AI adoption is accelerating across businesses of every size. But when it comes to measuring what it’s actually worth — and what it truly costs — the picture is murkier than the headlines suggest.

By now, most business owners we work with have adopted at least one AI tool. ChatGPT for drafting. Copilot in Microsoft 365. AI-assisted bookkeeping. Automated scheduling or customer service. The tools are multiplying, the subscriptions are adding up, and the question we’re starting to hear more often is a good one:

“Is this actually working?”

It’s the right question. And the honest answer — for most businesses under $50M in revenue — is that the full return on AI investment is still emerging. That’s not a reason to pull back. But it is a reason to be more deliberate about what you’re spending, what you’re measuring, and what risks you’re taking on.

Here’s what we know so far — and what every business owner should be tracking.

The ROI Picture in 2026: Promising, But Unproven for Most

The enterprise-level data on AI ROI is sobering. According to research from MIT, 95% of generative AI projects in large organizations have failed to show measurable financial returns within six months. A separate study found that while over 70% of companies report “positive” AI ROI, fewer than 1% report returns of 20% or more. Most are seeing 1–5% gains — and even those are measured in productivity estimates, not actual margin improvement.

For smaller businesses, the pattern is similar. Tools get adopted. Workflows get adjusted. Output gets faster. But when someone asks “how much is this actually adding to the bottom line,” the answer is usually a shrug.

That’s not because AI isn’t valuable. It’s because most businesses don’t have a framework for measuring it — and because many of the real costs are hiding in places that aren’t obvious.

The Known Costs: What Shows Up on the Books

Most business owners are tracking their AI spending at the subscription level. That’s a start — but it’s usually only a fraction of the true cost.

• Subscription fees. The per-seat or per-tool monthly costs that show up on the credit card statement. ChatGPT, Copilot, Claude, Jasper, Notion AI, and dozens of others. These add up faster than most owners realize when you count every department using something different.
• Token and usage-based costs. Many AI tools — particularly those built on underlying models via API — charge based on usage rather than a flat fee. Token costs can scale significantly as usage increases, especially for businesses building custom AI workflows or integrations. This is one of the least-understood cost drivers in the category.
• Implementation and integration costs. Getting AI tools to work within your existing systems takes time and often professional support. Whether that’s IT hours, consultant fees, or custom development, the setup cost is rarely included in the vendor’s advertised price.
• Training costs. Your team needs to learn how to use these tools effectively. That takes time away from billable or productive work. And in specialized fields, it may also require formal training or certification investments to use AI responsibly.

The Hidden Costs: What Doesn’t Show Up — Until It Does

This is where the CFO lens matters most. There are real costs to AI adoption that won’t appear in your software budget — but will eventually show up somewhere else.

• The “rework tax.” AI tools make people faster. But research suggests that a significant portion of that time gets spent fixing AI mistakes rather than doing new work. If your team is moving faster but spending 40% of that time correcting errors, you haven’t gained productivity — you’ve just shifted where the work happens.
• Errors and omissions risk. AI tools are not infallible. In fields where accuracy matters — accounting, legal, compliance, healthcare — an AI-generated error that goes unchecked can have real consequences. For professional service firms in particular, the liability question around AI-assisted work is still being defined. This is an emerging risk that should be on every business owner’s radar.
• Reputational risk. If a client receives work product that was materially shaped by AI — and that work contains errors, reflects bias, or lacks the nuance of a human judgment — the relationship damage can be significant. The reputational cost of an AI-related mistake is difficult to quantify and harder to recover from than the direct financial loss.
• Opportunity cost and distraction. Time spent evaluating, implementing, and managing AI tools is time not spent on the core business. For leadership teams at smaller companies, this is a real trade-off that rarely gets accounted for in ROI calculations.
• Data and security exposure. Many AI tools are trained on or have access to data you input. Understanding what data is being retained, how it’s used, and what your obligations are under applicable privacy laws is a cost of compliance that grows as AI adoption deepens.

So What Should You Actually Measure?

The businesses seeing the clearest returns from AI aren’t necessarily using more tools. They’re measuring more deliberately. Here’s a simple starting framework:

• Total loaded cost. Subscriptions + token/usage fees + implementation + training time (at loaded labor cost). Most businesses are underestimating this by 30–50%.
• Time recovered vs. time redirected. Did AI actually free up capacity — and if so, what is that capacity being used for? Time saved on low-value tasks only creates ROI if it’s being reinvested in higher-value work.
• Error rate. Is AI-assisted output requiring more or less review than non-AI work? Track this by function.
• Revenue or margin impact. Has any specific AI initiative contributed to a measurable revenue gain, cost reduction, or margin improvement? If you can’t draw a line between the tool and a financial outcome, you’re measuring activity, not ROI.

The Bottom Line: Invest Thoughtfully, Measure Honestly

We’re not saying don’t invest in AI. The businesses that are pulling ahead right now are the ones building AI into their operations in deliberate, measurable ways — not the ones waiting on the sidelines.

But we are saying: approach AI spending the same way you’d approach any significant business investment. Know your fully loaded costs. Define what success looks like before you deploy. Track outcomes against a baseline. And don’t let the speed of adoption outpace your ability to manage the risks.

The ROI of AI in 2026 is real — but it’s not automatic. For businesses in the $5M–$50M range, the difference between capturing that return and simply accumulating costs usually comes down to financial discipline and clear-eyed measurement.

That’s exactly the kind of conversation a fractional CFO should be part of.

Ascend Accounting Advisory works with private companies in the $5M–$50M range as a fractional CFO and outsourced accounting partner. If you’re trying to make sense of your AI investment — what it’s costing, what it’s returning, and how to manage the risks — let’s talk.