How Finance Teams Are Really Using AI and Automation in 2026
The conversation about Artificial Intelligence in finance is everywhere, but not always grounded in reality. There is a lot of talk about autonomous agents replacing accountants, perfect cash flow forecasts, and a future where everything is automated. In practice, though, the 2026 landscape is a lot more mixed: while most companies are already testing some sort of AI application, few are seeing a strong impact on their results.
A recent Gartner report shows that nearly 60% of finance teams are piloting or have already implemented AI projects. Even so, only 7% of CFOs say they feel a truly relevant effect from this technology on business performance. In other words: lots of proof of concept, lots of expectations, but still a long way to go before maturity.
Mohit Sharma, ACMA, CGMA, a finance executive who founded two AI startups focused on the sector, sums up this moment as a kind of identity crisis for the technology: its potential is huge, but the right way to fit AI into financial processes is still being figured out. Instead of magic solutions, what is emerging are concrete use cases in three main areas:
- Payment prediction and reduction of late payments;
- Unification and transformation of accounting data scattered across multiple entities and systems;
- Automation of manual processes at scale, built on reliable data.
Three leaders in finance and technology show, in practice, how they are using AI and automation to tackle very specific, everyday problems: getting paid on time, organizing the chaos of charts of accounts, and ending exhaustive manual checks across huge data sets.
Payment Prediction: AI on the Front Line To Reduce Late Payments
Sharma co-founded Pinaka AI in 2023 to solve a pain point he heard repeatedly during his travels and projects around the world: chronic delays in B2B payments. According to him, on average, around 60% of invoices issued in B2B environments are not paid on time. These delays are more than an annoyance: they can trigger a chain of issues such as contract renegotiations, loss of trust, the need to take on short-term debt, and constant stress on cash flow.
Instead of treating delinquency as something inevitable, Pinaka AI created a payment prediction product that flags, in advance, which customers are highly likely to be late and, more importantly, why that delay might happen. According to Sharma, the algorithm reaches up to 96% accuracy in its current tests, run with two large manufacturers in India.
To reach this level of precision, the platform integrates data from several sources:
- CRM (customer relationship history, negotiations, complaints);
- ERP (invoices issued, terms, payment methods, credit limits);
- External sources, such as credit bureaus and news that might affect a customer’s financial health.
Sharma sums up the role of AI in this context in a straightforward way: customer payment behavior is scattered across multiple systems; AI’s job is to integrate everything and create a single, reliable view. To do this, the tool combines four types of intelligence:
- Recommendation engine to prioritize customers and actions;
- Decision intelligence to suggest the most appropriate next step;
- Classification algorithms to categorize risk and behavior;
- Generative AI to create personalized messages and communications.
Hosted on Oracle Cloud, the solution not only predicts who may be late, it also recommends practical actions weeks before the problem shows up. The platform, for example, drafts and sends personalized emails for collections or adjustments to commercial terms, adapting tone and content to each customer profile.
In the past, something like this would depend on rigid automations and a much larger amount of manual development. With today’s AI tools, Sharma points out, it has become much faster and cheaper to build this intelligence, mainly because the models can learn from data without every single rule needing to be hard-coded.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
The focus, however, remains very pragmatic: it is not about replacing collections analysts, but about providing early visibility and actionable recommendations at a scale that would be impossible with human work alone.
Unifying Accounting Data: From Paperwork to AI Models
While Pinaka AI deals with cash-in forecasts, another big challenge is organizing the foundation that everything else depends on: the chart of accounts and the general ledger of dozens of entities around the world. That was exactly the scenario that Janice Stucke, CPA and CGMA candidate, found when she took over as CFO of CREW Network, an association that connects more than 14,000 women in commercial real estate in several countries.
Stucke inherited a finance department that still issued paper checks and relied on manual processes for practically everything. To make things worse, accounting data was spread across around 50 different charts of accounts, related to subsidiaries and affiliated entities in different countries. Each had its own structure, naming conventions, and formats.
The result was operational chaos: delayed payment processing, a high volume of manual adjustments, and a stack of custom code to keep integrations and reports running. For an organization with only 35 employees, the amount of technical debt was far too heavy.
Stucke decided to attack the root of the problem: create a consolidated chart of accounts for all entities. The goal was easy to state and hard to execute: standardize everything to make large-scale automation possible. That meant reclassifying and mapping more than 10,000 lines of accounting entries per month, rewriting history so that comparisons across periods and entities would be meaningful.
Generative AI as a Partner in Data Transformation
Following the traditional route, this transformation would take weeks, even with help from consultants and heavy use of spreadsheet macros or RPA. Instead, Stucke decided to try something different: use her corporate ChatGPT account to map and convert the old charts of accounts into the new unified structure.
She fed the model with the different charts and asked the AI to suggest mappings to the new standard, taking into account descriptions, contexts, and historical usage. On top of grouping accounts by period, the AI helped normalize naming variations — for example, several different ways of recording what was essentially the same type of event revenue.
The result was positive but not perfect. In multiple rounds, the AI performed the transformation consistently and, suddenly, on a new data set, started applying a different logic without a clear justification. On top of that, when Stucke tried to get ChatGPT to create reliable validation formulas inside the spreadsheets, the results were inconsistent.
The solution was to combine the best of both worlds: use AI to speed up mapping and reclassification while keeping tight internal controls with formulas and checks designed by her, just as she would when reviewing the work of a human or an RPA bot. In other words, the control process did not change; what changed was how quickly she could get to the first draft.
