Click for Takeaways: AI Adoption in Finance
- The AI literacy gap: most finance professionals were trained in the pre-AI world, creating two camps that block adoption: “excited but clueless” and “cautious but clueless,” and industry research confirms that education and skills gaps are the top barriers to AI adoption in finance
- The 80/20 flip: AI moves finance teams from spending 80% of time compiling and cleaning data to spending most time on strategy, insight, and influencing decisions
- Vendor reporting transformation: a 40-hour monthly process was reduced to several hours using a named entity recognition model; one analyst previously spent 25% of working time cleaning data across 10 countries where 80% of transactions were missing vendor names
- Foundation requirements: three non-negotiables before any AI implementation: clean, consistent, and connected data with standardized naming conventions and metric logic; integrated systems across HR, finance, sales, and operations; standardized processes
- The ambition-reality gap is real: research shows that while most CFOs expect significant AI ROI within two years, only a fraction report meaningful value today, largely because fewer than one in ten fully trust their enterprise data
- Change management sequence: start with enthusiasts, bring them in early, make them accountable; identify quick wins deliverable in days or weeks with clear before-and-after comparisons; bring skeptics on board by showing concrete results rather than making promises
The question came from the audience during a panel discussion at a finance conference. Someone asked about model training, and the person asking literally used “the bolts.”
That’s when it hit Mariya Guttoh, Director of FP&A and Treasury at PayJoy: most finance professionals have no idea how AI actually works.
“There’s not a lot of understanding what AI is, what a model is, what generative AI is,” Guttoh explains. “And it’s understandable because most financial professionals are trained in the pre-AI world. We trained on Excel, not Python. We trained to explain variance analysis, not trained models.”
This gap isn’t just about technical skills. It’s creating two camps in finance that are blocking AI adoption in finance: teams that are “excited but clueless” about what AI can do, and teams that are “cautious, well also clueless, but also they are convinced that the Skynet is coming for their job.”
In a recent episode of Datarails’ FP&A Today podcast, Guttoh revealed how she automated a 40-hour monthly process down to minutes, why foundation matters more than fancy algorithms, and what it takes to convert skeptics into evangelists when most finance professionals don’t understand the technology they’re being asked to adopt.
Her message is clear: AI adoption in finance isn’t failing because of bad technology. It’s failing because of bad education, unrealistic expectations, and change management that ignores how humans actually learn to trust new tools.
From $300 and Three Jobs to AI Implementation
Guttoh’s path to leading AI adoption in finance wasn’t linear.
She started in cosmetics marketing in Ukraine, managing brand launches and consumer research. At 23, she moved to the United States with $300 and one suitcase. Her Ukrainian master’s degree in accounting and economics didn’t matter. Her work experience didn’t count.
“The US job market politely declined my experience and my education from Ukraine,” Guttoh recalls.
So she did what she had to do: worked three jobs with no days off for three consecutive years while earning her MBA in finance and international business. She couldn’t get student loans because she had no credit history. Every class was paid for in cash.
“Throughout my MBA, I realized how analytical I am, and I fell in love with finance because it’s the control tower of the business where every decision and every number and every trend connect to the bigger picture,” she says.
But the marketing background shaped how she approaches finance. “What I brought from marketing is two main things. First, data storytelling matters, one of the defining shifts in modern FP&A trends. Numbers by themselves such as data. It’s only when you make them relatable to the audience, you can influence people. That leads me to the second point. You need to know your audience.”
This combination of analytical rigor and audience awareness is why Guttoh now focuses not just on building AI models, but on teaching finance teams how to trust them.
The AI Literacy Gap That’s Blocking Everything
When finance professionals don’t understand AI, they make one of two mistakes: they either overestimate what it can do or underestimate its value entirely.
“Overestimating what AI can do and not understanding how it works can lead to unrealistic expectations, followed by disappointment, followed by distrust,” Guttoh explains. “And just stating what AI can do actually means that teams are missing out and not investing in learning or experimenting with AI so they can miss the opportunity conflict plan.”
Both responses are problematic for AI adoption in finance.
The “excited but clueless” camp thinks AI will forecast everything perfectly with the push of a button. They don’t understand training data, model drift, or validation requirements. When AI doesn’t deliver magic, they abandon it entirely.
