Event-based fraud detection with direct customer calls using Amazon Connect (build/train/deploy walkthrough)
This AWS walkthrough describes an event-driven fraud detection flow that notifies a cardholder immediately when a suspicious credit-card transaction is detected. The solution integrates Amazon Fraud Detector with Amazon Connect so a customer can be called in real time and confirm whether a transaction was fraudulent. Customer attributes are stored in Amazon DynamoDB and updated by AWS Lambda during the call flow.
- Organization
- Not applicable
- Industry
- Finance
- Location
- United States
- Published
- June 2021
Reported outcomes
Strategic outcomes
Primary read
Use case focus
Showing 3 of 3
- 1Fraud detection
- 2Customer service automation
- 3Contact center automation
- Credit-card fraud is often detected after the card is blocked, leaving the customer without immediate context.
- Organizations need a low-latency workflow that can notify the cardholder and decide whether to block the card based on customer interaction.
- A transaction event is sent to AWS Lambda, which invokes an Amazon Fraud Detector model to score the transaction.
- If the transaction is flagged as fraudulent, Lambda retrieves customer data from Amazon DynamoDB and sends it to Amazon Connect to place an outbound call.
- The Amazon Connect contact flow lets the customer respond via DTMF and, if fraud is confirmed, invokes another Lambda function to set a block flag in DynamoDB.
- The walkthrough uses AWS CloudFormation to deploy the supporting resources and includes an Amazon SageMaker notebook workflow for building and training the Fraud Detector model.
- The reference implementation provides an end-to-end serverless demo for immediate fraud notification and customer verification.
- It demonstrates how prediction thresholds and downstream actions can be tuned and tested in a working call-and-block flow.
- The architecture reduces manual intervention by automating detection, customer contact, and account protection.
Architecture
Event-driven serverless architecture: transaction metadata triggers AWS Lambda, which calls Amazon Fraud Detector; if fraudulent, Lambda looks up customer data in Amazon DynamoDB and initiates an Amazon Connect outbound call; the contact flow uses DTMF to collect customer input; a second Lambda function can update DynamoDB to block the card; AWS CloudFormation provisions resources; Amazon SageMaker is used in the model build/train/deploy walkthrough.
Sources & evidence1
AI-generated summary. Verify important details with the linked sources before relying on this case.
Explore related AI use cases
Was this useful?