E-Sentinel is now part of the Eura technologies accelerator, Europe’s premier tech hub with a legacy of driving retail innovation. Nestled in Roubaix—France’s historic retail and e-commerce heartland, Euratechnologies has been the launchpad for industry leaders shaping the future of digital commerce.
Retail fraud is evolving at a rapid pace, with businesses facing rising threats fromrefund abuse, chargeback fraud, account takeovers, and logistics scams.
By joining forces with Eura technologies' vast retail ecosystem, E-Sentinel is accelerating its mission to deliver the most advanced AI-powered fraud prevention solutions— helping retailers stay one step ahead of fraudsters while maximising revenue protection and operational efficiency.
With deep industry expertise, cutting-edge technology, and a relentless focus onsecurity, we are setting a new standard for fraud detection in retail. This partnershipstrengthens our ability to combat fraud at scale, ensuring that businesses can growwith confidence in an ever-evolving threat landscape.
Welcome to the future of frictionless, fraud-free commerce.
Big Data and Machine Learning
Big data refers to a large volume of structured or unstructured data that is collected from various sources, such as social media, online transactions, and machine-generated data.
Big data is characterized by its volume, variety, and velocity, and it requires advanced technologies and techniques to process and analyze the data effectively. Big data is used to derive insights, patterns, and trends that can be used to make informed decisions and gain a competitive advantage in various fields
E-Sentinel needs to get a big amount of data, both internally and externally in order to draw relevant patterns and rules to fight online fraud.
Internal data refers to the data that is collected from E-Sentinel’s own systems, such as customer information, transaction data, and website traffic. External data, on the other hand, is data that is collected from external sources, such as public records and third-party data providers.
On the other hand, machine learning (ML) is a subset of artificial intelligence (AI) that involves building algorithms that can learn and improve from experience without being explicitly programmed.
Machine learning algorithms are used to process and analyze big data to identify patterns and make predictions or decisions. ML algorithms use statistical models to analyze data and make predictions based on the patterns and relationships found in the data. The more data that is available, the better the accuracy of the predictions.
Therefore, big data is a critical component of machine learning and is essential for E-Sentinel to identify patterns and trends that can be used to detect fraudulent activity.
Initial Training and Statistical Models
After gathering a large amount of data, preprocessing it to remove any noise or irrelevant information, ensuring it is in a format suitable for analysis and selecting a suitable algorithm, E-Sentinel will train its machine learning models.
Various algorithms based on different statistical models (such as Decision Tree, Logistic Regression, Naïve Bayes, Random Forest or Neural Network) will be tested in order to pick the most suitable ML model.
One of the main problematics will be to identify fraudulent transactions without applying subjective rules, in order not to influence the model and end up with a biased training ML model. A potential solution has been found to this problematic and will be tackled in a future article.
Another problematic is the unbalanced data that E-Sentinel needs to analyze. As a matter of facts, among the data collected, only a few (in proportion) will be labelled as fraudulent which could lead to ML models very precise but useless as the fraudulent transactions represent only a fraction of the total transactions analyzed.
Over sampling methods will be put in place in order to compensate this lack of fraudulent data, one of the methods that can be used is the SMOTE (Synthetic Minority Over-sampling Technique). This will help improve the efficiency of the model by increasing the recall, getting more prediction of the underrepresented group (fraudulent transactions).
Once the model has been trained and tested, it can be deployed to identify fraud in real-time transactions. As new data is fed into the model, it will use its learned patterns to classify each transaction as fraudulent or legitimate. Over time, E-Sentinel will continue to collect new data and use it to retrain the model to ensure it remains effective at identifying new types of fraud.
Patterns Identification & Predictive Model
Once the previous steps are done, the model that was trained to establish additional rules to counter fraud will be tested on the testing sets. This model is based on the various software’s parameters and highlight behavioural links among E-Sentinel’s features and rules.
If the tests are conclusive, these new rules and behavioural analyses will be added to E-Sentinel’s software.
The model will be trained at various levels:
- The first training will be on a global level, testing the data from all sources’,
- A second model will focus on the internal data only’,
- A third model will tackle e-commerce specific data, a model per client’,
- A forth model will be focused on industries/ Product categories data (when enough data is available), e.g. Electronic products, Apple products, specific brands...
The ultimate goal will be to identify global, local and specific trends to enhance the detection of patterns at all levels. E-Sentinel’s client will be able to add new rules and adapt the behavioural analyses on various levels.
These models will on the one hand enable E-Sentinel’s clients to get the best fraud detection software thanks to its supervised learning models but also on the other hand to benefit from a predictive fraud model using artificial intelligence based on reinforcement learning.
E-Sentinel models will give instant fraud prevention recommendations to each existing and new transactions and will highlight potential red flags that could lead to a fraud even before the fraud attempt takes place (Predictive Model).
NLP Models
Natural Language Processing (NLP) is a subfield of artificial intelligence that deals with the interaction between humans and computers using natural language. NLP tools, such as ChatGPT or LLaMA, can be implemented in E-Sentinel using an Application Programming Interface (API).
To implement NLP models, E-Sentinel’s team will need to first train the NLP tool on a large dataset of relevant data.
This dataset will need to be labeled and annotated to ensure that the model can accurately identify fraudulent behavior. For example, analyzing the fraud literature in order to get all documentations regarding existing and new fraud types and methods (Technology Watch).
Once these are stored and trained on the model, the model can be implemented to flag fraudulent methods and customers.
These models can provide valuable insights and assistance in identifying fraudulent activity and protecting e-commerce businesses from financial loss. There are several possible uses of such NLP models for E-Sentinel.
They can be used to analyze online information including social media. It can also be used to generate relevant recommendation for each specific case and contact stakeholders involved in the delivery process, for example, to generate automated correspondence.
Other uses include: automated chatbot, customer service, flag fake reviews, flagging suspicious interactions with customer support, etc.
Conclusion
E-Sentinel will use various artificial intelligence (AI) models to identify fraudulent transactions and the models cited above are the ones that will be implemented in the first place.
With time, the processes, checks and feature included in E-Sentinel analysis will evolve and become more and more accurate. The fraud detection ratio of 55% of total online fraud prevented will increase significantly Year-on-Year, especially thanks to its predictive fraud model.
Eventually, E-Sentinel’s clients will be able to give more freedom to their customers through lighter restrictions as their backs are covered by our fraud prevention software.
