AI is changing finance fast—but most companies are adopting it wrong. Here's what CFOs need to know about AI's real limitations and risks.
AI is changing finance fast—but most companies are adopting it wrong. Here's what CFOs need to know about AI's real limitations and risks.
Somewhere in the middle of all this AI noise, your team wants to know how they can start using it.
Maybe your senior analyst is running financial narratives through a generative AI tool before sending them up the chain.
Maybe it's your FP&A lead using an AI-assisted forecasting platform that came bundled with your ERP upgrade.
Maybe it's something as simple as Copilot quietly sitting inside Excel, autocompleting formulas and summarizing spreadsheets for your bookkeeping team.
Most of that….
It's probably happening without a governance framework, C-suite sign-off, and without any guardrails or understanding of where AI outputs end and human judgment begins
So what is the CFO’s job in the age of AI?
It's not to become a data scientist (although, a cool background).
It is surely true, though, to make sure the organization doesn't mistake a confident-sounding algorithm for a sound financial decision.
And that’s actually a harder job than it sounds, because AI is better at being confidently wrong than it is sheepishly right.
This guide cuts through the hype to focus on what actually matters: what AI does well in finance, where it quietly fails, and how finance leaders can build the oversight framework that keeps it from becoming a liability.
The problem here is the pace, not that of the technology but the speed of adoption without a proper governance mechanism in place.
For finance functions, pressure is mounting from boards who want real-time information. Agentic AI is scaling faster than guardrails that can be put in place.
At the same time, finance professionals face a stream of pressure
Pressure is mounting from management wanting quicker closure.
Pressure is mounting everywhere to do more with less, and AI technology delivers exactly that, and it’s being adopted by analysts, controllers, and FP&A at a much faster rate than most CFOs are aware.
The result is a disjointed AI implementation with inconsistent tools, inconsistent input data, inconsistent accuracy, and inconsistent risk, all without one common approach to validate the output results before action is taken.
This is not a technological problem; this is a leadership problem, and it falls on the shoulders of the CFO.
Whereas companies that implement good AI governance reduce their risk, companies that forego AI governance build on quicksand.
Before building any sort of governance strategy, its imported to be clear-eyed about AI capabilities and its risks. AI is genuinely powerful in certain areas of finance. In others, it's producing outputs that look authoritative but require significant human scrutiny.
AI can look very promising on paper. It is certainly possible to derive valuable insights from forecasting as well as data analysis with the help of artificial intelligence, yet some issues are often left unspoken by those selling the solution.
The first one is the matter of data quality, because the input into an algorithm dictates its output. The AI will work based on the data provided and will yield highly precise predictions, which may or may not be relevant to reality if your historical information is faulty, lacks precision, or consistency. Precision does not mean accuracy.
The model used in the forecasting process is calibrated to historical data and, in this case, may lack accuracy once put to use in the current economic climate. This is because if AI is working within parameters that no longer exist, its modeling won’t accurately reflect reality. Scenario modeling allows stress testing AI forecasts under circumstances that it has not encountered before, but not all do it.
Finally, confidence in AI-driven results is another aspect of the problem: while an algorithm produces the same amount of certainty in both situations, the person working with the information is not always clear about his or her situation.
It’s just what it sounds like. Using AI to solve problems is like looking at a black box. We don’t know exactly how it pulled the solution it did.
This is a common occurrence.
An AI application makes a prediction or detects something unusual, a decision is made on the basis of that, and after six months, somebody wonders why. Nobody knows how the model reached its conclusions, and the audit committee isn’t pleased.
Transparency is no longer just a matter of corporate governance. It’s becoming a matter of regulatory compliance as artificial intelligence becomes an increasingly pervasive presence in financial reporting and risk management. CFOs will have to justify their models to regulatory bodies, external auditors, and the board, and “because the model told me to” won’t cut it.
As CFOs investigate AI applications, they need to ask probing questions regarding explainability.
Generative AI, aka LLM’s like ChatGPT and Claude, have seemed to have entered finance through the side door. These types of AI produce written text, but still have implications within finance and across broad business use cases.
It is used to generate board packages, management commentary, and variance analysis. While the result generally seems well articulated, there could be hidden flaws in its logic.
This is a serious but hidden problem. Unlike other tools, generative AI technology is not aware of your business. Instead, it relies only on language models to build up a coherent narrative, including a plausible reason for a decline in gross margins in Q3, which may be completely off track.
Generative AI also suffers from ‘yes man’ syndrome. They are designed to agree, and to convince you they really feel that way even if the data does not support your decision. Thus, AI cannot be fully trusted to make decisions through talk points alone.
It is often the case in many organizations that the ultimate gatekeeper between the AI-produced narratives and boards or shareholders is the CFO’s function.
It’s not always easy to know what AI tools are right for your business in a market that's crowded with new entrants constantly. However, there are some steps you can take to ensure the product you are choosing fits your business needs and does not do more harm than good.
