How are finance teams actually using AI and automation in 2025?

The hype around Artificial Intelligence in the world of finance is easy to find. On one side, there are people worried about software agents replacing accountants. On the other, some dream of a supercharged finance department capable of predicting the future with unprecedented accuracy. But what is not so obvious is how companies are actually using AI in their financial operations in 2025 and 2026.

More than three years after ChatGPT catalyzed a new wave of interest and investment in Artificial Intelligence, many companies are still trying to figure out how the latest iterations of this technology can fit into their business models. And the numbers confirm this disconnect between excitement and concrete results.

A recent Gartner report revealed that while nearly 60% of finance teams are already piloting or implementing AI projects, only 7% of CFOs report a strong, significant impact from that investment. In other words, most are spending time and money, but few are seeing real returns.

You can understand the frustration, right? 😅

This is exactly the scenario that Mohit Sharma, ACMA, CGMA, sums up perfectly. Sharma recently launched two AI startups focused on the financial sector after decades in the field, and his assessment is straightforward: right now, AI is going through an identity crisis.

But hold on, because the story does not stop there. While the debate between hype and reality continues, some companies and professionals are already rolling up their sleeves and finding practical, concrete, and replicable ways to use AI to solve real problems in their financial processes. 🙌

In this article, you will learn about three real cases showing how financial automation is moving from concept to reality, with examples ranging from predicting late payments to consolidating fragmented accounting data and automating manual processes at scale. These are stories from people who tested, failed, adjusted, and found a path that works.

Getting paid on time with predictive Artificial Intelligence

Mohit Sharma co-founded Pinaka AI in 2023 to tackle a frustration he heard repeatedly from financial leaders throughout his international career. The problem was always the same, regardless of the country or company size.

About 60% of customer invoices generated in a B2B environment are not paid on time.

It sounds simple, but a late payment can trigger a cascade of consequences. The sales team may need to renegotiate a contract. The company may need to take on debt to cover cash flow. And so the cycle repeats, becoming what Sharma describes as a vicious problem that drains resources from multiple areas of the organization.

Pinaka AI, which employs about a dozen people, developed a software product that does something impossible at scale for a human team: predict which customers will be late on a payment and, more importantly, identify the specific reason why that customer will likely miss the deadline.

The product is currently being piloted by two major manufacturers in India, and the early results are impressive. According to Sharma, the company’s algorithm can make predictions with 96% accuracy.

Hosted on Oracle Cloud, the platform also uses generative AI to recommend actions the user can take and is capable of drafting and sending personalized emails to help collect payment before the due date.

In practice, the system provides recommendations on what to do to resolve the situation today, weeks before the payment deadline. This is possible because the tool integrates data that is normally scattered across different company systems.

Sharma explains that a customer’s payment behavior is spread across different platforms. And that is precisely the job of AI: integrate everything and create a single version of the truth.

Pinaka AI’s predictions are based on data from the client’s CRM and ERP systems, as well as external sources like credit agencies and news outlets. The startup combines four different types of AI to achieve this result: a recommendation engine, decision intelligence, a classification algorithm, and generative AI.

The company is still in the early stages of its journey, having recently completed an accelerator program and working through its pilots with the large manufacturers. But Sharma makes an important point: this is a task that traditional automation could have tried to solve years ago, but the rise of new AI tools made product development much faster and more cost-effective.

When you are solving complex problems, you need intelligent systems, he sums up. 💡

Unifying messy accounting records

Janice Stucke, CPA and CGMA candidate, took on the role of CFO at CREW Network last year, a trade association that connects more than 14,000 women in commercial real estate around the world.

Stucke inherited a finance department that was still issuing paper checks and needed a complete overhaul. It was exactly the kind of challenge she enjoyed, since she had implemented robotic process automation (RPA) and generative AI in her previous roles.

The first step, however, was not to implement any fancy tool. It was to face a data problem that seemed insurmountable.

