What's better, a classic anti-fraud approach or AI & ML-based one? Read on for the opinion of Volodymyr Todurov — the chief of data analytics at PIN-UP GLOBAL.

PIN-UP GLOBAL Holding specialises in multiple things, including anti-fraud protection. PIN-UP.TECH is one of its key segments, and its chief data officer, Volodymyr Todurov, is now focused on using the newest technology to create ultimate safety solutions.

Todurov explained how to PIN-UP anti-fraud models are created, (see here https://next.io/news/promoted/pin-up-global-ml-and-ai-combating-fraud/ to learn more).

Volodymyr Todurov & Risk Management: How It Works in Sports Betting

When asked about risk management in sportsbooks, Volodymyr Todurov focused on a few key points. He noted the main goal is to set correct individual limits considering multiple essential parameters like the specific sport, a tournament, an event and the betting markets.

Also, betting limits matter. The maximum bet amount and the time delay for accepting that bet should be considered, too. This goes mostly for managing the margins of the sportsbook.

Again, limits must be personalised, and different segments of clients may affect it, too. For example, different amounts may be set for beginners, regular customers, VIPs, high-rollers, etc.

The limits will also depend on the marketing actions and strategies. So, you need a flexible system that will satisfy the customer’s needs while adjusting the risks and evaluating profitability correctly. Outcomes

Volodymyr Todurov on Antifraud Tools: How Classic Pre-AI Solutions Work

Volodymyr was also asked about more classic schemes for detecting fraudulent behaviour, and he gave a clear explanation of the standard key principles.

In particular, Volodymyr mentioned that the standard approach always focuses on spotting and then handling anomalies.

Before you can identify them, you need to determine the norm, he says. Anything that differs from it a lot can be considered an anomaly that can and should be addressed.

That, for instance, is a huge change in customer behaviour during a tournament, an unexpected turnover in the pre-match and live bets distribution in one of the customer segments, skyrocketed average bets, skyrocketed maximum bets, etc.

All these trends are signs that you should examine what has happened much closer and carry out an in-depth analysis of a specific segment to identify the reasons for such a shift.

Also, Volodymyr Todurov highlights the importance of analysing customer-specific preferences.

Let’s say you have a client. How much does he usually bet? Which markets do they choose, and at what point in time do they bet? That’s how the system can spot a user who abuses errors in odds or uses the offer for any other kind of fraudulent behavior, for example, financial machinations, casino machinations, abuse of welcome and other kinds of bonuses, etc.

So-called arbitrageurs may also use software to benefit from this simple and common fraud scheme.

Still, despite its commonness, it’s not that effective for a fraudster because bookmakers also use special software to see arbitrage situations and take measures to restrict clients who choose only these specific markets to abuse the system.

Volodymyr Todurov’s Analytics Regarding AI: Which Benefits ML & AI Actually Offer

As noted by Todurov, gambling antifraud involves processing enormous volumes of information, and that’s when machine learning and artificial intelligence models become irreplaceable.

They can generalise data perfectly, and most importantly, they are really good at spotting different behavioural patterns, you won’t be able to spot by analysing transactions manually or using linear rules.

This improves and optimises the work for the entire team by reducing errors and saving time that experts may spend on other tasks while still gaining a lot of new information about the customers. Moreover, it’s not only about automation.

AI allows specialists to see patterns from new angles, dive deeper into various behavioural aspects, and improve their expertise.

So, the team can achieve better results without hiring more people. AI and ML also let the specialists focus on more interesting and important tasks, which can hardly be overestimated.

Volodymyr Todurov’s Anti-Fraud Platform: How PIN-UP.TECH Models Work

The model used by PIN-UP GLOBAL is designed to analyse many different criteria for every single transaction.

That goes for the amount that a user bets, the overall frequency of their bets, the events they pick, and the overall history of their actions on the platform. Moreover, the model not only analyses the above-listed parameters but also chooses the ones with the most predictive power and prioritises them over all other criteria.

Also, Todurov notes that the team uses the extreme gradient-boosting ML model, which is nearly perfect for structured data. This method implies building a group of models, each improving the other and hence constantly boosting the accuracy of final forecasts.

All in all, at the input, the team gives a history of bets made and gets the probability of 0 to 1 that a particular customer might be a fraud.

How the Model is Created: Insights into Building the Product

At PIN-UP GLOBAL, Volodymyr Todurov says, they focus heavily on the training data.

The more accurate and high-quality they are, the better the prediction. He notes that they hired a pro with deep expertise in gambling risk management. His main goal during over 3 months was to evaluate the segments of fraudulent customers and decide whether they were correct or not.

To meet that goal, he dealt with a lot of de-identified data from lots of real customers’ accounts, selecting the information that their training model would then use. Considering that the quality of this information is essential, such work greatly boosted the initial phase. Even the first models gave rather accurate results, showing acceptable levels of type 1 and type 2 errors.

The next significant step we needed to take, Todurov says, was the onboarding.

All our data science specialists learned enough about the system to understand how it works fully and use the software to boost results dozens of times.

We trained our models repeatedly, polishing them with information from different time periods in every iteration. That’s how our models became more advanced, Todurov notes.

Future of Betting Anti-Fraud According to Volodymyr Todurov

When asked about future plans, Todurov mentions that they are going to further enhance and advance their solutions based on generative AI and neural networks.

The process will be faster, he says, and the results are expected to be much more accurate.

The main benefit of choosing such an approach is that it’s capable of learning and constantly adapting. It provides an opportunity to get a full understanding of every single customer’s betting history and identify individual fraudsters or groups of members.

Volodymyr Todurov says there’ll be fewer routine tasks. Moreover, for big businesses, it’s just a top additional tool that allows you to segment and categorise clients. It’s a big relief for managers who’ll know exactly that they haven’t missed anyone suspicious.

Small businesses can benefit from the technology, too. It’s a great opportunity to optimise the processes without hiring more people.

Volodymyr Todurov is a chief analytics officer at PIN-UP GLOBAL. He has 8+ years of experience in gambling as a bookmaker, risk manager, anti-fraud, and data analytics. He specializes in DDDM, financial modeling, and making data-based predictions.