Christian Gralingen
Artificial intelligence was supposed to dramatically change the corporate finance function. Forecasts would become more accurate and more frequent. Closing cycles would shorten. Risks would be identified earlier. Scenario planning would evolve from an occasional exercise into a continuous capability. On the basis of those optimistic predictions, many finance leaders have invested heavily in the technology.
However, when CFOs speak in private, a different story emerges. There are proofs of concept that never leave their sandboxes. Models that looked promising in pilot sit unused when the pressure of the quarter hits. Dashboards are produced and refreshed but rarely shape the decisions that matter most. Finance is undeniably busier and more automated but not obviously more forward-leaning in how it helps the organization adapt.
It is tempting to blame the shortfall on technology issues: The data quality is not there yet; the tools are not sufficiently integrated; the models are not trusted; the vendors overpromised. All of those factors matter, but they do not explain why similar AI technologies, introduced under broadly comparable conditions, lead to very different outcomes in finance than in other corporate functions.
After several years of working closely with CFOs and their teams as they tried to apply AI in practice, another explanation became hard to avoid: In many organizations, the technology is moving faster than the way leadership actually works inside the finance function.
When new tools arrive, people tend to talk, decide, and behave much as they did before. Attention gravitates toward getting the close done, explaining variances, defending a single forecast number, and treating deviations as errors to be corrected rather than signals to be explored. AI is introduced into that environment and expected to transform it. Most of the time, it does not.
To understand why, and what might be done differently, it helps to look less at technology adoption and more at leadership practice. (See “The Research.”) In studying the question of AI adoption in finance, we took a simple but demanding view of leadership: Leadership is not a job title or an individual trait; it is the work through which people help their organization adapt under uncertain and changing conditions. In finance, that work shows up every time someone reframes a question, tries a different way of seeing the numbers, surfaces an uncomfortable signal, or helps colleagues adopt a better routine.
Viewed this way, leadership does not sit only with the CFO or a small circle of direct reports. It can be exercised by an analyst who notices something unusual and asks, “What might this mean?”; by a controller who proposes a trial of a different forecasting driver; or by a planning manager who brings several futures into the conversation instead of converging on one.
Leadership becomes visible in the way practices are introduced, tested, and shared. It also becomes visible in who feels empowered to initiate and sustain that work. This view aligns with broader work on digital transformation that frames leadership less as top-down control and more as the orchestration of attention, accountability, and learning across the organization.1
This way of looking at leadership has a sobering implication for AI in finance. If leadership is understood primarily as the CFO’s personal competence or as the formal hierarchy’s right to decide, then the function’s capacity to experiment, learn, and embed new ways of working will always be limited. If leadership is instead understood as shared work in practice, then AI becomes an opportunity to reshape how that work is done. However, the very nature of finance work itself can raise challenges for AI adoption.
When Finance Is Pulled Into a Paradox It Did Not Choose
Finance has always lived with a tension between control and change. Its core mandate is to ensure reliability: accurate reporting, regulatory compliance, and disciplined stewardship of capital. Over time, the function has built processes, controls, and habits designed to reduce surprise. A great deal of finance’s professional ethos is shaped by the imperative of not being caught out.
AI introduces a different dynamic into this environment. It allows finance teams to see more, and earlier. It makes it possible to scan wider sets of signals, test alternative assumptions at low cost, and explore uncertainty in ways that were previously impractical.
The result is that finance is pulled more deeply into a paradox it did not choose. It remains responsible for being the organization’s safe pair of hands while at the same time being asked to become more curious, experimental, imaginative, and adaptive. Finance must protect what is working, even as it helps reinvent what may soon no longer work.
Some finance functions have learned to live with this paradox. They develop ways of working that keep discipline and exploration in constructive tension. Others fall to one side or the other: They either protect the familiar and treat AI as an efficiency add-on, or they embrace every new tool and struggle to make anything stick.
What makes the difference is not simply the tools they buy but the pattern of leadership work that emerges inside the function.
Across many engagements, we saw four recurring activities that particularly mattered for finance teams learning to work with AI: staying alert to what is changing, experimenting in practice, thinking differently about the future, and embedding what proves useful. These are not stages in a process. They are different ways that leadership shows up in everyday finance work. Here, we will present four vignettes, drawn from our research, that show how leadership work around AI takes shape in everyday finance practice. Details have been anonymized and, where necessary, combined to protect confidentiality, but each vignette reflects patterns we observed repeatedly across multiple organizations, rather than isolated or exceptional cases.
When Vigilance Becomes Shared Work
At a European manufacturing company, the central finance team had invested in a sophisticated data platform that provided access to a wide range of external market and supply chain indicators. Over time, the volume of available information increased, but much of it remained in the background. The data was technically accessible, yet it was rarely featured in the conversations that shaped plans or decisions.