The organization had accounting data spread across roughly 50 different entities, each with its own chart of accounts. Some were subsidiaries, others were adjacent entities, distributed across multiple countries and, of course, in varying formats. This fragmentation was slowing down payment processing and required constantly updated custom coding just to keep things running — a mountain of technical debt for an organization with only 35 employees.

With so many entities and each one having its own chart of accounts, the organization could generate more than 10,000 general ledger transaction lines per month.

Stucke decided to create a consolidated chart of accounts for all entities, enabling system automation. But the process of migrating the entire activity history to the new structure for historical comparison purposes was, in her words, a massive project.

The time dividend that AI delivered

It would have taken Stucke’s team weeks, even with help from consultants, to transform the data from the old general ledger structure to the new one. The traditional approach would have been to implement spreadsheet macros or set up RPA bots. But Stucke figured both solutions would require excessive customization for the countless variations in data format and definitions.

She had a different idea: use her enterprise ChatGPT account to transform the charts of accounts. She was able to input the data and ask the AI to map everything to the new unified format. Beyond merging data by period, the tool could also resolve variations in how each entity labeled transactions, using generative AI to interpret the many ways the charts might categorize something like event revenue.

It worked. Up to a point.

Stucke also ran into common frustrations with generative AI. After executing the transformation flawlessly on 10 consecutive charts of accounts, the tool might implement a new, unwanted method on the next one. In practice, she had a hard time getting the software to cooperate 100% of the time.

There was also the issue of hallucinations and errors, something common in generative AI software. Stucke tried to get ChatGPT to include formulas that would show and verify its work, but it simply did not work correctly.

The solution? Implement her own verification formulas, exactly as she would with human or RPA outputs, to check the new unified ledger against the old data.

My internal controls process did not change, she explained.

Even with the frustrations, the AI approach delivered faster and still reliable results in a matter of days, allowing her to update the systems and keep moving forward with automating the finance function. And all of this happened with a relatively generic, accessible, and affordable product.

Stucke believes this new wave of AI products is empowering small and mid-sized businesses to experiment. Many smaller organizations feel they do not have the talent needed to implement these systems. However, it is possible to take a generic product and automate with far less effort than before, as long as the right vision is behind the project.

Fixing manual processes at continental scale

The most valuable advances in AI and automation simply do not happen without a unified and accessible data source. This has been a central focus of the work of Lawrence Amadi, ACMA, CGMA, partner and leader of the technology risk practice at KPMG Africa.

Among Amadi’s clients are some of the largest telecom operators on the African continent, including one with more than 85 million subscribers. Amadi had been working with them to transform their SIM (Subscriber Identity Module) systems, which authenticate and manage user identities and their devices on the network.

The company relied on manual processes to manage this massive volume of data. Roughly once a week, the team would download the data and manually check for incomplete or anomalous records.

These manual checks increase the risk that data is not fully exported or that the team suffers from what Amadi describes as audit fatigue. It becomes very stressful and demoralizing, he observed.

KPMG Africa’s goal was to automate the export, analysis, and flagging of exceptions from that same data. The project took seven months and involved people from different areas of the organization.

Amadi emphasizes that the project required professionals with diverse skill sets. They needed people who understood the product, knew the business rules, and understood how the data was produced. Strong data analysts were also essential — people capable of making sense of the data, unpacking it, analyzing it, and reorganizing it.

The project was built with Automation Anywhere, which describes itself as an agentic process automation system that combines the power of AI, automation, and RPA.

The result was a complete shift in workflow. Instead of periodic manual downloads, there is now an automatic export running in the system, an automatic analysis being executed, and exceptions being raised automatically.

According to Amadi, the new system delivered reduced errors, greater efficiency, and better quality in the reports presented to the board and the audit and risk committee.

