There is a tempting story unfolding in boardrooms right now: software is no longer just software. It is starting to behave like labor.
That is the real shift behind the move from traditional SaaS to AI-enabled SaaS. Companies are no longer buying tools that simply help people work faster. Increasingly, they are buying tools that promise to do portions of the work itself.
On paper, that sounds like a margin miracle.
In practice, it is often not.
The reason is simple: AI does not work for free. It may not show up in payroll, but it absolutely shows up in cost structure. Leaders still have to pay for models, credits, integrations, security controls, workflow design, testing, monitoring, and human oversight. Even the most powerful AI tools require thoughtful implementation. “Push a button and the robot figures it out” is not a finance strategy. It is wishful thinking.
That matters because many companies are layering AI costs on top of existing labor costs instead of redesigning work around it. When payroll stays flat and AI spend rises, margins can actually get worse before they get better. That makes this a CFO conversation, not just a technology conversation.
The opportunity is real. KPMG’s global research found that 71% of companies surveyed are already using AI in finance operations, and in the US, 92% of surveyed companies said their finance AI initiatives were meeting or exceeding ROI expectations. But the same body of research also makes clear that adoption is happening across specific finance activities such as accounting, planning, treasury, risk, and tax—not as magic, but as targeted workflow change.
So the right question is not, “Where can we use AI?”
The better question is, “Which work should be done by AI, which work should remain human, and what happens to the economics of the business when we change that mix?”
AI is changing the unit economics of work
Traditional SaaS generally improved productivity by helping employees do their jobs better. AI-enabled SaaS increasingly goes further by taking on task execution itself: drafting, summarizing, reconciling, triaging, forecasting, classifying, and routing.
That creates a different economic model.
Instead of paying only for seats, companies may now pay for usage, credits, tokens, workflow runs, compute, oversight time, and exception handling. In other words, labor is moving from humans to machines—but the machine labor still has a price tag.
This is why CFOs should resist the common trap of calling AI “efficiency” before they have actually costed it. A $20 monthly tool can become much more expensive when heavy usage, premium models, implementation support, API calls, data infrastructure, and review time are added to the equation. At enterprise scale, AI spending is not theoretical. It has become a major line of investment, and organizations are still learning how to govern and measure it.
At the same time, the upside is significant. McKinsey estimates that generative AI could add $2.6 trillion to $4.4 trillion annually across use cases, with especially large value pools in customer operations, marketing and sales, software engineering, and R&D. But McKinsey also notes that realizing that value requires changes to skills, workflows, and operating models—not just tool adoption.
That is exactly where finance leadership should step in.
The first rule: automate the tedious, not the critical judgment
A practical starting point is this: the best early AI candidates are usually the most repetitive, rules-based, time-consuming tasks—especially the ones that already frustrate your team.
Think:
- first-pass document summaries
- data extraction and categorization
- recurring reporting drafts
- account reconciliations with defined rules
- ticket triage and workflow routing
- FAQ response generation
- variance flagging
- transcription, note cleanup, and follow-up drafting
Those are not necessarily low-value activities. But they are often low-judgment activities.
That distinction matters.
Research on AI and productivity consistently shows that gains depend on the type of task, the quality of implementation, and the surrounding human system. OECD notes that AI has strong productivity potential, but the outcomes depend on complementary investments and are still subject to uncertainty. Research from NBER found meaningful productivity gains from generative AI in customer support, with the biggest improvements going to less-experienced workers rather than top performers. In other words, AI often works best as a force multiplier inside defined tasks—not as a replacement for expert judgment.
For finance leaders, that means AI is best deployed first where:
- the process is repetitive,
- the rules are reasonably clear,
- the cost of error is manageable,
- outputs can be reviewed,
- and the work consumes enough human time to matter.
Where humans should stay firmly in the loop
Not every function should be handed to AI just because it can be.
High-stakes decisions still require human ownership, especially when the work involves:
- judgment under uncertainty
- regulatory or compliance interpretation
- client trust
- pricing strategy
- material financial approvals
- exception handling
- scenario tradeoff decisions
- final signoff on sensitive outputs
That is especially true because organizations still report significant concerns around inaccuracy, privacy, cybersecurity, compliance, and intellectual property. McKinsey’s survey found that companies vary widely in how much human review they apply to gen-AI outputs, and many are still actively mitigating these risks.
A good finance rule of thumb is this: The higher the consequence of being wrong, the higher the required human involvement.
AI can assist. It can accelerate. It can draft. It can surface patterns. But if the decision has material financial, legal, or reputational implications, the human should still own the conclusion.
A CFO framework for deciding AI vs. human work
For leadership teams trying to get practical, here is a five-step CFO framework.
- Map the work before you buy the tool.