The last question matters more than it gets credit for. Several finance AI tools built on third-party large language models are subject to changes in those underlying models that the vendor doesn't control. An output that works a certain way today may work differently in six months.
The first red flag to watch out for when vendors talk about their tools is a focus on demo versus actual implementation. Demos rely on nicely structured data. Your data, on the other hand, will not be nearly as neat. The disparity in performance between demo and actual implementation is what causes most disappointments with AI.
You should also avoid vendors that fail to adequately describe how the technology works when asked about explainability. Saying that "the technology analyzes the data and generates insights" will do little more than raise an eyebrow. What is the technology doing, and is it capable of providing explainability?
It is not because the AI tool that mid-market businesses use in finance management is poor; it is simply that there are five good tools, each of which alone works well, but when used together form a fragmented data environment leading to poor results across all outputs.
The key consideration should be: will this tool contribute to improving the information infrastructure of finance operations or will it simply be a data silo? AI tools that can integrate easily with your ERP/Accounting System create more value, while those requiring manual exports often create more problems than they solve.
The talent dimension of AI adoption is where most CFO-level planning falls short. The question isn't just which tools to adopt — it's how the finance team needs to evolve to use those tools well.
AI changes the demand curve for finance skills in specific ways. It reduces the premium on tasks involving data collection, routine reconciliation, and standard report formatting. It increases the premium on tasks requiring judgment, context, and communication — interpreting outputs, identifying when a model is misbehaving, and translating analytical results into strategic recommendations.
That shift means the finance professionals who thrive in an AI-augmented environment are the ones who combine technical competence with strong business judgment. They can interrogate a model's output rather than just consume it. That's a reskilling challenge, not just a hiring challenge, and it requires intentional investment.
And can we be honest? A lot of finance teams are running with talent gaps that AI adoption is going to expose rather than fix. Plugging an AI forecasting tool into a team that doesn't have strong foundational FP&A discipline doesn't produce better forecasts — it produces faster bad ones.
This is where fractional CFO support creates disproportionate value. A seasoned financial leader who has already navigated AI adoption across multiple organizations can accelerate the process significantly — both in terms of selecting the right tools and building the team capabilities to use them well.
The organizations getting real value from AI in finance share a few consistent characteristics.
And they treated AI as a capability multiplier for strong finance teams, not a substitute for them.
The takeaway? Begin with a single friction area, leverage accurate data and tangible success metrics, conduct a rigorous test, measure, and consider scaling only after establishing governance processes and the right skills.
And then, of course, there is ROI.
While vanity metrics such as the number of adopted technologies or hours of paper reduction don't prove anything about AI's impact on financial decision-making, real metrics to focus on include forecast accuracy over time, close cycle length, and quality of analysis provided to senior leadership. They are difficult to track, but much more relevant.
AI in finance will be integrated into the built-in capabilities of ERP systems, financial planning applications, and other tools. Very soon, the choice between using or not using AI in your company will disappear — just like many others have in other industries.
The early adoption of governance models and the development of AI-related capabilities in your finance team will give you a serious head start when the time comes. Waiting to address these challenges will only mean having less time to do it, more pressure to deliver results quickly, and fewer opportunities to experiment responsibly.
So what's the message for finance leaders?
You are not supposed to be an AI specialist. But it would help a lot to understand how AI impacts the most important aspects of your role – namely, forecasting, reporting, and analytics.
This is exactly where experienced financial leadership transforms how you operate. Whether through interim CFO support during a critical technology transition, fractional CFO partnership for ongoing strategic oversight, or targeted finance team development that builds AI literacy from the ground up — the right guidance doesn't add complexity; it removes it.
Ready to turn AI complexity into a strategic advantage?
The difference between knowing about AI in finance and actually leveraging it comes down to having the right expertise when you need it most. Let's talk about how McCracken can help you navigate these decisions with confidence.
CFOs should not only choose an AI solution but also ensure that the impact of AI-produced results on finance activities is controlled properly. Thus, CFOs should define the principles of data governance for AI, determine responsibility for AI's results, and give finance teams the ability to verify AI-generated outputs.
Some main threats associated with the introduction of AI into the finance function are inaccurate data with a high level of precision, issues with model explainability in case of auditing, and false narratives provided by AI due to incorrect assumptions. For safe implementation of AI, CFOs need to make sure that validation processes will be put in place.
Certainly not – because the CFO's role requires more than just technical expertise. AI will certainly automate most routine operations and enable rapid analysis of large data sets, but it cannot provide necessary insights and strategic guidance to leadership. CFOs will still be indispensable and just evolve.
To evaluate an AI tool for the finance function, CFOs should take into account their organization's data requirements and how much it depends on the data quality, explainability of results, compatibility with current systems and workflow, and the workload required for implementation. Be careful with vendors providing only demos and not explaining models' work or outputs' validation.
.svg)
.svg){kind=icon}