That began to change when a financial planning and analysis manager proposed a small adjustment to how the team worked. Each Monday morning, two analysts were asked to bring one external signal they considered potentially important to a short discussion with colleagues. The conversation always started from the same question: “If this were the first sign of something bigger, what might it be?”
AI made it possible for the team to scan a much broader range of signals than before and to narrow that field to a manageable set for discussion. The more significant shift, however, was behavioral. Analysts and controllers began to see paying attention to early signals as part of their everyday responsibilities rather than as a specialist task. Over time, the team’s discussions started to influence budget assumptions and the way scenarios were framed for business partners.
No one on the team described this as leadership — yet it was. What changed was not the technology but the shared responsibility for noticing and interpreting what might be changing around the business. We observed similar practices across multiple finance teams. Where this shared vigilance took hold, AI was experienced as practical support embedded in everyday work rather than as an abstract promise. Where it did not, AI-driven signals tended to remain peripheral: available in dashboards, discussed in isolation, or quietly ignored when core planning routines took over.
When Experimentation Becomes Routine Rather Than the Exception
At an international consumer services company, the finance director had grown skeptical of large transformation projects that promised to reinvent planning with AI and delivered little beyond stress. Instead of launching another program, she encouraged her teams to think in terms of small, bounded trials that would generate insight without putting core processes at risk.
During one quarter, a business unit controller proposed running an explainable forecasting model alongside the standard statistical one. Using machine learning, the model suggested different revenue drivers for a specific product line. The trial was deliberately constrained: It would run for two sprints, it would not influence formal guidance, and success would be assessed not only by accuracy but by what the team learned.
The experiment did not outperform the existing model. What it did produce was a clearer understanding of which data the team actually trusted, where the conventional approach was more robust than expected, and where there might be room to rethink drivers in the next cycle. Crucially, no one was criticized for having wasted time. In the review meeting, the CFO asked a different question: “What did we learn that we could not have learned otherwise?”
That question quietly reset expectations around experimentation. Over time, teams began to treat trials not as projects to justify but as part of how finance learned. This mattered even more when generative AI tools became available.
In many finance functions we observed, the arrival of generative AI initially led to highly individual and largely invisible experimentation. People tried out tools on their own, unsure of what was acceptable, wary of failure, and sometimes concerned about what the technology might imply for their own role. Learning remained fragmented, insights stayed personal, and little of that experimentation translated into changes in shared routines.
In this team, the dynamic unfolded differently. Because experimentation had already been legitimized as collective work, generative tools were folded into the same discipline. Teams openly tested where AI-generated explanations or variance narratives genuinely improved shared understanding and where they merely added fluent but unhelpful noise. The tools were kept in a supporting role, helping teams reflect on results rather than replacing human judgment.
More broadly, we found that where AI remains peripheral, experimentation is treated as a temporary deviation from “real work.” Where AI begins to change practice, experimentation becomes part of how finance operates: disciplined, bounded trials carried out by people close to the business, using AI to learn what works, what does not, and where judgment must remain firmly human.
When the Future Becomes a Subject for Conversation, Not Prediction
At a regional utilities provider, the leadership team had always expected finance to deliver a single forecast that captured where the business was heading. As volatility increased in energy markets, the numbers they generated became harder to defend. After a series of painful forecast misses, the CFO tried a different approach.
Using AI-enabled scenario tools, the planning team constructed a handful of plausible futures for the next three years, each built around different combinations of input prices, regulatory decisions, and customer responses. Instead of producing one projection, finance brought several stories to the executive table, each with its own numbers and early-warning indicators.
The discussion shifted. Rather than debating which forecast was “right,” executives began asking, “What should we do if this scenario starts to materialize?” Finance’s role changed accordingly. It was no longer expected to predict the future with precision but to help the organization think through alternative futures and make deliberate strategic choices in the face of uncertainty.
AI made it easier to construct and analyze scenarios in greater depth. But the leadership move was to make it legitimate for finance to acknowledge uncertainty and to connect that uncertainty to concrete choices.
We observed a very different dynamic in finance teams that remained focused on prediction alone. There, AI was primarily used to refine a single forecast, optimize existing assumptions, and reduce apparent error. Paradoxically, this often increased people’s defensiveness. When models were challenged, teams responded by tightening assumptions rather than widening the conversation. Uncertainty was compressed into confidence intervals, and alternative futures were treated as distractions rather than inputs to decision-making. AI made the forecast more sophisticated but not more useful.
When Good Ideas Are Helped to Spread
At a large retail group, a finance team in one country had developed an AI-assisted routine for identifying unusual patterns in store-level expenses. It helped the team distinguish more quickly between genuine issues and benign anomalies. For some time, this remained a local success, known mainly to the people directly involved.
