your business will spend even more A.I. this year. Three-quarters of finance leaders get a raise 2026 technology budgetalmost half of them increased by more than 10%, with financial services companies leading all other sectors with an increase of about 15%.
Spending decisions are made. An even more difficult question is one that the budget process is not built to answer. “How do you predict costs that will change based on how your employees use the tool?”
Most companies get this wrong in two specific ways. Let’s look at both before budgeting for AI.
just from $107.88 Kiplinger Personal Finance $24.99
Become a smarter, more informed investor. Please subscribe from $107.88 $24.99 plus get up to 4 special issues
Click on the free question
Sign up for Kiplinger’s free newsletter
Profit and prosper with the best expert advice on investing, taxes, retirement, personal finance and more straight to your email.
Profit and prosper with the best expert advice straight to your email.
Mistake #1: Budgeting AI as a fixed cost
You’re used to software that costs the same every month. You buy a number of seats, pay a flat rate, and your bill becomes your bill.
AI doesn’t work like that. Costs track consumption rather than number of employees. Two advisors with the same license can generate significantly different invoices. This is because one person uses the tool to run long reports all day, and the other person questions the tool twice a week.
This will destroy your annual budget. You can’t set one number for January and stick with it. This is because usage increases as the skills of employees improve. AI tools Find even more uses.
Fixed line items guarantee one of two outcomes. Either you overfund and waste capital that could have been deployed elsewhere, or you run out of money and face overrun stories in near future quarters.
Treat AI the same way you would treat variable operating costs, not licenses. Build this as a range with a lower and upper bound, forecast on a rolling basis, and revised quarterly. Companies that operate this as a static annual number budget their variable expenses on a fixed budget, resulting in a budget deficit.
Mistake 2: Budget only token bills.
The number charged by the vendor is a tangible part of the cost. It is not the full cost.
All AI workflows incur a cost that never appears on a vendor’s bill: the humans involved.
- Someone needs to review the output before it reaches the client. trustee You can’t send unreviewed AI artifacts to people who trust and fund your company
- Someone needs to train staff on the tools
- someone has to maintain governanceAcceptable Use Policy, Compliance Records Required by Examiner
These are real costs, scale with deployment, and are on the same budget line as the tokens.
This is why ownership is just as important as numbers. Research on AI returns shows that companies whose technology teams solely own AI receive less value than those whose technology teams own AI. Finance and compliance share decisions.
The token bill is a technology expense (although it should be a business expense). Review time and regulatory risk are not related. If you do not budget for these things together, you will not have enough funds to avoid trouble.
Set a budget for honest profits, not promises
This is the number that determines how aggressively you want to fund this. Less than 1% of executives report AI returns over 20%the majority of reports have returns ranging from 1% to 5%.
Meanwhile, Gartner predicts that: 30% of generative AI projects are abandoned After the proof of concept stage. The spending is real. Most companies have yet to see proven benefits.
This is not an advocate of being sedentary. We insist on funding based on measured results, not vendor promises. Tie each AI budget line to a specific measurable outcome, such as time savings on named workflows or specific operational cost reductions.
We fund use cases that meet those criteria and reject those that don’t. The companies that will win over the next three years will be those that fund AI where the business case is strong and measurable, and refuse to fund it everywhere else.
how to assemble numbers
First, let’s measure the pilot. Use this tool to run your actual workflow for a month and review your invoices. This gives you the actual cost per task, which is the only accurate input to your prediction. multiply by a realistic value Adoptionis not a best case adoption. Then add costs that vendors never charge: time to review output, time to train staff, and work to maintain governance.
Build the result as a range. Set up a floor that covers reliable usage and a ceiling that can accommodate the growth you know is coming. Connecting a pay-as-you-go overflow slows down the workflows that advisors currently rely on when they exceed the limit, rather than blocking them. We then revisit the overall numbers quarterly. This is because the underlying price is changing and usage is moving faster.
AI is now a permanent part of operating budgets. Treat it like one. Forecast like a variable expense, fund it based on measured revenue, and modify it based on a schedule. CFOs who do this place capital where it will yield returns. A CFO who sets a fixed number in January will have a year to account for the variance.
Related content
This article was written by and represents the views of our contributing advisors and not of Kiplinger’s editorial staff. To check your advisor’s records, SEC or together finra.
