SaaS CFO Says AI Improves Finance Judgment Only When Leaders Stay Inside the Workflows
ithin Shetty explains how CFOs can use AI to improve financial judgment by placing it inside workflows while keeping accountability and final decisions human.
Nithin Shetty
CFO at Scrut Automation
Workflow Design: CFO uses AI after personal analysis to challenge variance explanations, stress-test scenarios, and draft communications.
Human Judgment: AI supports finance decisions, but capital allocation, fundraising, relationships, and final board messaging remain human responsibilities.
Failure Modes: AI can produce plausible errors, miss company context, and weaken forecasts during volatile startup transitions.
Team Shift: Finance hiring increasingly prioritizes skepticism and interpretation over production skills, moving controllers toward review and challenge.
CFO Advantage: Hands-on AI adoption can improve strategic thinking, but careless deployment risks governance failures and weaker leadership judgment.
Nithin Shetty is CFO at Scrut Automation, where he oversees treasury, FP&A, cross-border compliance, and revenue operations simultaneously. He started there when revenue was at $500k ARR, and now it's at $20M ARR.
In this conversation, Nithin shared how he uses AI at the right times in the right workflows so that his judgment and accountability are maintained.
Not Approaching the AI Transformation as an Observer
My name is Nithin Shetty. I am a Chartered Accountant and CFO who has spent the last decade building finance functions from zero inside high-growth, VC-backed SaaS companies.
Create a Free Account to Read More
You'll also get access to a growing community of modern CFOs and finance executives accessing proven frameworks, tools, and insights to navigate AI-driven finance.
Have an account? Log In
My journey was shaped less by a linear career path and more by repeated exposure to chaos at scale. I started with three ventures during my CA articleship, spanning manufacturing, FMCG, and seafood supply, not strategically, but because I could not separate thinking about business from running one. That instinct never left.
After qualifying as a CA, I took on fractional and interim CFO mandates across five industries, including manufacturing, SaaS, mobility, media, and EdTech. Each presented a distressed business model, requiring me to quickly build financial clarity and make consequential decisions with incomplete information. This remains the core of what I do.
Today, I'm CFO at Scrut Automation, a GRC and compliance SaaS company. For three years, my entire operating context has been a business that helps enterprises manage risk and automate compliance, and AI is now restructuring every assumption that business was built on. I have watched that disruption from the inside while simultaneously running the financial infrastructure that funds it.
I am not approaching AI transformation as an observer. I have lived inside it at the company level, watching how AI changes product economics, sales cycles, customer expectations, and unit economics in real time.
Organizational Complexity
My finance organization is part of Scrut Automation, a Series B GRC and compliance SaaS company backed by Lightspeed, MassMutual Ventures, and Endiya Partners. We scaled from $500K to $20M ARR over three and a half years, requiring the finance function to grow and restructure multiple times within a single tenure.
I lead a 20-person operation across five verticals. The core Finance team has 12 people, supported by 3 in Revenue Operations, 2 in CS Operations, 1 in Legal and Compliance, and 2 in Payroll and Tax. My direct reports are a Financial Controller and an FP&A Lead. Company Secretarial and Legal services are handled by a structured retainer outside the headcount.
Our operations have cross-border complexity. We operate across an Indian holding company and a US subsidiary. I manage consolidated reporting, transfer pricing, intercompany settlements, US GAAP to Ind AS reconciliation, Delaware compliance, and state tax filings as part of our regular operating rhythm. Revenue recognition covers USD and INR, spanning subscription ARR, professional services, and usage-based components, structured under both Ind AS 115 and ASC 606.
The scope makes this function slightly unusual for a Series B company. Revenue Operations and CS Operations report to finance, not by convention, but because I was the single owner of performance data across systems. Owning GTM and customer health data made it operationally logical to also own those functions. This structure has held and proven its value across every board cycle.
How AI Reduces Drafting and Interpretation Time
Before systematically using AI, financial communications drained the most time from my calendar without proportionally improving output. Writing a board update, an investor memo, or a management commentary on monthly numbers required context-switching from analysis to writing. The drafting process itself consumed time that should have gone into the underlying thinking.
