The short answer

There is no defensible single “average AI budget” for a small business. A two-person team using one general assistant has a different cost structure from a company connecting product data, a CRM, customer requests, and approval rules to an API-supported workflow.

The useful budget separates four categories: licenses, usage, implementation, and ownership. Vendor pages publish the first two. The organization must estimate preparation, integration, testing, training, review, maintenance, and the staff time needed to keep the workflow dependable.

The right question is not only “What does the tool cost?” It is “What does the current work cost, what improvement is realistic, and what will it take to operate the new process responsibly?”

What the published prices are

These figures were checked against official vendor pages in August 2026. Promotions, billing terms, minimum seats, included features, and enterprise agreements can change, so confirm the linked source before approving a budget.

PlanPriceNotes
ChatGPT Business$20 / user / mo annual; $25 monthlyTwo-user minimum. API usage is separate.
Claude team plansCheck current plan and contract termsAnthropic's pricing experience varies by team size, usage, and contract; confirm the displayed offer directly.
Microsoft 365 Copilot BusinessCheck current annual and monthly termsRequires a qualifying Microsoft 365 plan; promotions may affect the displayed price.
Google Workspace BusinessStandard tiers shown at $7, $14, and $22 / user / moGemini features are included by tier; promotional pricing may appear.
ChatGPT EnterpriseContact salesOpenAI does not publish a universal seat minimum or price.
Do not compare by seat price alone. Review data controls, administration, identity, retention, contractual requirements, support, and whether the plan matches the actual work. A lower license price can still be the more expensive choice if staff cannot use it consistently or safely.

The part nobody prices

Every cost above is a line item you can see. The ones below are the reason AI budgets overrun, and none of them appear on a pricing page.

Data preparation

If your product information lives in three systems that disagree, no model fixes that. Someone reconciles it first. On a messy catalogue this is frequently the largest single cost of the project.

Integration

A subscription gives you a chat window. Connecting it to the systems where work actually happens — your storefront, CRM, inventory, help desk — is separate work with a separate cost.

Training and rework

Staff time spent learning, plus output that is revised or discarded before everyone agrees what good looks like. Measure this directly instead of applying a generic multiplier.

Per-seat, API, or build?

Three routes, and most companies default to the first without asking whether it fits.

Per-seat subscriptions

Often appropriate when the work is open-ended and human-driven—research, drafting, analysis, and different people doing different things. Business plans can provide stronger administration and data controls than personal accounts, but the organization still needs approved-use rules, staff guidance, review standards, and a clear policy for sensitive information.

API access

Often appropriate when the work is repetitive and structured—the same operation running many times with defined inputs and reviewable outputs. API economics depend on volume, model choice, prompt and output size, caching, retries, monitoring, and the engineering or platform work needed to operate the process.

A built workflow

Often appropriate when the task has a defined input and a checkable output and runs frequently. A built workflow costs more up front and may reduce seat dependence, but it can still carry recurring API, platform, hosting, monitoring, support, and maintenance costs. It can remove some manual steps; it should not be described as eliminating all human work or judgment.

The test: list your ten most repetitive tasks. If most are open-ended and done by different people, buy seats. If most are the same operation repeated, seats are the expensive answer and you are paying per person for something that should run once.

A worked example

This illustration is not a client promise. Consider a twelve-person consumer-products company selling on two marketplaces plus its own store.

  • Seats: 6 people × $20 = $120/month. Not everyone needs one.
  • Workflow tools: two category tools at roughly $60–150/month combined.
  • Visible total: about $250/month, or $3,000/year.
  • Implementation and operating cost: estimated separately after documenting preparation, integration, testing, review, training, and support.

Now the other side. If two people each spend six hours a week on listing content at a $45 loaded hourly cost for 46 working weeks, that task consumes about $24,840 a year. Recovering one third would represent approximately $8,280 of annual capacity. That is recoverable capacity, not guaranteed cash savings.

That gap — between what the software costs and what the manual work costs — is the actual business case. Most owners have never calculated the second number.

Run your own numbers in the ROI calculator →

What to do first

  1. Price the problem, not the tool. Work out what one repetitive task costs per year before looking at any subscription page.
  2. Buy one seat and use it properly for a month. Do not roll out five tools to twelve people. One person, one real task, thirty days.
  3. Write down what good output looks like before anything reaches a customer. Most of the rework cost is the absence of this one document.
  4. Only then decide whether the answer is more seats, an API process, or a built workflow.
When AI is the wrong answer: if the task requires judgement more than production, if the inputs are inconsistent every time, or if the process is undocumented, automating it will produce bad output faster. Fix the process first. Sometimes the honest answer is that a $250 monthly stack cannot solve a problem that is actually about who owns which decision.

Sources

Primary pricing sources checked August 2026: OpenAI Business, Anthropic plans, Microsoft 365 Copilot, and Google Workspace. Vendor prices and terms change; confirm current figures before budgeting.