The “cautious but clueless” camp sees AI as an existential threat. They’re convinced automation will eliminate their jobs. Because they don’t understand what AI actually does well (and what it doesn’t), they resist any adoption efforts.
“The breach to solve this problem is education,” Guttoh says. “Because if people don’t understand it, they will not trust it. And if they do not trust it, they will not use it.”
This isn’t about turning finance professionals into data scientists. It’s about building enough literacy to ask the right questions, guide outputs, and trust results.
“You don’t need to be data scientists to thrive in AI powered finance, but you do need to understand enough to ask the right questions, guide the outputs, and build trust in the results,” Guttoh emphasizes.
Her approach: skip the technical jargon. Talk about AI in plain finance language through real examples. Show how it connects to daily work. “Once people see how it connects to their everyday work, how it simplifies it, the fear kind of starts to fade and they get, they understand it a little bit better and that’s when the real adoption can possibly begin.”
The Junior Analyst Analogy That Actually Works
When explaining AI to finance teams, Guttoh uses an analogy that resonates: AI is a junior analyst who never sleeps and doesn’t require coffee.
“Imagine you’re hiring a junior analyst, just the one that never sleeps and doesn’t require coffee,” she tells skeptical finance professionals. “The same as a junior analyst, you train an AI model, you provide it with historical data and with the correct answers.”
Just like a junior analyst learns from experience and feedback, machine learning works the same way. You check results and provide corrections. Over time, the junior analyst gets better at spotting errors just by looking at numbers, before even checking formulas. AI does the same thing.
This framing shifts the conversation from threatening replacement to helpful augmentation. Finance teams already know how to work with junior analysts. They understand training, feedback loops, and gradual improvement. Positioning AI the same way makes adoption less intimidating.
“I draw really easy parallels to make it more engaging for finance folks and make it more understandable how the process works in the background,” Guttoh says.
From 40 Hours to Minutes: The Vendor Reporting Win
Theory only goes so far. Real AI adoption in finance accelerates when people see actual results.
At PayJoy, Guttoh’s team had a massive monthly headache: producing detailed expense reports by vendors for each function across 10 countries, a common challenge in global cash management. It sounds straightforward until you understand the reality.
Only three countries were fully integrated into PayJoy’s ERP. The other seven closed books in their own systems and uploaded data later. This meant 80% of transactions, hundreds of thousands monthly, were missing vendor names in the dedicated field.
Vendor information existed somewhere in description fields or memos, but these were freeform text across multiple languages with inconsistent spelling and formatting. Producing meaningful vendor reports required manual data cleaning.
One analyst spent 40 hours every month on this process. A quarter of his working time went to cleaning data just to provide basic information to department heads.
“We operate across 10 countries and only three of them are fully integrated within our ERP,” Guttoh explains. “The rest are closing the books in their owner’s piece and then not uploading their data. So basically 80% of data, 80% of transactions, which is hundreds of thousands on a monthly basis, were missing the vendor name and its dedicated field.”
Guttoh scratched her head and realized she knew enough from her University of Texas postgraduate certificate in AI and machine learning to solve this.
She built a process that cleans, translates, and normalizes the data. Then she used a pre-trained named entity recognition model to pull potential vendor names from the cleaned data and normalize them. Finally, the model clusters transactions without identifiable vendor names into groups based on descriptions.
“So now the entire process takes several hours, including populating publishing reports versus 40 hours a month,” Guttoh says.
The analyst who previously spent a quarter of his time on this work? Now he’s working with the accounting team to fix underlying data that typically requires account reconciliation software.
“This is time we’re actually using to work with the accounting team to fix their underlying data and to fix their process, which in my point of view is going to add far more value in the long run than spending it on cleaning the data,” she notes.
The team called it “vendor reporting week.” Now it’s vendor reporting hours.
This is AI adoption in finance at its most practical: eliminating soul-crushing repetitive work so analysts can focus on problems that require judgment and business insight.
Foundation First: Why AI Fails Without Clean Data
Guttoh is blunt about what must exist before implementing AI: your foundation.
“Before you start thinking about AI, implementing AI and what tools to use, you need to make sure you have your foundation in place,” she says. “Because I would say that AI is a rule, but your foundation is three main things. Your data, your process, and your systems.”
Data must be clean, consistent, and connected. Same naming conventions. Same timestamps. Same metric logic. You cannot have different calculations for the same metric across teams. Create a data dictionary and lock it.
