AI Won’t Take Your CFO’s Job

But it will change the CFO role. Here’s how.

By Chris Fenster, Founder & Executive Chairman

The Radiologist Paradox

In 2016, Geoffrey Hinton (the “Godfather of AI”) declared that hospitals should “stop training radiologists now.” At the time, his logic seemed airtight: if machines could read X-rays faster and more accurately than humans, why would we need expensive, highly trained doctors?

Nearly a decade later, radiology residency programs are at record capacity. Radiologist salaries have jumped almost 50% since 2015, and demand for human radiologists is higher than ever.  

What happened? And more importantly, what does AI in radiology have to do with the future of finance?

The story of radiology offers a surprisingly precise preview of what’s coming for finance. Both fields work with structured data. Both involve pattern recognition. Both have clear benchmarks for accuracy. And both are being transformed by AI … just not in the way the doomsayers predicted.

The AI Reality Few People Anticipated

It turns out that lab accuracy doesn’t equal field accuracy. AI models trained on standardized test images struggled in real hospitals with real patients. Stakeholders demanded accountability that machines couldn’t provide (malpractice insurers refused to cover autonomous AI diagnoses.) And radiologists spend only about a third of their time actually reading scans; the rest involves consulting with other physicians, explaining results to patients, and reviewing complex cases where context matters.

The Era of Context

The radiology paradox points to a deeper truth that’s now shaping the AI landscape: context is the new moat. Aaron Levie, CEO of Box, recently articulated this shift in what he calls “The Era of Context”:

“AI models are trained on public or generally available data sources. By default, they know basically everything about anything, other than your specific workflows and business. For AI agents to be effective in the enterprise, they need your enterprise context.”

This is the insight that separates transformative AI deployments from expensive experiments. Generic AI tools can generate a P&L statement from transaction data with impressive speed, but ask one to explain why gross margins compressed in Q3 when product mix shifted, a key supplier raised prices mid-quarter, and you’re strategically investing in a new channel you’ve discussed extensively with your board, and suddenly you’re debugging hallucinations instead of prepping for your board meeting.

Dharmesh Shah of HubSpot called context graphs “a beautiful idea that will matter eventually”, then pointed out that asking companies to capture decision traces right now is like asking someone to install a three-car garage when they don’t own a single car. This is because most companies are still struggling with basic data unification: they’re so early in their AI adoption that a three-car garage feels like overkill.

He’s right about the readiness problem, but for the companies we serve—venture and growth-stage companies who’ve chosen embedded teams rather than mostly W2s for their finance function—the calculus is different. It makes zero sense for them to try to build what we can provide—faster, better, and more securely at a fraction of the cost. The context graph isn’t something they need to construct themselves; it’s something that accumulates naturally when you’re consistently present at decision time, across enough similar decisions, that patterns emerge.

The Output Gap

The Deloitte Q4 2025 CFO Signals survey confirms that AI is now a boardroom priority: 87% of CFOs believe AI will be extremely or very important to their finance operations in 2026. But here’s the catch—only 21% of active AI users report clear, measurable value, and just 14% have fully integrated AI agents into finance functions.

This isn’t a technology problem. It’s a context problem. The bottleneck has shifted. The question is no longer, “Can AI do this task?” because the answer is increasingly yes. The new bottleneck is whether you can vet AI’s outputs.

This is what we call the output gap: the lag between what AI can theoretically do and what organizations can actually deploy. Though an MIT study last year found that 95% of enterprise AI initiatives fail, it also highlighted what the successful 5% are doing right: they’re focusing on human enablement, strategic alignment, and disciplined execution.

McKinsey’s CFO Yuval Atsmon put it bluntly: AI can now handle up to 30% of tasks—including faster research and better summarization—but “you can’t really do a full strategic analysis yet.” The constraint moved from, “Can we afford to do this task?” to, “Do we have anyone who can tell if this task was done correctly?”

The firms that will capture value in this environment aren’t the ones with the most sophisticated AI. They’re the ones with the deepest context, the most embedded relationships, and the human judgment to bridge the gap between what AI outputs and what decisions require.

The Unautomatable

So: what can AI never replace? Research from The Secret CFO identifies exactly three zones where human expertise remains irreplaceable.

High-stakes decisions. The judgment calls that define company trajectory: when to raise, when to cut, when to bet big. These require synthesis of incomplete information, risk tolerance calibration, and accountability that algorithms cannot bear.
Personal relationships. The trust with investors, the credibility with boards, the rapport with operators. The test of a good relationship is whether you have enough credit with stakeholders to ask for something highly unreasonable, and get it. AI can’t build that credit.
Accountability. Someone must own the outcome. When the forecast is wrong, when the close is late, when the board loses confidence, a human must stand accountable. This is the “irreducible core” of executive function.

Of these three, relationships are the most durable. “Even decision-making skills and accountability could become more commoditized eventually,” notes The Secret CFO. “But the quality of your internal and external relationships will forever be unique to you.”

The Iron Man Suit, Not the Terminator

This brings us to a critical distinction about how to think about AI in finance. There are two models:

The Terminator model: AI replaces humans entirely

The Iron Man suit model: AI amplifies human capability

Siqi Chen, CEO of Runway, frames this well:

“Instead of AI finance tools being a godlike Terminator thing, what if it felt more like an Iron Man suit? It helps you make the complexity of the business more understandable so that you can develop a better intuition of that complexity yourself.”

This distinction matters because it points to an emerging truth that Andrej Karpathy, former Tesla AI Director, describes as, “We’re summoning ghosts, not building animals.” Generic AI tools are powerful statistical distillations of human knowledge, but they’ve never actually done a month-end close or navigated a difficult board conversation. They’ve read about these things; they haven’t lived them.

