The Real Cost of AI: How to Measure True ROI for Your Firm
When you budget for an AI tool, you budget for the bill the vendor sends. That bill is the visible part of the cost. It is also the smaller part.
The expenses that decide whether AI pays off for your firm never appear on the vendor’s invoice at all, and most firms do not budget for them until they arrive.
This is the part of AI economics that catches finance leaders off guard. The token cost is the tip. The real cost sits below the surface, and it is made of your people’s time and your firm’s regulatory exposure.
Sign up for Kiplinger’s Free Newsletters
Profit and prosper with the best of expert advice on investing, taxes, retirement, personal finance and more – straight to your e-mail.
Profit and prosper with the best of expert advice – straight to your e-mail.
The cost of review
Every piece of AI output that reaches a client must be checked by a human first. This is not optional for a fiduciary. You cannot send an AI-drafted client communication, an AI-generated summary or an AI-assisted recommendation to the people who trust your firm with their money without a qualified person reviewing it.
The technology does not change the standard of care. It changes who does the first draft.
That review is a labor cost, and it scales with how much AI you use. The more your advisers generate, the more there is to check. A firm that measures only the token bill sees AI getting cheaper per task while the review burden quietly grows.
If you do not budget the review time, you have not budgeted the tool. You have budgeted half of it.
The cost of training
A tool your staff cannot use well is a tool you are overpaying for. I see this all the time with firms that roll out Microsoft Copilot without any training around how to use the tool and get the most out of it. These firms quickly find the costs without the benefits.
Getting real value out of AI requires teaching your people how to prompt it and how to judge what comes back, including when to distrust it. That training takes time, it takes a person to deliver it, and it repeats every time the tool changes or a new hire arrives.
This cost is easy to skip and expensive to skip. Untrained staff produce worse results from the same tool, which makes the AI tool look like a poor investment when the real problem is the absence of training. The token bill is more expensive when your people use the tool badly, reducing your return on the investment.
The cost of governance
This is the line that protects the firm, and it is the one most likely to be missing from the budget.
Using AI responsibly in a regulated business requires an acceptable-use policy that classifies which tools are approved and which data may be processed. It requires vendor due diligence documentation for every tool that touches client data, mapped against your regulatory obligations.
It requires updated supervisory procedures showing how AI-assisted work is reviewed before it reaches a client. It also requires a training record an examiner can inspect.
None of that builds itself. Each piece takes time from compliance and operations staff, and it must be maintained as the tools and the rules change. The off-channel communications enforcement wave taught the industry an expensive lesson about applying existing rules to new technology after the fact.
AI governance is the same lesson waiting to be learned again. The firm that funds the tool but not the governance around it is buying the upside and leaving the downside unbudgeted.
Why ownership decides the outcome
These costs fall across three parts of your firm. The token bill belongs to technology. The review burden belongs to the leadership team. The governance work belongs to compliance. When one of those groups owns the AI budget alone, the costs that live in the other two go unfunded.
Research on AI return makes this concrete. According to the Mavvrik report AI Cost Statistics 2026: Forecasting, ROI, and Budget Risk, firms where technology teams own AI spend by themselves capture less value than firms where finance and compliance share the decision. The reason is exactly this fragmentation.
A technology-only budget sees the invoice and misses the iceberg. A shared budget sees the whole cost, funds it correctly and gets a real answer about whether the tool is worth it.
How to budget the whole cost
Start by writing down every cost a single AI workflow creates, not just the one the vendor charges for. Put the token estimate at the top. Then add the hours of review the output will require, the training to get staff using it well and the compliance work to govern it.
That full number is the real cost of the tool. It is the only number that tells you whether the investment returns anything.
I want to point out that this cost will always be less than a human cost, but it should be clearly measured.
Then assign each cost to the group that incurs it and bring those groups into one budget conversation. The token line is a technology decision. The rest is not.
The firm that budgets the whole iceberg will know what its AI use costs and whether it pays dividends on the investment.
The firm that budgets only the tip will be surprised twice, once by the hidden costs and again by the return that never materialized because the tool was never properly supported.
