Amazon Web Services (AWS) has launched the Agentic Shopping Assistant (ASA). This new AI retail solution allows retailers to build their own conversational shopping assistants using technology developed for Amazon’s Alexa for Shopping service. The offering includes architecture guidance, starter code and deployment support from AWS, enabling retailers to create AI-powered shopping experiences tailored to their product catalogues, customer data, and brand identity.

Amazon built ASA on AWS services — including Amazon Bedrock, AgentCore, and OpenSearch — drawing on lessons learned from operating Alexa for Shopping. The company said more than 300 million customers interacted with its AI shopping tools last year, helping generate nearly $12 billion in incremental sales. AWS said retailers can deploy customised shopping assistants within weeks, rather than spending years building similar infrastructure independently, while retaining control over customer relationships, branding and proprietary business insights.

What is it? The agent is effectively a retailer-facing version of Amazon’s own AI shopping assistant infrastructure. Instead of relying on keyword searches and filters, shoppers can describe what they are looking for in natural language and receive product recommendations through conversational interactions.

Retailers can customise the assistant using their own catalogues, business rules and brand voice. AWS claims conversational shopping sessions achieve conversion rates 3.5 times higher than those of traditional keyword searches.

Unanswered questions: AWS has not disclosed how much of the Alexa for Shopping architecture it exposes to customers versus what it keeps proprietary. The company also has not explained how it processes, retains and uses customer conversations, prompts and behavioural data when retailers deploy the system. Similarly, AWS has not clarified whether shopper interactions are available to underlying model providers or whether all data remains within a retailer-controlled environment.

Questions for AWS: MediaNama has emailed AWS seeking clarification on the following:

1. What customer data, prompts and metadata do AWS, model providers, or third-party services process through ASA deployments?
2. Do retailers and shoppers retain their interactions, and can anyone use that data to train models or improve products?
3. Which foundation models are supported by default, and can retailers deploy third-party or self-hosted models?
4. What safeguards exist to prevent manipulative recommendations, dark patterns or preferential product ranking by AI assistants?
5. Does AWS provide transparency tools that allow retailers and consumers to understand why specific products are recommended?

We will update the article when we receive a response.

Also read

• Paytm bets on agentic AI and wealthtech for future growth: Q4 FY26 earnings call
• “We Are Moving From AI Systems That Recommend to That Act”: Mastercard on Agentic AI
• Ola’s Krutrim quietly shuts down its agentic AI assistant ‘Kruti’