I started using Claude as a drafting and interpretation layer for financial communications and FP&A outputs. I would complete the analysis, then use AI to structure the narrative, stress-test the logic of my number presentation, and check whether the underlying data supported my story. I also began using it for first-pass contract and document reviews, which reduced the time between receiving a document and having a structured set of questions ready for legal.
As a result, my time allocation shifted meaningfully. The drafting cycle for board communications has compressed significantly. More importantly, the quality of challenge improved because I used AI to argue against my own interpretation before presenting it. This surfaced gaps I previously would have only caught in the room.
A Real-World Monthly Reporting Workflow
I will describe the monthly board and investor reporting cycle in detail. It has two connected phases: the numbers phase and the narrative and scenario phase. AI is involved in both, playing a different role in each.
The cycle starts when the Financial Controller completes month-end close. Then, I have the actuals for revenue, cost, burn, and the core SaaS metrics: ARR, NRR, burn multiple, and CAC payback. First, I manually review the numbers myself before AI touches anything. That sequence matters. I need my own read of what happened before I introduce any external interpretation, because if I go to AI first, I risk anchoring on its framing rather than developing my own.
After my initial review, I use Claude to interpret variances. I provide actuals against budget, prior period comparatives, and key movements. I then use AI to pressure-test my interpretation of what drove each material variance. At this stage, I'm not asking what happened — I already know that. Instead, I'm asking if my explanation for why it happened is the most accurate and complete version. AI surfaces alternative explanations I may have overlooked and flags logical gaps in the narrative before it goes to the board.
That phase's output directly feeds into the scenario modeling phase. Based on the month's actuals and variance drivers, I update the live financial model. Then, I use AI to stress-test the forward assumptions. This is where I develop scenarios. I run growth, base, and conservative cases. I use AI to challenge the assumptions underlying each one, particularly on revenue trajectory, burn, and runway. Here, I ask if the board will find a hole in any of these scenarios that I haven't already addressed.
The final phase involves drafting. I use AI to produce the first draft of the board narrative, management commentary, and investor update. I then review and rewrite that draft multiple times before finalizing anything. The AI draft serves as a starting point and structural scaffold, not a finished product. My rewrite incorporates the voice, judgment call about what to emphasize and contextualize, and an understanding of what each investor needs to hear at that moment.
Why CFOs Must Be Deliberate About Where AI Sits Within a Workflow
I am deliberate about where AI sits in the workflow. It sits before my judgment, not instead of it. I review and own every output.
I am deliberate about where AI sits in the workflow. It sits before my judgment, not instead of it. I review and own every output. Maintaining that boundary has been important, especially in board and investor reporting, where precision and accountability are non-negotiable.
But I don't have a fixed list of AI tasks versus human tasks. It's just about where the cost of being wrong is recoverable and where it isn't.
For AI-informed tasks, the highest leverage areas for me are variance analysis interpretation, scenario modeling, and financial communications. Variance analysis provides a good example. The calculation itself was never the problem. The problem was always the speed and quality of the narrative around it: understanding what the number actually says, what drove it, and the right management response. AI has meaningfully improved the speed and depth of that interpretive layer.
Scenario modeling and stress testing are other areas where AI earns its place. I use it to pressure-test assumptions, generate alternative cases I might not instinctively reach for, and challenge the model's logic before it goes to the CEO or board. It functions as a structured devil's advocate. FP&A support and first-pass contract review follow similar logic. High-volume, repeatable analytical work where the output feeds into a human decision rather than replacing it.
Areas that remain explicitly human include capital allocation, fundraising judgment, board and investor communication at the final mile, and any decision with a relationship or reputational dimension. Let me explain why.
Capital allocation at a Series B company is not a math problem. It is a judgment call about which bets to make with constrained resources, made under uncertainty, and with incomplete information. The CFO owns that judgment and is accountable for it. AI can model scenarios, but it cannot carry the accountability.
Fundraising is the clearest example of a fully human domain. Investor relationships, term sheet negotiation, reading the room on a deal, and knowing when to push and when to hold are not things I would delegate to or even heavily inform with AI. The stakes are too high, and the variables are too human.
