Prices checked October 16, 2026. Vendors change prices and plan names often, so confirm on the vendor's page before you budget.
This article prices one AI setup: the plan you pay for, how the bill is worked out, and the staff time around it. For the wider picture, including what other firms spend on AI and what outside help costs in Canada, read What AI costs a Canadian business: tools, help and time.
The short answer for a firm of 50 to 300 people: give staff who use AI daily a per-seat plan, such as Claude Team, ChatGPT Business, Microsoft 365 Copilot or Gemini in Google Workspace. Consider usage billing only for an automated job that runs many times a day without a person starting it. Then budget the staff time to set it up, learn it and check its work, because that usually costs more than the licence.
Two ways to pay for AI
A subscription charges a fixed fee per person each month. Each person gets a usage allowance. If someone uses the tool heavily, they may hit the limit and have to wait, but the bill does not change. Claude Pro and Max, Claude Team seats, ChatGPT Plus and Business, Microsoft 365 Copilot and Gemini in Google Workspace all work this way.
Usage billing charges for the amount of text the AI reads and writes, measured in tokens. A token is a small piece of text: about four characters, so 100 tokens is roughly 75 words (OpenAI). Prices are quoted per million tokens, and writing costs more than reading. Software that calls the AI directly (the API) is billed this way, and so are automated workflows built on it.
Some plans combine the two. Claude Enterprise has a fixed seat fee of US$20 per person a month, billed annually, with a 20-seat minimum. Every token your team uses is then billed separately at standard API rates. ChatGPT Business seats use their plan allowance first, then draw on a shared pool of credits (OpenAI rate card).
What one task costs on usage pricing
Here is a worked example using Anthropic's published API rates for Claude Sonnet 5.5: US$2 per million tokens read and US$10 per million written. The task is ours, chosen to be typical: read a 7,500-word document and write a 1,500-word summary. The arithmetic is ours too.
| Step | Tokens | Cost (USD) |
|---|---|---|
| Read a 7,500-word document | About 10,000 | 10,000 × $2 per million = $0.02 |
| Write a 1,500-word summary | About 2,000 | 2,000 × $10 per million = $0.02 |
| One task | About 12,000 | $0.04 (on Claude Opus 5.5, about $0.08) |
| 1,000 tasks a month | About 12 million | About $40 |
| 50,000 tasks a month | About 600 million | About $2,000 |
One task costs about four cents, and the last row shows the risk. Once a workflow runs by itself, on every incoming email or every new file, the number of tasks grows and the bill grows with it. A seat costs the same whether the person uses it twice or two hundred times.
Which one fits your firm
| Situation | Better fit | Why |
|---|---|---|
| Staff use AI themselves to draft, summarise and research | Seats | A fixed monthly cost per person that finance can forecast. Most teams of 50 to 300 people belong here. |
| A few people use it heavily all day | A higher tier for those people | Claude Max offers 5 or 20 times Pro usage, and Claude Team and ChatGPT Business both offer premium seats. Keep everyone else on a standard seat. |
| An automated workflow runs hundreds or thousands of times a day | Usage | Software is doing the work, so no person needs a seat. You pay for each run. Set a monthly spend limit before it goes live. |
| Enterprise controls plus heavy use | Seat fee plus usage | Plans such as Claude Enterprise charge both. Ask for an estimate based on your expected usage and set spend limits. |
Many firms end up with both: seats for staff, and a usage account for one or two automated jobs. If you take on usage billing, check the bill monthly for the first quarter. Claude lets admins set spend limits, and both Anthropic and OpenAI sell prepaid credits.