“Consistency is the key,” Guttoh emphasizes. “Without it, the model won’t know what to trust and neither will you.”
Systems must be integrated. If HR, finance, sales, and operations don’t talk to each other, AI models only see half the picture. Try producing accurate forecasts without knowing the sales pipeline or inventory levels. You’re just guessing.
Processes must be standardized. If accrual methodology changes month over month, if CapEx versus OpEx booking varies by country, AI will produce “very interesting results that will have no correlation with reality.”
“AI is learning from the pattern,” Guttoh explains. “So if there is no pattern, if your accrue method is changing months over months, AI is gonna just be confused.”
Get these three things right before even thinking about AI. Otherwise, you’re building a roof with no walls.
“Don’t get distracted by the AI high, build the foundation first and the roof will hold,” she says.
Local Currency Forecasting: The Second Win
After cleaning vendor data, Guttoh tackled another pain point: forecasting expenses in USD across 10 countries with different currencies.
The traditional approach had finance functions allocating expenses in USD at consolidated levels. Easy for them to project travel and other expenses without splitting by country. But completely divorced from reality.
“In reality it works in Excel, but in reality it really didn’t work because if Mexican Paso appreciates our USD expenses in pesos, USD is actually higher, but so is our revenue,” Guttoh explains.
The other problem was granularity. Teams were projecting at consolidated GL level without detail.
Once vendor data was clean, Guttoh had the foundation to build an AI model that forecasts fixed expenses at the vendor level in local currency. Even when vendors work across multiple countries, the model forecasts each country specifically in local currency, incorporating past seasonality, trends, and local inflation.
Then it translates everything to USD and provides summaries by GL and function.
“So now the function still sees their forecast the same way they used to, nothing changed for them, but we as FP&A have a full breakdown behind every number,” Guttoh says. “And instead of just reporting variances, we actually can explain what’s driving the variance to the budget and segregate the effects impact from a specific vendor or project.”
This demonstrates mature AI adoption in finance: solving real business problems while maintaining familiar interfaces for end users who don’t need to understand the underlying complexity.
The Change Management Strategy That Actually Works
Technical capability doesn’t drive AI adoption in finance. Change management does.
Guttoh’s approach is structured and repeatable. “First things first, I identify the real pain, something that makes the team groan every month, every week,” she says. “And once you know the pain, you truly need to understand the cause of the pain, otherwise you will be treating the symptoms and not the cause.”
Next, involve the right people from the beginning. People who understand the problem and can translate it clearly. Otherwise, IT builds solutions that don’t match business needs.
“Tech people can actually build a solution or workflow that does not translate a reality and doesn’t solve our business needs,” Guttoh notes.
Stay agile. Business needs change faster than multi-month development cycles. Build constant check-ins, short cycles, testing by end users. Pivot and adjust until it works.
“In our environment, PayJoy’s business needs change faster than we can say balance sheet honestly,” she says. “And tech cannot just disappear for three months and come back with a solution that we don’t need anymore.”
Finally, share quick wins. Don’t wait for a full rollout. Show progress immediately.
“Even if it falls just a little part of the pain, this visible progress turns skeptics into believers and creates momentum for even bigger change,” Guttoh explains.
Selling AI to Enthusiasts First
Successful AI adoption in finance doesn’t start by convincing skeptics. It starts with enthusiasts.
“I start with the enthusiasts and I bring them in early, I involve them, I engage them, I make them accountable,” Guttoh says. “I teach them so they understand and that’s the people who actually will help you to steer this shape of change.”
From there, look for quick wins: projects deliverable in days or weeks, not months, with results you can show in before-and-after comparisons.
“The trick is to make the benefits so always that no one can argue with them,” she explains.
Once you have a few quick wins, bring skeptics on board. Skip the hype and buzzwords. Go straight to what matters to them.
“Look, this process that used to take your hours, sometimes even days now takes you minutes,” she tells skeptics. “That’s when the enthusiast who already trained also steps in and helps me in delivering this message. And that’s when the light bulb goes on and they see that, oh yeah, that actually helps and that actually can move the business into the right direction.”
Communication matters. Celebrate even the smallest wins. They build momentum and trust.
This approach recognizes that AI adoption in finance is fundamentally a people problem disguised as a technology problem. Solve for the people and the technology follows.
