|Workflow platform that allows customer teams to build and update their fraud processes without needing to write code; enables set up of workflows that automate any type of fraud detection achievable on Sift's platform.
|Enables the classification of multiple types of fraud and abuse simultaneously on a single account via the use of global and custom models, and ensemble of models. A distinct fraud score can be computed for each listed type of fraud.
|Enables the detection of account misappropriation and produces a risk score that indicates when an account may be being used by someone other than the original creator.
|Enables migration from an old risk scoring model to a new risk scoring model for a given customer to address changes and trends in fraud patterns. The calibration keeps score distributions stable even when Sift migrates customers between model types.
|Provides an analytical framework for evaluating anomalous shifts in risk scores for a given customer, allowing Sift to validate a new scoring model for a given customer before deployment and blocks deployment until validated.
|Prevents interferences between customer analysts reviewing a transaction within the Sift platform, providing real time updates to systems and client browsers interacting with the review queue.
|Varies the weights on a per customer basis of the models that make up Sift’s global scoring model to generate more accurate and specific risk scores.
|Produces a prediction that includes a risk score for posted content, and may include multiple distinct models that operate together as a unified risk model to predict whether abuse or fraudulent content is likely to occur.
|Provides a proxy model for interpreting complex black box models by constructing a surrogate model that mimics outputs of a black box model.
|Provides automatic multi-factor authentication to Sift's service, enabling both direct verification requests and verification requests triggered by an automated workflow. Verification data may be used as training data to improve customer-specific models.
|Enables a customer to Sift's service to determine false positive rates in declines or adverse decisions output from automated workflows.
|Allows for automated anomaly detection in decisions output from automated workflows.
|Implements text clustering models and techniques to surface fraudulent or abusive patterns in online content across users.
|Selectively identifies salient signals for card testing and converts those signals into learnable features that may be added to an existing machine learning.
|Enables robust testing of workflow routes for identifying optimal routes for improving automated proposals for digital handling.
|Enables bulk labeling of corpora of data samples using a variety of techniques for exploring and identifying groups or networks of fraudulent and legitimate data samples.
|Enables the creation of a connected component graph or network for exposing potential large scale attacks, such as bot attacks.
|Introduces an active learning-informed data sampling technique for creating a labeled corpus of samples for effectively training a model.
|Provides several mechanisms for automatically creating workflows and workflow routes for new and existing Sift customers.
|Creates and enables an automated agent for accelerating chargeback disputes by automatically scoring the success of a chargeback based on known transaction evidence and proposing transaction evidence that may improve the probability of success.
|Enables customers and partners to integrate with and interchange data through Sift's systems by an extensible webhook service.
|Second patent allowed
|Identifies anomalies in risk score distributions including shifts or drifts to generate an explanation for the anomalous behavior(s) together with corrective actions taken to mitigate the anomalies.
|Second patent pending
|Identifies fraudulent automated bot activities and generates a unique bot signature for each distinct bot that is detected and which can be leveraged in real-time bot identification to accelerate detection and threat mitigation posed by malicious bots.
|Allows customers to evaluate and respond to events based on multiple scoring criteria by introducing a technique that assigns both an event score and one or more percentile scores to each evaluated event, and enables customers to design related workflows. This invention involves using a T-Digest algorithm for calibration.