Claude seat prices
Claude as the worked example, because it sells both seats and usage, from Claude's official pricing page. Prices are in US dollars per person per month. Claude says Pro is charged in your local currency where supported, but it does not publish a Canadian dollar price.
| Claude plan | Price (USD per person per month) | Who it is for |
|---|---|---|
| Free | $0 | Trying it out |
| Pro | $20 monthly, or $200 a year billed annually (about $17 a month) | One person |
| Max | From $100 | One heavy user: 5 or 20 times Pro usage |
| Team, standard seat | $25 monthly, or $20 billed annually | Teams of 2 to 150 people |
| Team, premium seat | $125 monthly, or $100 billed annually | Heavier users on the same team plan: five times the usage of a standard seat |
| Enterprise | $20 per seat billed annually, 20-seat minimum, plus usage at API rates | Larger firms that need enterprise controls |
Our arithmetic for a 120-person firm on Claude Team standard seats billed annually: 120 × US$20 = US$2,400 a month. At the Bank of Canada rate for October 15, 2026 (1 USD = 1.4196 CAD), that is about CAD 3,407 a month.
For seat prices in Canadian dollars for ChatGPT Business, Microsoft 365 Copilot and Gemini in Google Workspace, see What AI costs a Canadian business. They follow the same per-seat model: ChatGPT Business costs the same as a Claude Team standard seat, and Copilot and Gemini come as part of, or an add-on to, a Microsoft 365 or Google Workspace plan.
Compare plans on what they let staff do with company information, the controls your IT provider needs and the usage allowance, as well as price. Check what you already pay for too. Many firms already have Microsoft 365 or Google Workspace, which changes the real added cost. Our comparison of Copilot, Claude and ChatGPT covers the differences.
The costs around the tool
A licence is a fixed, visible number. The larger cost is staff time, and it rarely appears in the budget. We do not publish a standard hours estimate, because it depends on the task and the people. Budget each of these as a line of its own:
| Cost | What it involves | Who pays it |
|---|---|---|
| Setup | Writing instructions, gathering example documents and templates, deciding what information the AI may use, and testing it on a real task. | The person who owns the task, plus whoever helps configure it |
| Learning | Practice until the new way is faster than the old one. Leaders need it first, then staff. | Every person who will use it |
| Checking | Reading and correcting the output before it goes to a client or into a decision. This continues after launch. | The person accountable for the result |
| Integration and admin | Connecting email, files or other systems, granting access, setting policies and managing seats. | Your IT provider or internal admin |
Setup and learning are the ones most often left out. AI adoption has a setup tax explains why and how to fund the time. For the learning line, how to train employees on AI sets out a leaders-first plan with protected time for champions.
A draft produced in two minutes that needs forty minutes of correction may not save anything, so count checking time before you decide whether the setup pays off.
Price one recurring task
Pick one task your team repeats every week, such as a proposal, a client summary or a monthly report. Price the first 90 days, with cash and staff time in separate columns:
| Line | What to enter |
|---|---|
| Seats for three months | Number of people × the plan price × 3. Note the billing currency and whether annual billing is paid up front. |
| Usage for three months | Only if an automated workflow is involved: expected runs a month × cost per run × 3. Use the worked example above as a guide. |
| Outside help | A setup or training fee if you use one. Leave it at zero if you do it yourself. |
| Staff time (separate column) | Hours for setup, learning, checking and admin. Keep this beside the cash figures, not added to them. |
Then run a few normal examples of the task the new way. Count preparation and correction time, and how often the output was usable. Keep time saved separate from money saved. Freeing an hour does not reduce payroll or create billable work unless someone decides what that hour is for.
After it is running
Name who will update the instructions when the task changes, replace out-of-date examples, review who has access and fix a broken connection. If outside help set it up, ask what you keep at handover: the instructions, configuration and documentation, and the right to keep using them. Subscriptions and usage keep billing after the help ends, so check them every quarter.
If you want help pricing a setup for your team, book a free 30-minute call. Bring the task, the tools you pay for and what makes it difficult today. The first conversation establishes whether Solo, Solo for Execs, Fractional AI Leadership, another route or no project makes sense.
Free resource: The Sidekick Starter Kit.
