Virtual Assistant Fraud Detection via Audio Analysis

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Individuals are increasingly vulnerable to fraudulent activities via telephone calls, as they often unknowingly provide sensitive information to potential scammers, leading to issues like identity theft, and existing systems lack effective methods to efficiently detect and prevent such fraudulent interactions.

Innovation Solution

A system utilizing a virtual assistant application that interacts with callers, cross-references caller data with databases of known fraudulent activities, and employs artificial intelligence to determine potential fraudulent calls by analyzing audio data and prompting callers with simulated voice audio to gauge the purpose of their calls, thereby routing suspicious calls to designated parties or deleting voicemails associated with fraudulent activities.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional call screening systems are used, then basic call filtering is possible, but they cannot effectively detect sophisticated fraudulent activities or predict potential fraudsters

Engineering Contradiction:

Improvefraud detection accuracyVSAvoidsystem complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

A virtual assistant application is introduced as an intermediary component between the incoming call and the user. The virtual assistant performs preliminary analysis by interacting with the caller through simulated voice audio, analyzing audio data characteristics, and cross-referencing with databases of fraudulent activities before the call reaches the user, thereby improving detection accuracy without requiring complex manual review processes

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system replaces manual call screening and human judgment with automated AI-driven analysis. The virtual assistant uses machine learning models to analyze audio patterns, speak-to-text processing, and automated database cross-referencing to detect fraudulent activities, substituting the mechanical process of human operator review with intelligent automation that can process multiple calls simultaneously

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If manual call screening is performed, then user control over calls is maintained, but it is time-consuming and cannot scale effectively

Engineering Contradiction:

Improvecall screening efficiencyVSAvoidtime for call analysis

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The virtual assistant performs preliminary analysis and classification of incoming calls before they reach the user. By pre-screening calls using audio data analysis, pattern recognition, and database cross-referencing, the system identifies fraudulent or suspicious calls in advance, allowing the user to receive only verified legitimate calls or be alerted to potential threats, thereby significantly reducing the time required for manual screening

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system enables self-service call screening where the virtual assistant autonomously analyzes incoming calls, makes determinations about fraudulent activity, and takes appropriate actions such as blocking calls or alerting the user without requiring manual intervention. This automation allows the system to scale effectively handling large volumes of calls while maintaining consistent detection accuracy

Inventive Principle:

Principle #25Self-service

3Measurement precision

If call data is collected and analyzed, then fraud detection capability is improved, but user privacy and data security concerns arise

Engineering Contradiction:

Improvefraud detection precisionVSAvoiddata security risk

Core Design Contradiction:

Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The system extracts and analyzes only the necessary audio data characteristics and call metadata required for fraud detection, separating this analysis function from the main communication flow. By taking out only the essential features for analysis (such as audio patterns, timing, and content snippets) rather than collecting complete call records, the system achieves precise fraud detection while minimizing data retention and reducing security exposure

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The virtual assistant acts as an intermediary that processes call data through secure, isolated analysis functions before any potential user interaction. This intermediary layer ensures that sensitive audio data and call information are processed in a secure environment with controlled access, reducing the risk of data breaches while maintaining the precision needed for accurate fraud detection

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS11902463B1Systems and methods for detecting fraudulent calls using virtual assistants

Publication Date: 2024.02.13 UNITED SERVICES AUTOMOBILE ASSOCIATION (USAA)

AI summary

A system may include a processor that may execute computer-executable instructions that cause the processor to receive caller information regarding an incoming communication from a caller and receive a request from a user to route the incoming communication to a virtual assistant application. The virtual assistant application is configured to interact with the caller and determine whether the caller is associated a fraudulent caller activity stored on databases accessible by the processor. The processor may then receive an indication from the virtual assistant application that the caller is associated with the fraudulent caller activity and forward the incoming communication to another party in response to receiving the indication.