The practice began to travel when the organization created space for teams to share how they worked, not just what they delivered. During an internal learning session, a finance manager from another country heard about the routine and asked for the code and the checklist the team was using. Within weeks, a slightly adapted version was running in two more countries.
Crucially, this early diffusion was noticed and reinforced. Several months later, someone in the head office suggested integrating the approach into the retail group’s standard monthly review, with a small number of governance guardrails agreed to jointly with internal audit. That move did not mandate adoption, but it signaled that the practice was legitimate, safe to reuse, and worth building on.
Nothing in this process required a major program. The AI component itself was neither complex nor revolutionary. What mattered was that people saw it as acceptable to borrow and adapt one another’s ways of working, and that senior leaders took an active interest in how a good idea could become normal practice rather than remain a local innovation.
The Quiet Power of the CFO
Across all of these patterns, including shared vigilance, routine experimentation, strategic direction setting, and the spread of local practices, the CFO’s influence is both indirect and decisive.
Some of it is visible: The CFO decides where to invest, which initiatives to sponsor, which skills to hire for, which projects to stop. These choices shape what is possible. But another part of the role is less obvious and, in the context of AI, just as important: the tone they set about what counts as “real work” in finance.
When a CFO consistently asks only about accuracy and speed, people learn that the safest way to succeed is to avoid anything that might introduce uncertainty. When a CFO shows interest in what teams are learning from experiments, or in which weak signals might matter, people learn that thinking and trying are also part of their job. When a CFO insists that every pilot demonstrate a clear return on investment before it starts, experimentation dies in the planning stage. When a CFO is prepared to back a modest trial to see what happens, even if the payoff is not guaranteed, experimentation becomes possible.
Leadership in this sense is not dramatic. It is expressed in questions, in the allocation of a little time here and there, and in the willingness to protect a practice that is still fragile. Over time, those small acts accumulate. They determine whether AI finds a place in the real routines of the finance function or remains stuck in presentations and proofs of concept.
What the Numbers Can Reveal, if Read Differently
Our earlier work on digital finance maturity examined which organizational conditions tend to be present when finance teams make sustained progress with AI, such as ongoing experimentation and clear accountability for decisions supported by models.2 Our follow-up work showed why many teams nevertheless stall as AI expands: Those conditions often fade in day-to-day routines, even when the technology itself continues to improve.3
The evidence in this article adds another layer. It shows how those same conditions are not abstract capabilities but rather the result of everyday leadership work inside the finance function. Whether experimentation, accountability, or learning persists depends less on formal design choices than on how people are encouraged, protected, and listened to in practice — and on how those same people, in their everyday work, help shape and carry new practices forward. The constraint, in other words, is how leadership work takes shape.
Finance leaders are used to interpreting diagnostic surveys on digital maturity or AI readiness at face value. Read that way, such data shows where functions have invested, which tools they have adopted, and how they assess their own progress.
There is another way to read the same data: as indirect evidence of where leadership work is, and is not, taking place.
When external data rarely influences planning discussions, the issue may not be data availability but whether shared vigilance is part of everyday work. When many organizations report running pilots but few of them see changes in core forecasting or planning routines, the problem may lie less in experimentation than in the ability to turn local learning into shared practice. And when finance professionals say they have access to advanced tools but hesitate to surface the uncertainties those tools reveal, the constraint is often one of leadership tone and permission rather than technical capability.
It is important not to overstate what such numbers can prove. They do not establish causal relationships between specific leadership behaviors and AI outcomes — but they do reveal patterns. Read carefully, they make it plausible that differences in how leadership shows up in everyday practice help explain why similar technologies produce very different results across finance functions.
Changing How Leadership Works Inside Finance
AI will continue to advance. Tools will become more accessible. Vendors will refine their offers. Regulatory expectations will grow. None of that guarantees that finance will transform. What will matter is whether finance functions change how leadership works.
If leadership remains concentrated in a few places and focused primarily on protecting existing routines, AI will mostly be used to continue old ways of working, only faster. If leadership is understood as shared work on practice, such as watching the horizon, trying things in small ways, shaping strategy under uncertainty, and helping good practices spread, then AI can become a powerful ally in reshaping what finance does and how it contributes.
CFOs do not need to brand any of this as a new leadership model. They do not need to talk about frameworks at all. They do need to ask themselves some plain questions:
- Who in our function feels responsible for noticing and discussing early signals?
- Where do people feel safe experimenting with new ways of forecasting or analyzing risk?
- How do we encourage thinking about alternative futures rather than defending a single number?
- What do we do to ensure that local innovations become shared practice when they prove their worth?
- What am I, as CFO, doing day to day to make those forms of leadership more likely, or less?
The latter question is perhaps the most important. But until the answers to all of those questions begin to change, AI will keep pressing against the glass of the finance function without fully entering the room. When they do change, the technology that once felt like an external pressure will start to feel more like an instrument that teams can pick up and use as they learn together how to work differently.