Custom AI Development Case Study
Client Profile
Intro
For a fast-growing European payment platform, regulatory compliance can never be an afterthought. Serving customers across 37 countries in the EMEA region means complying with 37 distinct sets of regulations in 21 languages. A slight misstep can draw the ire of regulators and undermine customer trust.
The Compliance Challenge
The Challenge
Each jurisdiction where the payment platform operates has its own set of financial regulations, anti-money-laundering laws, and payment processing requirements. All of these are prone to change at an unpredictable pace.
Before the regulatory monitoring system was put in place, the team of 12 people spent 60+ hours a week browsing regulatory websites, newsletters, and official gazettes to catch applicable changes in time. Despite all that work, it took an average of 28 days to detect a regulatory change.
Manual monitoring, while time-consuming, was far from the only bottleneck for the compliance team. As regulations came in 21 languages, they had to be translated and interpreted before anyone could assess their impact.
The human factor wasn't helping, either: different team members routinely reached different conclusions when evaluating the impact of a new regulation. The result? Some regulations triggered false alarms, while other critical legislative changes flew under the radar, catching the team by surprise later on.
Solutions That Didn't Work
The company had tried to solve its compliance challenge several times, including reaching out to a variety of AI development companies.
In 2023 and 2024, the company increased the team's headcount, implemented off-the-shelf compliance software, and attempted to develop an internal proof of concept for a custom system.
Wake-Up Call
In Q1 2024, the inefficiencies in regulatory monitoring caught up with the company, landing it in hot water.
The team missed a critical change in AML reporting requirements in a central European market. So, when it was time to submit the AML report, the company didn't provide all the necessary information to the regulator.
This single missed regulation almost led to a hefty fine and a PR disaster for the platform.
The root cause behind this close call was evident: ineffective manual monitoring.
This was the company's wake-up call: something had to be done about the regulatory monitoring and compliance. The C-suite decided to pause market expansion until they could be confident in the timeliness, precision, and effectiveness of regulatory monitoring processes. The solution was clear: the company needed an automated regulatory monitoring system adapted to its needs. However, at that time, nobody considered AI implementation as a valid solution.
Building the Foundation of the Custom AI Solution
The Solution
Ensuring the new regulatory monitoring system would indeed be effective for the company required an in-depth assessment of both the existing workflows (including a comprehensive jurisdictional analysis) and the feasibility of the custom AI solution.
The insights gained during the discovery phase also informed the regulatory taxonomy and impact scoring framework. A thorough analysis of historical regulatory data over the past six months ensured the framework was tailored to the compliance team's needs.
From a technical perspective, we discovered that artificial intelligence is the right toolset for such a challenge. First, AI agents can consistently collect the required regulatory updates, and a custom AI solution can be built to detect and classify potential impacts.
You may read about the details about AI Advisory Service here
Custom AI Solution Features
The AI-powered pilot for an intelligent regulatory monitoring system was designed to solve three key tasks:
The AI system consisted of three key components: AI agents capable of automated source monitoring, intelligent classification built on custom model development, and relevance and impact scoring, a complex AI mechanism that served as the system's core.
During the discovery stage, 80+ regulatory sources in 21 languages were identified as relevant for the payment solution provider. Country-specific AI agents regularly check each of these sources for new publications and promptly store new updates. After that, the specific AI translation agent ingests and automatically translates publications from 21 languages into German, French, and English.
As a result, the compliance team reports spending up to 40% less time on manual monitoring. The time saved is expected to increase over time as the solution's output gains the team's trust.
Read more about how AI Strategy Discovery helps uncover similar challenges
The client's historical regulatory assessments served as the training datasets for the underlying natural language processing (NLP) model. The training ensured that the generative AI model's output aligned with the compliance team's established approach to classification and impact analysis.
Thanks to this alignment, the team receives information on every new regulatory change in a familiar format, without having to adapt to a new taxonomy.
The intelligent AI solution automatically classifies clauses by several parameters: theme (AML, Licensing, Cross-Border, etc.), modality (obligations versus prohibitions), necessity (exceptions, waivers), and time horizons.
Read more about AI/LLM model customisation and training
Once the clause is categorised, the AI solution assigns a relevance and impact score between 0 and 10 to predict the consequences of the regulatory change for the client's operations. Such classification required a significant amount of time during the AI software development timeframe. Both the business team and the AI development team worked closely to review numerous real-world scenarios, building unique and meaningful scoring criteria.
In line with responsible AI tenets, each impact score comes with a confidence indicator. Thanks to it, the team doesn't have to guess how reliable the score is and can double-check the impact assessments if necessary.
That said, training the AI model on historical assessments enables it to return consistently precise impact scores.
"The relevance scores are surprisingly accurate. I start relying on them now." - Compliance team member
How Successful AI Implementation Works in Practice
The regulatory monitoring system continuously checks the connected 80+ sources for relevant changes in legal requirements. Once it detects one, a compliance team member receives an alert on their dashboard. The alert specifies jurisdiction, a relevant set of documents with working links, the number of applicable restrictions and obligations, the date of entry into force, and an impact score with a confidence indicator.
Here's an example of the alert generated purely by an artificial intelligence solution:
13 March 2024. A new EU regulation [2024/886] has been published, with 4 new obligations and 2 restrictions, effective on 9 October 2025. Impact score: 8, confidence 94%.
Thanks to these automated alerts, the client's team members can focus on proactive analysis of the regulatory landscape instead of manually monitoring sources. The dashboard brings all alerts into one centralised place, reducing cognitive load and context switching.
User Training & Feedback Loops
To ensure the compliance team is fully aware of the AI solution's strengths and limitations, end users were trained on its use. Its curriculum was tailored to the team's existing processes so that the system could be introduced seamlessly into them, without much disruption. We also held a training session in which our AI experts explained how AI tools and applications work, what AI integration is, how systems leverage AI, and what the team can expect from such human-to-AI interactions.
This training enabled end users to make the most of the system and accelerate adoption. At the same time, it helped avoid confusion, frustration, and resistance to change, as most people were nervous about the AI performance and the probability of AI tools replacing professionals.
When the system became operational, we scheduled monthly system performance reviews and continued capturing feedback from users. Based on those reviews, we continued to enhance the AI model's accuracy and relevance algorithms.
Business Benefits
In early 2025, the client rolled out the regulatory monitoring system as a supporting tool. It is now used alongside existing business processes and other AI tools that we delivered.
Following the system's launch, the compliance team checks regulatory change sources twice a week instead of daily. That represents a perceived decrease in time spent on manual monitoring of up to 40%. The team expects the time savings to accrue over time as members become more accustomed to it, and the high output accuracy builds trust in the system as a whole and in artificial intelligence as an inevitable part of it.
The system also enables the team to pinpoint critical regulatory changes nine times more quickly. Instead of identifying a relevant change weeks after it's published, the team now detects it in 3 days on average. Most importantly, the team missed zero critical regulations since the system was deployed.
As the system consolidates all information on upcoming regulatory changes in one place, the team could add proactive regulatory landscape analysis to its reporting to the board. This provides executives with the visibility they need to make informed strategic decisions.
Now confident in the company's compliance-monitoring capabilities, the board approved an expansion into five new markets in 2026.
Engineering Team & Responsibilities
