With 30 years of experience, expert.ai stands as a seasoned player in the financial services industry. Yet, according to Andrea Ricotti, SVP of Sales North Europe & Channel Development EMEA, the company still considers itself a startup due to its flexibility and agility to meet modern requirements.
The AI landscape is evolving rapidly, and the key for many organisations is to stay flexible. By doing so, many organisations are able to thrive in the changing market.
As regulatory requirements grow more complex, particularly around KYC and AML, Ricotti outlined how expert.ai is supporting financial services firms by providing explainable, regulated, and trustworthy AI solutions designed to stand up to real-world scrutiny.
Remaining compliant
A vital aspect of any company’s mission in the fincrime market of 2025 is to be able to remain compliant with evolving AML and KYC regulations whilst offering AI-based solutions. How does expert.ai do this?
“The first step,” Ricotti explains, “is to be prepared—to know as much as we can about the different regulations and requirements in each market.”
“We need to understand exactly where AI is going to be used,” he continues. “AML processes in banks and financial institutions are strictly regulated. That means any AI solution must be trustable, explainable, and traceable. And in our vision, it should be also human-oriented as AI will not replace human beings, and will not replace the subject matter expert that use it,”.
Ricotti emphasizes that AI must be transparent to satisfy regulatory requirements. “You need to be able to explain to a regulator why a specific customer has been rejected or flagged. That’s why our unique selling point is that our AI grid matches those explainability and compliance requirements.”
He also warns against oversimplifying enterprise AI. “Everyone has used ChatGPT or other generative tools, but the big mistake is thinking that the same ease of use applies to enterprise processes like AML. It doesn’t.”
Instead, Ricotti advises firms to be cautious: “Evaluate the expertise and tools behind the AI solution—make sure it can prevent hallucinations, inaccuracy, or other problems that might arise from a purely generative AI approach.”
Hybrid AI approach
In an age where having the upper hand in the fight to lead the AI space is becoming ever more vital, expert.ai states that it offers a hybrid or composite AI approach, that combines different techniques.
The expert.ai hybrid AI solution suite provides a thorough understanding of language, from complex documents to social media messages, and turns it into knowledge and insight. This makes for faster, better decisions without all the manual, time-consuming work.
Ricotti said, “We use symbolic and deterministic AI to control the results produced by other AI components like generative AI, which can summarize, create reports, or extract information. But these outputs must always be monitored.”
The company also integrates agentic AI, which can act autonomously to execute tasks or coordinate the different pieces of technology. The result is an AI framework that is both powerful and safe.
According to Ricotti, the benefits are substantial. “Productivity is boosted, because the auto-summarisation of risk reports and news summaries that would take hours of manual effort can be done quickly. At the same time, our approach improves capacity. We’re living in a moment when the number of weak signals and news items that need to be monitored has exploded. From this perspective, AI can be a fantastic tool to increase the capacity, quality and nature of these controls.”
Yet the goal is not to replace human decision-making: “We are not speaking of an agentic AI flow that replaces human beings. This has never been, and will never be, our vision at expert.ai.”
Ricotti stressed that whilst AI supports and increases productivity, the final decision and the final click will always be taken by a human.
“We make sure that the early steps are done in a controlled way, limiting false positives and hallucinations, so the subject matter expert can review a clean, pre-approved analysis and make a confident decision,” said Ricotti.
Fairness and control
One of the most well known problems in AI surrounds the challenge of bias, accuracy and hallucinations.
This is a key challenge for expert.ai too, and through its EidenAI Suite – which seeks to transform how organisations leverage knowledge through the strategic use of AI -seeks to ensure fairness and control.
As Ricotti outlines, AML and KYC process are tightly regulated, which means that its vital to not accept hallucinations or bias – with the results firm get from AI must be controlled and monitored.
He explains that this is built into the hybrid AI framework: “For every task in the process—whether it’s information extraction, summarization, or classification—we use the most suitable form of AI, that are most important for some specific tasks within a process. Maybe I would use symbolic AI if I needed to extract information for a rules-based approach; while if I want to summarise or to create an abstract, I would use GenAI.”
The company’s EidenAI Suite gives the subject matter expert full control at every stage, and if classification or extraction seems off, users can isolate and fix that task without disrupting the entire process.
Ricotti also highlighted that the firm has included factchecking layers to analyze news and customer data, helping to prevent misinformation, with expert.ai’s tools mirroring the steps a human would take.
Challenging false positives
A big challenge in the AML market today is false positives. How does expert.ai held reduce false positives whilst improving the efficiency of AML procedures?
In order to tackle the challenge of false positives, expert.ai tackles them at two levels. The first is entity matching.
“Say I’m searching for Mr. Jack Ryan. The AI needs to determine whether the Jack Ryan mentioned in the news is the same one as my customer. We use atomic data points— age, occupation, city, date of birth—to verify the identity,” said Ricotti. This data is extracted and then matched with the customer record. At the end, the system gives you a score: 100% match, 60% probability, for example.
The second level is news severity scoring. “In financial crime, we know for sure that standard bad news malpractices like terrorist financing and money laundering are critical. From that perspective, each news that could be a signal of this malpractice can be immediately notified with a certain level of severity.
The company evaluates how serious the reported event, with expert.ai’s technology considering the nature of the crime and the role of the individual, then generating a severity score.
These two scores, Ricotti explained, are entirely customisable. He remarked that the client can define the threshold for notifications or decide to ignore results below a certain match probability.
Adoption timeline
What does a typical AI adoption timeline look like for a large enterprise working with Expert.ai? In this area, the first step for the company comes with managing expectations.
Ricotti remarked, “We often have customers asking for 100% accuracy in four weeks, because they think that GPT, for example, can be used and implemented everywhere That’s not possible—especially not in AML or financial crime, which are highly regulated and complex processes. You need to set expectations – you cannot expect to reduce false positive rate by 100% and expect to have a solution in two weeks.”
Ricotti explained that the firm splits the overall timeline into two different phases. It starts with the demonstration phase, where the advantages of using AI are demonstrated to the customer. “We ask the customer for sample documents, a set of clients, and examples of past analyses. We use that to demonstrate how our hybrid AI framework works on their real data and improve the quality of extractions and classifications granting the highest level of accuracy.”
Following this comes the delivery stage, where the company seeks to demonstrate to their customers that AI can scale and can work on millions of documents or customers, and how expert.ai can industrialise the solution.
Here, Ricotti says, expert.ai’s decades of experience come into play: “We are working with different banking groups in EMEA and some of them have more than 20 million retail customers. Imagine monitoring that at scale —daily— while minimizing false positives and ensuring high accuracy.”
Keep up with all the latest FinTech news here
Copyright © 2025 FinTech Global