Board communication at the final mile stays human for a different reason. I have used AI to draft and stress-test the logic of my presentation. But what I deliver to Lightspeed, MassMutual Ventures, and Endiya Partners is my read of the business and my framing of the risks. That means my credibility is on the line. AI sits upstream of that moment, not inside it.
How AI's Limitations Affect Finance Teams
I've already addressed many of the benefits of AI in finance. On the negative side, here are three honest observations.
AI is confidently wrong in ways that are harder to catch than obvious errors. A calculation error in a spreadsheet is visible. A plausible but subtly flawed narrative interpretation of a variance is not. Early on, I caught two instances where AI-generated commentary mischaracterized a cost variance driver in a way that would have been embarrassing in a board context. That recalibrated how carefully I review AI outputs before they move forward.
The second limitation is context. AI does not carry institutional memory. Every time I use it, I reconstruct context from scratch. For a finance function where decision history, investor relationships, and the strategic narrative are deeply contextual, this is a genuine constraint. It means AI works best on discrete, well-scoped tasks rather than on anything requiring a deep understanding of how we got here.
The third is adoption within the team. Getting a finance team to use AI well, rather than avoiding or over-relying on it, has required active management. The instinct to paste a number into AI and accept the output without interrogating it is a real risk, particularly with junior team members. Building the right habits around AI use has been as important as the tools themselves.
Three Places Where AI Should Not Be Used in Finance
The area where AI has most clearly underdelivered for me is anything requiring judgment about people and relationships.
Early on, I attempted to use AI to structure and frame difficult conversations: compensation decisions, performance conversations with senior team members, and communications to investors about missed targets. The outputs were technically coherent but completely wrong in practice. They were calibrated for a generic professional context, not for the specific relationship history, the individual's personality, or the unspoken dynamics that govern how those conversations land. I stopped using it for that category entirely.
Another area is forward-looking risk assessment in a startup context. AI excels at identifying risks with historical precedent and pattern recognition. It poorly identifies the specific risks that matter in a high-growth, capital-constrained, rapidly pivoting business, where threats often haven't occurred in that specific configuration before. When we navigated the enterprise pivot at Scrut and simultaneously managed a revenue reforecast, the AI-generated risk frameworks felt generic. The real risks in that moment were specific to our investor relationships, our sales team's capacity to execute a segment change, and our burn position relative to the fundraising timeline. AI could not usefully surface any of that.
And one final area is forecasting accuracy in volatile periods. People implicitly assume AI will improve forecast accuracy in FP&A. In a stable, high-data environment, that may be true. In a Series B SaaS company undergoing a strategic pivot with a sales team in transition and a rapidly evolving product, historical patterns are not useful. The inputs are too noisy, and the business changes faster than any model can usefully learn. AI gave me more scenarios, but not better ones, during that period.
How Finance Teams Are Changing with AI
Some people define their value through output volume, and AI reducing the need for that output has required actively managing how they think about their own contributions.
It has shifted what I hire for without yet shifting how many I hire.
The technical bar for joining my finance team has not dropped. But the nature of the technical requirement has changed. I am less focused on whether someone can build a model or produce a variance analysis and more focused on whether they can think critically about outputs that are not their own.
The specific capability I now explicitly screen for is productive skepticism toward AI: the instinct to interrogate rather than accept. A team member who treats a coherent AI output as correct is now a liability, unlike two years ago, because the volume and plausibility of AI outputs mean uncritical acceptance creates risk at scale.
The second shift is that the controller and FP&A lead have moved upstream. This involves less time on production and more time on review, interpretation, and challenge. That transition has not been seamless for everyone. Some people define their value through output volume, and AI reducing the need for that output has required actively managing how they think about their own contributions.
The team shape has not changed yet. The talent requirements inside that shape have changed significantly.
Why Claude Is a CFO's Most Important AI tool
Claude is my go-to tool. Not because it is the most specialized tool in my stack, but because it is the highest leverage one. Every other finance tool I use serves a specific workflow. Claude sits across all of them.
