How We Implement ML To Enhance Products, Workflow, and Customer Experience

An AI-first approach is a core pillar of MD Finance's business strategy, reflecting how AI/ML technologies enhance both internal processes and product efficiency.

Serhii Serednii, Head of AI/ML at MD Finance, notes that the past decade has brought a breakthrough in fintech thanks to innovative technologies. Below, he shares how these technologies help MD Finance outperform competitors.

Let ML Decide: The Credit Scoring Evolution

Automation in lending and scoring has existed since the mid-20th century, but when we talk about digital lending today, we no longer refer to traditional rule-based scoring with questions like “Do you have a higher education?” or “Are you married?” – we are talking about a comprehensive decision tree and its reliable automation.

The shift towards analyzing increasingly large volumes of data is driven by a fundamental difference in business models. Traditional banks typically provide customers with a few large loans in their lifetime (such as mortgages or car loans), while fintech companies focus on small amounts that clients can receive monthly.

Accordingly, the approach to lending and risk management differs. In fintech, the depth of the decision tree can reach thousands of nodes. To automate this process, we develop proprietary ML algorithms based on models like Gradient Boosting Machines.

These ML methods are particularly effective for estimating each client's lifetime value and evaluating the potential for long-term cooperation. This allows us to balance desirable conditions for customers, attract new users, and maintain strong profitability.

Behind Limit Strategies

ML also helps fintech companies answer a critical question once a client is approved: How much money can be lent? It might be helpful to try ML-driven limit strategies inspired by Recommendation Systems, models typically used in products that adapt to user behaviour and preferences. Almost everyone has experienced such models in action: the content suggested to you on TikTok or YouTube is powered by them. At MD Finance, we've adopted a similar approach, tailoring it to determine the optimal credit limit for each user.

The comprehensive strategies we create allow us to assess the business impact on credit risk (the risk of a client not repaying a loan) and the risk of under-receiving profit (i.e., missed opportunity from issuing loans with lower limits or declining them entirely).

Over the past year, our AI/ML team has designed and tested more than 10 robust models to explore ways of improving product workflows across selected markets in Europe and Latin America.

See, Verify, Confirm

Since 2020, we have been gradually integrating deep learning-based verification technologies into our product ecosystem to strengthen fraud prevention and streamline identity verification workflows. At that time, such tools were not yet widespread across the industry – we adopted them out of practical necessity, not following trends.

One of the systems we developed focuses on verifying a user’s face during onboarding or transaction approval. It combines facial recognition with liveness verification to ensure the input comes from a real person, not a photo or a video. Engineered for speed and reliability, the tool delivers results within milliseconds.

What makes this solution effective is its ability to cross-check each face against a growing database. This helps us detect users attempting to submit multiple applications under different names, which is a common tactic in fraud cases.

Built to be adaptable across diverse environments, the system supports real-time verification without delays or adding friction for the user.

As always, all operations are conducted in full compliance with data protection regulations, including the GDPR. For us, legal, ethical, and technical constraints are not boundaries, but a framework for responsible innovation.

Controlling the Quality of 100% of Phone Calls

Our team experienced something off the beaten track working with LLM models, which are now used in a new tool to assess call center employees.

Previously, in a large call center, supervisors could only randomly review a small percentage of calls to ensure professionalism and adherence to the call script. However, with a large team, reviewing even 10% of calls was physically unfeasible.

Now, the AI tool allows the call center to:

• Transcribe 100% of incoming and outgoing calls in the local language and translate the transcripts into English.

• Assess call quality based on predefined criteria, such as detecting deviations and incorrect responses compared with the original call script.

• Generate individual reports to support supervisor feedback and employee training, and management reports to monitor the call center’s effectiveness and overall performance.

This tool is perfect for scaling a business’s quality standards across multiple countries. It allows for transparent, data-driven assessments of employee performance and offers insights for the leadership team on user problems and service issues.

At MD Finance, we have already been successfully applying machine learning and artificial intelligence to scoring, fraud detection, and customer service, resulting in clear improvements in operational efficiency.