What the available evidence does not establish
The AI spend statistics page distinguishes enterprise-leader surveys, technology-professional surveys, vendor datasets and worker-reported AI use. Those populations and measures are not a direct sample of AI budgets in firms of this size.
That is a limit on the sources reviewed, not proof that no relevant evidence could exist anywhere. A survey percentage about usage or unexpected charges should not become a percentage-of-revenue budget or a savings target for your firm.
How much should a firm spend on AI: start with the arithmetic
Period budget = paid seat charges + expected billable usage + required base software + other applicable costs. Calculate each line in the same currency and time period. Keep fixed commitments separate from assumptions about variable usage.
The scenarios below use ChatGPT Business Standard at US$20, the current Copilot Business add-on at its US$21 list rate, and Microsoft 365 Business Standard with Teams at US$14 per user per month under annual plans. The Microsoft suite is one named qualifying base option, not the only eligible plan. All rates were verified October 29, 2026.
Source verified October 29, 2026: OpenAI, ChatGPT Business overview.
Source verified October 29, 2026: Microsoft, Microsoft 365 Copilot business pricing.
Source verified October 29, 2026: Microsoft, Microsoft 365 Business Standard.
Three firm sizes, worked
Illustrative examples: seat counts, headcounts and usage allowances below are hypothetical. They are not industry averages or recommended budgets. The assumed base suite covers every employee in each scenario; change that assumption to match your actual licenses.
| Firm size | Paid AI mix | Monthly seat equivalent | Usage allowance | Base suite equivalent | Formula-visible total |
|---|---|---|---|---|---|
| 10 people | 4 ChatGPT + 2 Copilot | 4 × US$20 + 2 × US$21 = US$122 | US$50, hypothetical | 10 × US$14 = US$140 | US$122 + US$50 + US$140 = US$312/month; × 12 = US$3,744 |
| 25 people | 10 ChatGPT + 5 Copilot | 10 × US$20 + 5 × US$21 = US$305 | US$100, hypothetical | 25 × US$14 = US$350 | US$305 + US$100 + US$350 = US$755/month; × 12 = US$9,060 |
| 100 people | 40 ChatGPT + 20 Copilot | 40 × US$20 + 20 × US$21 = US$1,220 | US$400, hypothetical | 100 × US$14 = US$1,400 | US$1,220 + US$400 + US$1,400 = US$3,020/month; × 12 = US$36,240 |
An annualized usage allowance is a scenario, not a contracted annual usage price. These totals exclude taxes, specialist tools, implementation and other charges. If the base software is already in your budget, do not count it again as incremental AI spend.
For the 25-person example, the incremental seat-and-usage scenario is (US$305 + US$100) × 12 = US$4,860. The US$4,200 base-suite commitment is separately visible. No difference is presented as a saving or forecast.
Estimate the seat commitment
Use separate counts for Standard and Premium seats. Formula: (Standard seats x Standard rate) + (Premium seats x Premium rate) = monthly equivalent; monthly equivalent x 12 = annual equivalent. A monthly plan's twelve-month total is a scenario, not an annual commitment.
Billing period
Review this estimate with the AI Renewal Checklist, or explore the AI Spend Control System for a manual working toolkit.
Excluded from the seat subtotal: taxes, API usage, workspace credits, extra usage and separate base licenses. Google Workspace is a complete suite price. Copilot uses its add-on list price, with its base license excluded unless the separate scenario is shown.
ChatGPT's current 200-seat limit applies to workspaces created from August 24, 2026; older workspaces retain their previous limit. Official seat-limit scope, verified October 29, 2026.
Calculated in your browser. Nothing you enter leaves this page.
Set your own budget
- Inventory the subscriptions and billing accounts using the AI tool inventory template. Name the person who confirms each line.
- Price the actual seat mix and contract term. Read the AI seat versus usage pricing guide and the relevant vendor cost guide before mixing fee types.
- Estimate usage from a representative period, noting seasonality, workload changes and any enforceable limits. Keep uncertain inputs visible.
- Set the approval threshold and next review date under the firm's own rules. Record the evidence and required action before the notice deadline.
Questions about the budget
Is there a rule of thumb, like a percentage of revenue?
The sources reviewed here do not establish a reliable percentage for U.S. professional-service firms of 10 to 250 people. Build a budget from justified seats, measured usage, required base software and the firm's own approval rules.
Should we budget per employee or per seat?
Use paid licenses for fixed seat charges and the correct units for usage. Employee count is useful context, but not every employee needs the same paid plan and one person can hold multiple subscriptions.
How often should the budget be reviewed?
Review it before renewal notice deadlines and when pricing, staffing or the approved workload changes. Keep a recurring review date and assign an owner for any unexpected usage charge.
Turn the budget into a review decision
This illustrative example is a starting point for your own records. Start with one invoice and one upcoming renewal. Use the AI Renewal Checklist with no signup, then follow the AI subscription audit guide to build the wider review.
Keep the evidence, approver, notice deadline and Keep, Resize, Test or Retire decision in your own records or the free companion templates. These are not dedicated fields in the paid Control Center or workbook. Compare AI subscription review options if you need a different collection method.
The AccruePilot AI Spend Control System costs US$59 once for a local browser Control Center, editable Excel workbook and 12-page field guide. It is for one purchasing U.S. organization's internal use. Your team supplies and maintains the records; the toolkit does not discover accounts, enforce budgets or cancel services.
Start the free renewal checklist
Published by AccruePilot, which sells the manual toolkit discussed here. Vendor names identify the services reviewed and do not imply endorsement.