI spending falls into three layers, and the most visible one is usually the smallest:

• Per-seat subscriptions for off-the-shelf assistants: on the order of USD 18 to 30 per user per month, billed annually.
• Custom or managed build cost when you want AI fitted to your own workflows rather than a generic chatbot: a one-time or ongoing project cost that varies widely with scope.
• Hidden costs: data preparation, integration, training your team, and ongoing governance. These rarely appear in any quote, and they routinely dwarf the subscription.

The trap is budgeting only for the first layer. A tool nobody adopts, or one that quietly mishandles customer data, costs far more than its monthly fee. Demand for AI is real, so this is worth getting right: Statistics Canada reports that 12.2% of Canadian firms used AI to produce goods or deliver services in 2025, double the share from the year before, with another 14.5% planning to adopt within twelve months. The spend is coming for most businesses; the question is how to make it count.

The Three Ways to Buy AI

Before pricing anything, it helps to know which of three things you are actually buying. They cost very differently and solve different problems.

• Per-seat assistants. A licence per employee for a general-purpose chatbot or productivity add-on (the AI in your office suite, a standalone chat tool). Lowest entry cost, fastest to switch on, least tailored to your business.
• Custom AI. Software built around your specific workflow: a quoting assistant that knows your pricing, an intake bot that follows your process, a report generator that reads your data. Higher upfront cost, far higher fit.
• Managed AI. Someone else builds the custom AI, runs it, monitors what it costs and how well it works, and keeps it secure, for a predictable fee. You pay for an outcome instead of assembling and babysitting the pieces yourself.

Most businesses end up with a mix. The mistake is assuming the cheapest option (per-seat tools) is automatically the right one, when the real cost depends on what you are trying to achieve.

Per-Seat AI Tools: The Sticker Price

Per-seat tools are the easiest to price because the vendors publish it. Here is where the mainstream business options sit as of June 2026, in US dollars on annual billing. Always confirm the current rate and your regional price before you budget.

• Microsoft 365 Copilot Business lists at USD 18 per user per month on annual billing for businesses up to 300 users (a limited-time price; the standard rate is USD 21). It requires a qualifying Microsoft 365 plan underneath, so the true cost is the Copilot add-on plus the base licence. See Microsoft's Copilot pricing page.
• ChatGPT Team from OpenAI lists at USD 25 per user per month on annual billing (USD 30 month to month), per OpenAI's published ChatGPT business pricing.

So for a ten-person team, a per-seat assistant is roughly USD 180 to 300 a month before tax, plus whatever base subscription it sits on. That is a real but modest number. The problem is what it does not include: the tool arrives as a blank box, and getting value out of it is on you. That is where the hidden costs start.

It is also worth noting that per-seat AI is the most common source of the same waste we already see with software licences in general. We dig into how unused seats pile up in our piece on Microsoft 365 licence waste, and AI seats follow the same pattern.

Custom and Managed AI: A Different Model

Custom AI is priced as a project, not a per-seat sticker, so there is no single number to quote, and any provider who gives you one without understanding your workflow is guessing. The cost scales with scope: a narrow assistant that automates one repetitive task is a modest build; a system that touches several departments and integrates multiple systems is a larger one.

The honest way to think about custom AI cost is not "what is the price" but "what is the work worth." If a build automates an hour of skilled work per person per day, the value is measured against salary, not against a subscription. That is the calculation that actually tells you whether to proceed, and we walk through it in detail in our guide to the ROI of managed AI for a small business.

Managed AI changes the cost shape again. Instead of a big upfront build plus an open-ended question of who keeps it running, you pay a predictable ongoing fee that covers building, running, monitoring, and securing the AI. The point is not that it is cheaper than a tool subscription (it is not), but that it bundles the hidden costs below into one accountable line instead of leaving them to land on you unplanned.

The Hidden Costs Nobody Quotes

These are the costs that turn a cheap subscription into an expensive mistake, or a fair build into a bargain. None of them appear on a pricing page.

• Data preparation. AI is only as useful as the data it can see. Getting your information clean, organized, and connected so the AI can use it is often the single biggest line of effort, and it is real work whether you do it internally or pay someone.
• Integration. A tool that cannot reach your systems gives generic answers. Connecting AI to your email, files, and line-of- business apps so it works from your real context takes setup and maintenance.
• Adoption and training. A licence is not a habit. If your team does not change how it works, the spend produces nothing. Training, change management, and the weeks where productivity dips before it rises are a genuine cost.
• Governance and security. Deciding what staff may put into AI tools, keeping customer data out of public models, and meeting your obligations under Canadian privacy law all take time and policy. Skipping this is not a saving; it is a deferred risk. (We cover the rules in AI and PIPEDA.)
• Maintenance and drift. AI tools, models, and prices change. What worked last quarter may need re-tuning. Someone has to watch quality and cost over time, or the value quietly erodes.

The Licences-Unused Trap

The most common way small businesses waste money on AI is the same way they waste money on any software: buying seats that nobody uses. It is easy to roll out an AI assistant to the whole company, see a burst of early curiosity, and then watch usage quietly fall off as people drift back to old habits. Three months later you are paying full price for licences that sit idle.

This is why adoption is a cost line, not an afterthought. A per-seat tool only pays for itself if people actually change how they work, and that does not happen on its own. The fix is to start narrow (a few power users, one real workflow), prove the value, and expand only where usage is real, rather than buying for everyone on day one and hoping it sticks. Reviewing usage and trimming idle seats should be a recurring habit, the same way it is for the rest of your software.

How to Think About What to Spend

Instead of asking "what does AI cost," ask these questions in order. They will tell you what to actually spend.

• What specific work am I trying to change? Name the task, the hours it consumes, and who does it. If you cannot, you are not ready to buy yet.
• What is that work worth? Multiply the hours by what the people doing them cost. That number is your budget ceiling, and it is usually far larger than any subscription.
• Will a generic tool do it, or does it need my context? If a per-seat assistant solves it, start there. If the value depends on your data and process, a custom or managed build will return more than it costs.
• Who is accountable for adoption and governance?If the answer is "nobody," budget for it or expect the spend to underperform.

Spend in proportion to the value of the work you are changing, not in proportion to the lowest sticker price you can find. A USD 25 tool that nobody adopts is more expensive than a fitted build that saves real hours.

When AI Is Not Worth the Cost

Honest cost advice has to include the cases where the answer is no. AI is not worth the spend when:

• The task you want to automate is rare, low-value, or already fast. The cost of building and governing the AI exceeds what it saves.
• Your underlying data is a mess and you are not willing to fix it. AI will amplify the mess, not fix it, and you will pay for the privilege.
• You have no one to own adoption. Without that, you are buying licences that will sit unused, which is pure cost.
• The work involves sensitive personal information and you have not worked out the privacy and security side. The right move is to sort that out first, not to spend and hope.

Walking away, or waiting, is a legitimate and sometimes cheaper answer. The goal is value, not adoption for its own sake.

If you would rather not carry the hidden costs yourself, managed AI folds the build, the integration, the monitoring, and the governance into one predictable fee, so the cost is planned rather than discovered. Either way, the right first step is the same: a clear-eyed look at the work you want to change and what it is worth. We are happy to help you think it through, with no obligation to buy anything.