And now that the company has organized and centralized an important data source, it is ready to expand the application of automation and AI to other areas of the business. 🚀

What separates AI projects that work from those that stay on paper

Looking at these three cases, it is easy to spot a pattern that differentiates successful initiatives from the majority of projects that stay stuck in the pilot phase. In every example, there was a specific, well-defined problem being addressed, with clear success metrics established before any implementation.

Pinaka AI did not try to reinvent the entire financial cycle — it focused on predicting B2B late payments. Janice Stucke did not try to automate the entire CREW Network at once — she started by unifying the charts of accounts. Lawrence Amadi and KPMG Africa did not propose transforming the telecom’s entire operation — they tackled SIM data management first.

This incremental approach is far more sustainable than trying to make everything work at the same time, and it is what explains why these projects achieved real impact while so many others are still stuck in eternal pilot mode.

Another common element is data quality as a non-negotiable starting point. In all three cases, there was a significant investment in cleaning, structuring, and standardizing data before putting any AI model to work. This foundational work is tedious, time-consuming, and unglamorous, but it is what determines whether the model will work or not. Companies that skip this step end up with models that produce inconsistent results, which erodes the team’s trust in the technology.

Finally, all of these cases kept the human at the center of the process, especially for decisions involving ambiguity or high impact. AI was used to amplify the team’s capacity, not to replace it. And this combination — machine doing what it does best and human doing what they do best — is what is producing real results.

Three AI implementation tips for finance professionals

The three professionals interviewed shared valuable lessons learned in practice. If you are thinking about launching your own AI project in the finance space, these points are worth paying attention to.

Understand the real costs

Mohit Sharma learned many lessons from his startup. One of the most important is how much money AI costs to operate.

A poorly configured product can burn through tokens — which function as a practical measure of computational power and cost — at an alarming rate. Running an AI product also introduces new risks that need to be accounted for: how much will it cost if a generative AI product makes a mistake or causes a controversy, perhaps using inappropriate language with a customer?

There are layers of cost, direct and indirect. The moment you lose visibility, you lose money.

These costs and risks need to be weighed against a realistic forecast of the revenue and efficiencies to be gained from the project. It does not matter whether it is OpenAI or Google technology. The reasoning needs to be financial: what is the return for my business? When does the solution hit break-even?

Build organizational buy-in

Especially in large organizations, an AI or automation project may require combining skills and data from different parts of the company. This simply does not happen unless people understand and accept the mission, according to Lawrence Amadi.

You need buy-in from everyone. The why is crucial.

That why can range from better control coverage to greater business visibility. Regardless of what it is, there will be no progress if the purpose and means of the project are not clear to everyone involved.

Let others learn along the way

Janice Stucke used generative AI to complete a daunting data transformation in just a few days, by herself.

I was able to accomplish in four to five days what would have taken an entire team of consultants and my staff two to three weeks.

But that efficiency came with a cost that is not always obvious: my entire team was not taken on that journey, Stucke acknowledged. In other words, the team did not get the opportunity to practice using AI or to see exactly how and why the tool was so effective.

AI proficiency will continue to be an essential skill for finance teams, especially as automation absorbs more work. Involving everyone now can help future-proof careers and accelerate the digital transformation of the finance function as a whole. ✅

The picture taking shape for the future of finance with AI

The three cases presented in this article paint a realistic and at the same time hopeful picture of the current state of AI in finance. The technology works, yes, but not magically. The most meaningful results come from teams that did the hard work of preparing their data, defining clear problems, and maintaining realistic expectations about what AI can and cannot deliver on its own.

The Gartner figure about the 7% of CFOs reporting strong impact should not be read as a sign that the technology has failed. In reality, it is a reflection of where most companies are on the maturity curve. As more organizations follow the lead of the cases described here — focusing on specific problems, investing in data quality, and keeping humans at the center of the process — that percentage is likely to grow significantly.

AI in finance does not need to be an instant revolution. The best results are coming from those who treat this journey as a continuous evolution — project by project, problem by problem, with each win paving the way for the next. And that is something any finance team, of any size, can start doing right now. 🎯