Combine machine learning recommendations with generative AI to serve personalized product suggestions and content to each shopper across every channel in near real-time.
Overview
This Guidance helps retailers deliver individualized experiences to both identified and anonymous consumers by combining real-time behavioral signals with machine learning recommendations and generative AI. The system processes browsing events, transactions, and in-store interactions in real time through Amazon Kinesis Data Streams, feeding Amazon Personalize for tailored product recommendations and Amazon Bedrock for personalized content generation. An AI Shopping Assistant powered by Amazon Bedrock AgentCore enables conversational product discovery, while automated retraining workflows detect and respond to model drift to maintain recommendation quality. You can increase conversion rates and customer lifetime value by delivering relevant experiences across every touchpoint, regardless of whether shoppers are logged in or anonymous.
Benefits
Deliver individualized experiences at scale
Automate your personalization operations
Deploy self-healing ML pipelines that detect model drift and trigger retraining automatically. Reduce manual intervention while keeping recommendations accurate as customer behavior evolves.
Unify omnichannel retail signals instantly
Ingest and act on data from e-commerce, in-store IoT, POS, and CRM systems through a single streaming pipeline. Update customer profiles in near real-time to power timely engagement.
How it works
This architecture diagram shows how to build a retail hyper personalization platform that combines real-time data streaming, machine learning recommendations, and generative AI to deliver individualized shopping experiences.
Download the architecture diagram.
Step 1
Shoppers interact with the retail experience through the e-commerce site, in-store IoT touchpoints, and POS systems. Retail signals, including browsing events, transactions, and CRM records, flow into Amazon API Gateway, over TLS-encrypted connections, for downstream processing.
Amazon Kinesis Data Streams processes events in real time while Amazon Data Firehose delivers raw events to the data lake.
Amazon S3 stores training datasets and raw events. Amazon Athena federates queries across sources for analytics-ready data.
AWS Lambda processes streaming events, updating profiles in Amazon DynamoDB and forwarding signals to Amazon Personalize for identified and anonymous users.
Amazon Personalize delivers tailored product recommendations and rankings based on real-time interactions and historical behavior.
Amazon Bedrock, a fully managed service with security, privacy, and responsible AI controls, uses Retrieval Augmented Generation (RAG) to generate personalized content grounded in near real-time customer profiles and product data. Amazon SageMaker AI builds, trains, and deploys custom ML models using training data from Amazon S3.
The AI Shopping Assistant is deployed and operated using Amazon Bedrock AgentCore, a comprehensive set of services for securely running AI agents at scale. The AgentCore Runtime provides low-latency serverless environments with session isolation, supporting any agent framework, for conversational product discovery. Responses return via AWS AppSync.
Amazon EventBridge captures scheduled triggers and model drift events, sourced from Amazon SageMaker Model Monitor. This detects performance degradation, routing them to AWS Step Functions to initiate automated retraining workflows and campaign refresh pipelines.
Amazon ECS on AWS Fargate hosts microservices for A/B testing, campaign management, and recommendation blending.
AWS AppSync serves near real-time recommendations via GraphQL. Amazon Pinpoint delivers personalized email, SMS, and push campaigns.
Amazon Cognito authenticates users. AWS KMS encrypts data at rest. Amazon CloudWatch provides metrics and logs while AWS X-Ray enables distributed tracing across the request path. AWS IAM enforces least-privilege access.
Deploy with confidence
Everything you need to launch this Guidance in your account is right here.
Let's make it happen
Ready to deploy? Review the sample code on GitHub for detailed deployment instructions to deploy as-is or customize to fit your needs.