The missing ingredient is grounded context that accumulates over time and generates continuous feedback. This is why the future belongs not to AI that replaces humans, but to human agent teams: professionals who combine deep domain expertise with the ability to orchestrate AI agents, interpret their outputs, and exercise judgment where it matters most.

The Two Paths Forward

The finance profession is bifurcating into two distinct paths. This isn’t new: I’ve written about the talent paradox that makes most first finance hires structurally mismatched for the stages that follow. The AI era accelerates this pattern.

The Super Controller is architecture-minded, data-literate, automation-obsessed person. They direct agents toward the right problems, validate their work, handle exceptions, and maintain process excellence. They’re capable of delivering 10X the impact of a traditional finance professional through operational leverage.
The Super CFO is relationship-native, judgment-oriented, and influence-focused. They concentrate human effort in the unautomatable zones: high-stakes decisions, personal relationships, and accountability. Their value increases as AI handles the mechanical work, freeing them for what matters most: interpreting signals, advising leadership, building trust with investors, and shaping strategy.

These are two distinct talent pipelines with different raw materials. Super controllers think in systems, workflows, and scalable processes. Super CFOs think in relationships, influence, and strategic context. Both are essential. Both become more valuable in the AI era. But confusing one for the other—hiring a controller and calling them CFO, expecting a super controller to evolve into a super CFO—is a recipe for the stage mismatch problem that derails so many growing companies.

As Box’s Aaron Levie puts it:

“AI agents are a force multiplier for your skill level in any particular field. Those that deeply understand their field will always have the leg up because they understand what’s happening behind the scenes, they know when to intervene with the agent, they can make judgment calls about what works and what doesn’t.”

The Radiology Paradox, Revisited

The radiology story gets really interesting when you look at what happened after hospitals adopted digital imaging in the early 2000s. Radiologist productivity jumped dramatically: 27% for plain radiography, and 98% for CT scans. Yet no radiologists were laid off. Instead, something unexpected happened: utilization rates for imaging increased 60% over the next eight years.

This is Jevons Paradox: efficiency gains lead to increased consumption. When something becomes faster and cheaper, you do more of it. Before digital systems, the median reporting time for X-rays was 76 hours. After digitization, it dropped to 38 hours. Suddenly, CT scans that were once reserved for exceptional trauma cases became routine.

The same dynamic is emerging in finance. As AI handles more of the mechanical work, finance teams shift toward higher-value activities: strategic guidance, scenario modeling, investor relationship management, and board preparation. The monthly close call transforms from a one-hour retrospective into a strategic financial review powered by real-time analytics and forward-looking insights.

This isn’t about doing the same work faster; it’s about doing fundamentally different work: work that couldn’t exist when humans were constrained by bandwidth.

The Future: Finance-grade AI

Here’s what generic AI vendors miss: technical capability is only half the equation. Finance demands what we call “finance-grade AI”: not the probabilistic, 90%-accurate consumer chatbot, but systems built to meet deterministic requirements:

Lineage: Where did this number come from?

Detail: What transactions comprise it?

History: How has it changed?

Reconciliation: Does it tie to source systems?

As AI systems generate more financial insights, the CFO’s role evolves from decision-maker to decision auditor: the executive responsible for ensuring governance frameworks validate AI outputs, checking for bias, and maintaining accountability. This reframes human oversight from limitation to elevation. The CFO isn’t slowed down by governance requirements; they’re the quality assurance layer that makes AI trustworthy.

The Embedded Advantage

So how do you build the context that makes AI valuable? This is where the radiology analogy becomes most instructive.

Not every hospital benefited equally from AI advances. The ones that succeeded did something specific: they combined cutting-edge technology with deeply embedded clinical expertise. They didn’t replace radiologists with algorithms; they built systems where AI handled the pattern recognition while humans provided context, relationships, and accountability.

The same principle applies to finance. The future doesn’t belong to software companies that treat AI as a product feature or consultants who parachute in for projects. It belongs to embedded partners who accumulate context continuously across hundreds of concurrent engagements: who sit at the decision-making table and gain visibility into strategy, execution, and the nuances that determine outcomes.

This embedded advantage creates something no stand-alone tool can replicate: pattern recognition from cross-company data, institutional knowledge that compounds rather than resets with each engagement, and the trust that comes from multi-year partnerships through both growth and crisis.

The Inflection Point

The CFO function is experiencing what Andy Grove called a “strategic inflection point”: a transformation where the old rules stop working, and new rules haven’t fully formed yet.

Some firms will try to defend the status quo, billing by the hour as AI makes hours worth less. Others will adopt generic AI tools that fail to deliver because they lack the context and change management to drive adoption.

The winners will be those who recognize this isn’t a technology problem or a people problem; it’s a transformation that requires both, together: the analytical power of AI, combined with the judgment, relationships, and accountability that only humans provide.

Later this month, Raymond will be sharing more about how we’re bringing this vision to life: a platform that unifies everything we’ve built over 17 years into something our clients will experience daily. It’s the most significant client-facing initiative in Propeller’s history, and it’s built on a simple thesis: the best AI isn’t the smartest, it’s the most embedded.

The radiologists who thrived weren’t the ones who denied change or the ones who feared it. They were the ones who figured out how to harness it, combining machine precision with human judgment to deliver outcomes neither could achieve alone.

Your financial statements are already better when AI reads them. The question is: what will you do with that advantage?

Want to explore what finance-grade AI looks like for your company? Let’s talk.