Customer support · Production

Tyson Foods elevates customer search experience with an AI-powered conversational assistant built on Amazon Bedrock

The problem

Tyson Foodservice had limited direct engagement with over 1 million unattended operators who purchased through distributors without direct company relationships, and its keyword-based search frustrated foodservice professionals by failing to match culinary terminology to catalog descriptions, driving lost revenue opportunities.

First attempt

The earlier keyword-based search on the Tyson Foodservice website failed to handle terminology mismatches, causing chefs searching for 'pulled chicken' to miss products labeled 'shredded chicken,' and those looking for 'wings' to miss 'party wings' or 'drummettes.'

Workflow diagram · grounded in source
1
User submits search query
trigger
“A user uses the search bar in https://www.tysonfoodservice.com/”
2
Query vectorized via Titan Embeddings
ai_action
“The query string is converted to embeddings using Amazon Bedrock and the Amazon Titan Text Embeddings model”
3
k-NN vector search in OpenSearch
ai_action
“The search application performs a k-nearest neighbors (k-NN) vector search to find relevant results in Amazon OpenSearch Serverless and return those results to the website”
4
AI assistant processes natural language
ai_action
“The query is processed by the agent node using Anthropic's Claude 3.5 Sonnet on Amazon Bedrock. Depending on the subject of the query, the agent might orchestrate multiple agents to return relevant information to the user.”
5
Tool execution returns data
integration
“This node implements a generic tool executor that connects to various tools. Whenever a tool call is issued by the agent node, this node handles the execution of the tool call. The tool calling node executes the tools, which are defined …”
6
High-value actions logged to RDS
feedback_loop
“an Amazon Relational Database Service (Amazon RDS) database cluster to persist the user high-value actions for analytics purposes”
7
Conversational response delivered
output
“This AI assistant delivers a seamless conversational search experience that offers comprehensive support across Tyson's extensive range of products, recipes, and articles, providing contextual guidance through natural conversation”
Reported outcome

Tyson Foodservice deployed a generative AI assistant and semantic search on its website, enabling direct engagement with previously unattended operators, dramatically improving search relevance, and capturing high-value customer interaction data for business intelligence.

Reported metrics
Unattended operators without direct engagementover 1 million
Search relevance improvementdramatically improved search relevance
Customer interest measurement precisionunprecedented precision
Reported stack
Amazon BedrockLangGraph
Source
https://aws.amazon.com/blogs/machine-learning/tyson-foods-elevates-customer-search-experience-with-an-ai-powered-conversational-assistant?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Tyson Foodservice deployed a generative AI assistant and semantic search on its website, enabling direct engagement with previously unattended operators, dramatically improving search relevance, and capturing high-val…

What tools did this team use?

Amazon Bedrock, LangGraph.

What results were reported?

Unattended operators without direct engagement: over 1 million; Search relevance improvement: dramatically improved search relevance; Customer interest measurement precision: unprecedented precision (source-reported, not independently verified).

What failed first in this deployment?

The earlier keyword-based search on the Tyson Foodservice website failed to handle terminology mismatches, causing chefs searching for 'pulled chicken' to miss products labeled 'shredded chicken,' and those looking fo…

How is this customer support AI workflow structured?

User submits search query → Query vectorized via Titan Embeddings → k-NN vector search in OpenSearch → AI assistant processes natural language → Tool execution returns data → High-value actions logged to RDS → Conversational response delivered.

Related customer support cases