What is the AI confidence score? 

Risk Learning is an innovative feature that employs machine learning to enhance the application's ability to identify potentially risky transactions.  It analyzes transactional and subject data, as well as outcomes from our risk analytics, to continuously enhance its predictive accuracy.

The resulting AI Confidence score is a straightforward percentage representing the model's confidence in the likelihood of a risky transaction. The higher the percentage, the more likely the transaction will be risky. This capability not only provides a new data point but also increases operational efficiency by helping users make more informed decisions.

Your continued use of the system and your feedback play crucial roles in enhancing the AI model's risk assessment. The more transactions you score, review and provide feedback on, the more precise the AI confidence score becomes. 

Adding AI Confidence to your Threshold Settings 

Admins can utilize the Transaction Inbox settings to establish the AI confidence scores as a factor considered for determining which transactions are displayed in the Transaction Inbox default view. This can be done as an addition to the Risk Scoring (Risk Score AND AI Confidence) or as an alternative (Risk Score OR AI Confidence).

To enable/disable Risk Learning, see the instructions found in the Transactions Inbox Settings.

How to Enhance Risk Learning

As a user, your continued engagement with the product will continually drive the improvement of Risk Learning; however, there are several proactive strategies that users can adopt to enhance the AI's predictive accuracy. Here are key actions that can improve the model's performance:

Review of Transactions: Regularly reviewing transactions directly contributes to the model's learning process. When you start a review, you provide valuable data that helps the model recognize patterns associated with risky transactions. Additionally, closing transactions further trains the model on what is not risky. This continuous engagement is critical in calibrating the AI to identify true risks based on real-world outcomes.

Historical Analysis: Revisiting past transactions that led to investigations or identified issues such as fraud, bribery, or corruption is crucial. By beginning reviews on the transactions that these historical cases are related to, the model gains insights into complex risk patterns and learns from past issues or wrongdoings. This retrospective examination not only enhances the model's awareness but also its sensitivity to similar future transactions.

Broadening the Scope of Review: Expanding your review beyond transactions that initially meet pre-set risk thresholds can significantly improve the AI's learning range. Investigating transactions with lower or borderline risk scores can uncover subtle risk indicators that might otherwise be overlooked. This approach ensures a more comprehensive learning spectrum for the AI, enabling it to make more nuanced risk assessments.

Feedback on Risk Analytics: Utilize the feedback mechanisms, such as the thumbs up or thumbs down icons, adjacent to each risk analytic result. This input helps fine-tune the model by indicating whether the risk identified was accurately assessed or if it represented a false positive. Such direct feedback aids in adjusting the analytical parameters of the model, enhancing both its accuracy and reliability.

Implementing these strategies will not only refine the AI's capabilities but also extend its understanding of risk in varied contexts, ensuring a robust and effective Risk Learning system. This ongoing process is fundamental to achieving continual improvements in the model, aligning it closer with operational goals and compliance standards.

Difference from the Risk Score

The AI Confidence Score is related to but not the same as the Risk Score.  The scores can be different, with each one having merit.

• The Risk Score is a cumulative figure based on the settings for each underlying Risk Analytic. The more a customer configures the Risk Analytics to their specific risks, the more likely the Risk Score will highlight transactions that should be reviewed.
• The AI Confidence Score is a model-driven prediction that a given item would be reviewed, based on the past behavior of all Lextegrity’s customers.
• The AI Confidence Score benefits from Lextegrity’s Risk Analytics but, unlike the Risk Score, does not depend on the specific setting for each analytic. The model can determine the impact of each Risk Analytic based on review behavior without requiring the Risk Settings to be set up perfectly.
• This model is trained on over thousands of transactions and will continue to grow in accuracy as more transactions are reviewed.  

In leveraging the AI Confidence score, users are equipped with a powerful data point that can further enhance decision-making. This score helps identify transactions that bear the hallmarks of potential risk, similar to previously analyzed transactions, thus providing a critical layer of insight that complements our existing Risk Score. As the system continues to learn and improve, it becomes an even more vital component in a user’s arsenal to prevent or detect fraud, bribery, corruption, and other transactional risks, ensuring that our clients can maintain the highest standards of compliance.

Beta Status

The current beta status of our Risk Learning predictive model reflects a strategic approach toward achieving ultimate precision and reliability in risk assessment. At this stage, our focus is on enriching the model's training dataset with expanded real-world transaction data, which is crucial for refining its predictive algorithms.

By engaging in a beta phase, we aim to gather comprehensive feedback from our users, focusing on practical outcomes and user experiences. This feedback is invaluable as it provides direct insights into how the model performs under varied operational scenarios and highlights areas for potential enhancements.

Additionally, this beta period is indicative of our commitment to continuous development and improvement. It serves as a proactive period during which we actively refine and evolve the model’s capabilities, ensuring that it meets the highest standards of efficiency and accuracy.

Our goal during this beta phase is not merely to observe but to actively learn and adapt, guaranteeing that when fully launched, the model will be a robust, dependable tool in the arsenal of risk management strategies. We appreciate the collaboration of our users during this phase, as their contributions are vital to the successful enhancement and sophistication of our Risk Learning feature.