Break processes into tasks, not job titles. “Payroll,” “customer success,” and “FP&A” are too broad. Identify what people are actually doing minute to minute. - Sort tasks by judgment, risk, and repetition.
The sweet spot for AI is usually high repetition, low-to-moderate judgment, and controllable risk. The more ambiguity or consequence involved, the more human review you need. - Fully cost both sides.
Do not compare AI license cost to salary alone. Compare the full cost of human execution versus the full cost of AI execution, including implementation, model usage, credits, integration work, oversight, security, quality assurance, and exception handling. - Define the accuracy threshold.
How right does the output need to be? 80% may be useful for a first draft. It is unacceptable for compliance reporting or financial signoff. Accuracy requirements should determine whether AI drafts, assists, recommends, or executes. - Reallocate labor intentionally.
If AI makes a human role more efficient, the financial gain does not appear automatically. Leadership has to decide what happens next: reduce labor need, increase output, improve service quality, or redeploy talent to higher-value work. If none of that happens, AI becomes an added cost layer instead of a margin lever.
That final point is where many companies stumble.
Efficiency without labor redesign is not margin expansion
This is the uncomfortable truth: AI can improve productivity without improving profitability.
If a team saves 20% of its time but headcount, workflows, and operating expectations stay exactly the same, then the company has not captured economic value yet. It has only created latent capacity.
That is not bad. In many cases, that capacity should be reinvested into higher-value work: deeper analysis, better client support, more proactive forecasting, stronger controls, faster iteration, or new revenue-supporting activity.
But leadership must choose.
Otherwise, the business ends up paying for both the old labor model and the new AI layer.
The CFO’s role is not to slow AI down. It is to make it real.
A strong CFO is not the person saying “no” to AI. A strong CFO is the person forcing clarity around the economics.
What work is actually being automated?
What is still being reviewed?
What is the total cost per task?
Where is margin expanding—and where is spend simply shifting from payroll to software and credits?
What new capabilities are required in security, governance, and process design?
And how will human roles evolve once AI removes part of the workload?
Those are the questions that turn AI from hype into operating leverage.
Because the companies that win this next phase will not be the ones that buy the most bots.
They will be the ones that understand the true cost of machine labor, redesign work accordingly, and use human talent where judgment creates the most value.
References
- KPMG, Global AI in Finance Report and US AI in Finance findings. https://kpmg.com/xx/en/our-insights/ai-and-technology/kpmg-global-ai-in-finance-report.html
- McKinsey, The Economic Potential of Generative AI. https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier
- OECD, The Impact of Artificial Intelligence on Productivity, Distribution and Growth. https://www.oecd.org/content/dam/oecd/en/publications/reports/2024/04/the-impact-of-artificial-intelligence-on-productivity-distribution-and-growth_d54e2842/8d900037-en.pdf
- Brynjolfsson et al., NBER, Generative AI at Work. https://www.nber.org/papers/w31161
- McKinsey, The State of AI: How Organizations Are Rewiring to Capture Value. https://www.mckinsey.com/~/media/mckinsey/business%20functions/quantumblack/our%20insights/the%20state%20of%20ai/2025/the-state-of-ai-how-organizations-are-rewiring-to-capture-value_final.pdf
- KPMG, Artificial Intelligence in the financial sector. https://kpmg.com/de/en/insights/digital-transformation/artificial-intelligence/artificial-intelligence-in-finance.html
Top 5 FAQ: AI Is Not Free Labor
1. Is AI actually cheaper than human labor?
Not always. While AI can reduce time spent on tasks, it introduces new costs such as model usage, API credits, integrations, implementation, monitoring, and human oversight. Many companies see costs rise initially because AI is layered on top of existing labor instead of replacing or redesigning it.
2. How should CFOs measure the ROI of AI?
CFOs should evaluate AI by comparing the full cost of human work vs. the full cost of AI execution—not just salaries vs. software licenses. This includes implementation, usage, security, quality control, and exception handling. ROI is only realized when AI leads to measurable changes in output, cost structure, or labor allocation.
3. What types of tasks should be automated with AI?
AI is most effective for repetitive, rules-based, and low-judgment tasks such as data extraction, reporting drafts, summarization, and workflow routing. These tasks consume time but require less critical thinking, making them ideal for automation with human review.
4. What work should remain human instead of AI-driven?
High-stakes work should remain human-led—especially tasks involving judgment, compliance, financial approvals, client trust, and strategic decision-making. The higher the risk or consequence of being wrong, the more important human oversight becomes.
5. Why doesn’t AI always improve profitability?
AI can increase productivity without improving margins if companies don’t adjust their workforce or workflows. If labor costs stay the same while AI costs are added, profitability may decline. Real financial impact only occurs when leaders intentionally reallocate labor, reduce costs, or increase output based on AI efficiencies.