Even with its limitations, the approach cut the project timeline from weeks to just a few days. In four or five days, Stucke achieved progress that, using the traditional model, would have required two to three weeks from a full team plus external consultants. That created a clean enough data foundation for new automations to be built on top of standardized information.
She also highlights an important point: generic AI tools that are accessible and relatively low cost are starting to empower small and midsize businesses that previously had no capacity to invest in large platforms or massive IT projects. The challenge now is making sure teams keep up with this shift, learning to use AI in a practical and critical way, without giving up strong controls and validations.
Automation at Scale: From Manual Downloads to Continuous Monitoring
While some teams struggle with charts of accounts and overdue receivables, others operate at a volume level that makes manual checks impossible. That is the situation faced by clients of Lawrence Amadi, ACMA, CGMA, partner and leader of the technology risk practice at KPMG Africa, who works with some of the continent’s largest telecom companies.
One of these companies, with more than 85 million subscribers, needed to constantly control and clean a massive volume of SIM card data – the chips that authenticate users and devices on the network. All this control depended on a critical, fully manual weekly process: employees downloaded large data sets, ran filters, and tried to identify incomplete, inconsistent, or suspicious records.
This kind of routine brings two obvious risks:
- Incomplete or corrupted data at the time of export;
- Audit fatigue, when the team stops spotting problems because it is overloaded.
Besides being exhausting, the process was fragile. Skipping a single step or applying a filter incorrectly was enough for a large portion of errors to slip through. KPMG’s proposal was to redesign everything using an intelligent automation system built on the Automation Anywhere platform, which combines AI, automation, and RPA in a single environment.
The project took about seven months and involved people from several areas, not just IT. Amadi points out that the success of automation at this scale depends on three profiles working together:
- People who deeply understand the product and the associated business rules;
- Data specialists who can unpack, analyze, clean, and reconstruct large volumes of information;
- Technology and risk professionals who ensure process robustness and compliance.
In the new setup, what used to be a manual weekly download turned into a continuous flow:
- Automatic data exports at defined intervals;
- Automated analysis based on rules and trained models;
- Automatic generation of exceptions and alerts within the system itself.
The impact was straightforward: fewer errors, more efficiency, and far more reliable reports for the board and the risk and audit committee. And perhaps most importantly, the company now has a more organized and auditable data foundation, ready to support other layers of automation and AI in different areas, such as billing, fraud prevention, or usage behavior analysis.
Three Lessons To Implement AI in Finance Without Getting Lost in the Hype
From the experiences of Sharma, Stucke, and Amadi, you can pull out some very practical pointers for running AI in finance projects without falling into common traps.
Understand the Real Cost of AI
For Sharma, one of the biggest shocks for newcomers is discovering the operational cost of AI. Generative models and advanced algorithms can consume a huge amount of tokens and computing resources if they are poorly designed. Beyond that, you have to consider less obvious risks, like the financial impact of a serious error made by a generative model when interacting with customers.
He stresses that everything needs to be viewed through the finance lens: what is the expected return, when is the breakeven point, and at what point does the extra cost stop paying for itself. It does not matter whether the technology comes from OpenAI, Google, or any other provider – without financial clarity, the project can easily turn into just another recurring expense that is hard to justify.
Building Buy-In Matters as Much as Choosing the Technology
Amadi emphasizes that, in large projects, nothing moves forward if the right people are not committed to the initiative. Different departments need to open up their data, review rules, and adjust processes. Without understanding why the transformation is happening – whether it is for more control, better visibility, or risk reduction – the likely outcome is silent resistance and political roadblocks.
He is blunt: without a clear, shared reason, there is no real progress. In this context, internal communication and alignment among finance, IT, and the business are just as important as the AI engine running in the background.
AI Needs To Be a Team Skill, Not Just an Individual One
In Stucke’s case, using ChatGPT to transform the chart of accounts came with a side effect: she managed to do in a few days what would have taken weeks for an entire team. But by concentrating this learning in herself, the team missed the chance to get familiar with the tool, test its limits, and understand how AI can support their daily work.
As automation takes over more transactional tasks, mastering AI tools will become an important part of the finance professional profile. Bringing the team into the process, creating room for safe experimentation, and ensuring that internal controls remain solid are key to avoiding excessive dependence on a few people or a single tool.
AI in Finance in 2026: Less Fantasy, More Real-World Use Cases
The current picture of Artificial Intelligence and automation in finance is far less glamorous than the marketing pitch, but much more interesting for anyone focused on practical outcomes. On one hand, data shows that only a minority of CFOs see a strong impact on results. On the other, we are starting to see concrete stories from teams that:
- Cut payment delays with reliable forecasts and proactive actions;
- Turn chaotic charts of accounts into a unified foundation ready for automation;
- Replace manual, exhausting processes with automated, trackable, continuous workflows.
The path that works usually follows a few simple principles:
- Start small, with a well-defined problem, and then scale as the model proves itself in practice;
- Work data and processes with the same priority as the technology itself;
- Keep internal controls strong, treating AI as just another executor to be audited, not an infallible oracle;
- Share learning across people, so AI skills spread throughout the team.
In this context, technology stops being an end in itself and goes back to basics: helping the finance function become less reactive, gain advance visibility into risks and opportunities, and free up more time for the decisions that truly move the business. It is not instant revolution; it is continuous evolution – but for those who manage to fit AI and automation into their daily routines, the difference is already showing up in the numbers and in the way work gets done.