The 80/20 Flip That Changes Everything
Most finance professionals spend 80% of their time compiling, cleaning, and checking data. Only 20% goes to understanding what the numbers actually mean.
AI flips this.
“AI flips this threshold along us to actually spend most of the time and understanding the numbers influencing strategy and turning insights into actionable advice for business to improve working capital and decision-making,” Guttoh says.
This is the promise of AI adoption in finance: not replacing finance professionals, but freeing them from mechanical work that consumes time without adding strategic value.
Building FP&A from scratch, Guttoh would embed AI into data-heavy repetitive tasks from day one. But she’s clear-eyed about limitations.
“The FP&A of my dreams cannot be built in isolation,” she notes. “We’re not a siloed team and an AI-centric team like FP&A requires a strong foundation. And much of that foundation is actually built across organizations by different teams. So it really takes a village to create an AI ready business.”
The human element remains critical. “Not everything that counts can be counted and AI still needs human guidance to focus on what truly matters,” Guttoh says. “So I don’t think AI will replace as soon but knowing and being curious, experimenting with that actually can shift FP&A from being just budget gatekeepers to one of the most strategic future ready functions in our organization.”
The shift moves FP&A from cost center to profit center. “That’s not just a tech update, it’s a true real mind shift in what FP&A is and its role and the role it plays and driving business forward.”
The Upskilling Roadmap
AI adoption in finance requires new skills, both hard and soft.
Hard skills include AI literacy (understanding models, overfitting, drift), data tools (Power Query, SQL, Python), process automation concepts, and governance around sensitive data.
“You need to be data smart. You need to be comfortable with tools like power, acquire SQL, even a little Python, just enough to have meaningful conversations with other teams and IT team specifically,” Guttoh says.
But soft skills are equally important, maybe more so.
“As I mentioned before, storytelling is huge because executives don’t care how pretty your report is. They care what it means for me and the business,” she explains. “And you need to be the one with answers.”
Cross-functional collaboration matters. Finance professionals must speak not just finance language, but product, risk, fraud, and engineering languages. Business acumen, adaptability, critical thinking, and listening skills round out the requirements.
“That’s the mix that will actually make future finance professionals successful,” Guttoh says. “Because hard and soft skills now share the driver’s seat. One shows you the way and the other one gets you there.”
The Bottom Line for Finance Leaders
AI adoption in finance is stalling not because of bad technology, but because of an education gap creating unrealistic expectations and paralyzing fear.
The finance leaders who will drive successful adoption:
- Build AI literacy without requiring data science degrees
- Start with foundation: clean data, integrated systems, standardized processes
- Identify real pain points and understand root causes, not symptoms
- Target quick wins that deliver results in days or weeks, not months
- Convert enthusiasts into evangelists who help sell change internally
- Use analogies (like junior analysts) that connect AI to familiar concepts
- Focus AI on eliminating repetitive work, not replacing strategic judgment
- Celebrate small wins to build momentum and convert skeptics
The ones who demand AI solve every problem without understanding limitations will create disappointment and distrust. The ones who ignore AI entirely because they don’t understand it will watch their teams fall behind competitors who iterate toward better solutions.
The opportunity is clear: AI can flip the 80/20 ratio, freeing finance teams to spend most of their time on strategy and insight instead of data compilation. Getting there requires education, foundation-building, and change management that meets people where they are, whether they’re excited but clueless or cautious and convinced Skynet is coming.
Where Datarails Fits
Datarails is an Excel-native FP&A platform that helps finance teams automate reporting, consolidate data, and build AI-ready foundations, so teams can stop cleaning data and start driving strategy. To learn more about how finance teams are using AI to accelerate planning and analysis, explore Datarails AI.
This article is based on Mariya Guttoh’s appearance on the FP&A Today podcast.
Mariya Guttoh is Director of FP&A and Treasury at PayJoy, a global FinTech company expanding credit access through smartphones to underserved customers across emerging markets in LATAM, Africa, and Southeast Asia. The company has served over 15 million customers and is on track to reach $650 million in revenue by end of 2025. Prior to PayJoy, Mariya spent nearly a decade at Bank of the West in senior FP&A and strategic finance roles within a $90B+ organization. She holds an FP&A certification, a postgraduate certificate in AI and Machine Learning from the University of Texas, and an MBA from the University of South Florida.