I use it specifically for variance interpretation and scenario stress testing. Those two activities sit at the center of every consequential finance conversation I have: with the CEO, with the board, with investors. Being able to challenge my interpretation of the numbers and pressure-test my model’s assumptions before those conversations has improved how I participate in them.
It wins over every other tool because it improves my thinking, rather than just speeding up my output. Every other tool I use optimizes for production. This one optimizes for judgment. In a CFO role, that distinction matters more than any efficiency metric.
Does AI Make CFOs Replaceable?
The question I wish you had asked is whether AI is making CFOs more replaceable or less.
The common answer is less replaceable. AI handles the analytical work, and the CFO moves up the value chain, focuses on judgment and strategy, and becomes more indispensable.
I think that is partially wrong.
CFOs who will become less replaceable use AI to develop sharper judgment, stronger board relationships, and deeper strategic thinking, and invest in the human capabilities AI cannot replicate.
CFOs who will become more replaceable use AI to produce better outputs without developing better thinking, mistake faster analysis for deeper insight, and allow the tool to substitute for the intellectual rigor the role has always required.
The uncomfortable truth is that AI is not sorting CFOs into relevant and irrelevant. It is accelerating the separation between CFOs who were genuinely thinking and CFOs who were primarily producing. That distinction existed before AI. It was just harder to see.
I think about that regularly. It is the question that keeps me honest about how I use these tools and why.
How CFOs Should Navigate AI-Driven Changes
My advice starts uncomfortably…The CFO who rushes to deploy AI into high-stakes workflows without the governance infrastructure to catch errors, and the first serious mistake in a board or investor context will set the function back significantly.
My advice starts uncomfortably, so I will lead with that.
Most CFOs will get this wrong by moving too slowly and then overcorrecting. I already see a pattern of long, cautious observation followed by rushed, poorly governed adoption when competitive pressure makes inaction untenable. Both ends of this pattern are expensive. The CFO who waits too long cedes ground on productivity, talent, and strategic relevance. The CFO who rushes to deploy AI into high-stakes workflows without the governance infrastructure to catch errors, and the first serious mistake in a board or investor context will set the function back significantly.
- Get into the work personally and immediately. Do not use a task force, delegate exploration to a junior team member, or read about it. To develop genuine judgment about where AI belongs in a finance function, use it yourself on real work and experience both its leverage and its failure modes firsthand. A CFO who has not personally used AI on a board update, a variance analysis, or a scenario model has no basis for making good decisions about where to deploy it at scale.
- Govern the boundary explicitly before you need to. Define in advance which decisions AI can inform and which decisions remain exclusively human. Do not leave that boundary implicit. In a finance function, a blurred boundary costs you: board communications carry subtle errors, risk assessments miss context-specific threats, and capital allocation decisions optimize for historical patterns rather than forward-looking judgment. Write the boundary down, communicate it to your team, and review it periodically as the tools evolve.
- Treat AI adoption as a team capability question, not a tools question. Access to AI tools will not be the limiting factor in most finance functions; the team's judgment and ability to use them well will be. Junior team members risk over-reliance, accepting AI outputs without interrogating them. This creates quality risk in the very workflows where you seek efficiency. Building the right habits—the instinct to challenge AI output rather than validate it—is a management problem requiring active attention.
- This advice is specific to CFOs in high-growth or VC-backed companies. Your investors are watching how you think about AI, not just how your company uses it. A CFO who can speak fluently and specifically about AI integration in the finance function, including where it has not worked, signals operational sophistication that matters in board relationships and the next fundraise. Vague enthusiasm about AI is as damaging as ignorance of it.
- And most CFOs will resist this advice. AI's arrival does not reduce the premium on human judgment in finance leadership; it increases it. As AI absorbs more analytical and drafting work, what remains exclusively human becomes more visible and consequential. Relationship intelligence, accountability, the ability to read the room in a board meeting, the judgment to know when the model is wrong, even when it looks right. AI is not automating those capabilities; it is amplifying their importance precisely because everything around them is being automated. CFOs who thrive in this moment will invest in those human capabilities while integrating AI into their workflows.
Follow Along
You can follow Nithin Shetty's work on LinkedIn.
More expert interviews to come on The CFO Club!
