What challenges did Yahoo! face before AI implementation?

Legacy keyword expansion struggled with current vocabulary, semantic understanding, and sometimes produced no expansions. Yahoo needed broader yet semantically relevant audience targeting for search retargeting.

What was the impact of the AI solution?

Up to 600x increase in keyword expansion rates. Up to 5x growth in addressable audience reach. Improved campaign performance and conversions.

Source

ProductionEvidence: Medium50/100

Yahoo improves search retargeting audience keyword expansion using Amazon Bedrock generative AI

Use case typeCustomer targetingUpdated Jan 1, 2025

Yahoo DSP enhanced search retargeting keyword expansion with generative AI on Amazon Bedrock. The new workflow replaces a Word2Vec plus LSH approach with LLM-based expansion, embedding similarity checks, and sensitive keyword filtering. It launched in production in Q1 2025 and improved keyword expansion rate and addressable audience reach.

Organization: Yahoo!
Industry: Tech & Comms
Location: United States
Published: January 2025

Reported outcomes

Keyword expansion rate increase: Up to 600×

Other quantified impact

Addressable audience reach growth: Up to 5×

Strategic outcomes

Scale & capacityExpanded advertisers' keyword sets at much larger scaleCustomer experience & trustKept output semantically aligned with original intentRisk & complianceAdded sensitive-keyword guardrails for policy complianceCompetitive differentiationStrengthened Yahoo DSP's competitive edge in programmatic advertising

Why do we believe this?Outcome claims, sources, and evidence checks

Normalized claim

Keyword expansion rate increase: 600 x increase

AWS Machine Learning BlogJan 1, 2025Blog postExplicit claimMedium evidence strength

Up to 600 times increase in keyword expansion rates.

Normalized claim

Addressable audience reach growth: 5 x increase

AWS Machine Learning BlogJan 1, 2025Blog postExplicit claimMedium evidence strength

Up to 5 times growth in addressable audience reach.

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks

Customer: Yahoo!
Provider: AWS
Maturity: Production
Linked source: AWS Machine Learning Blog

It launched in production in Q1 2025 and improved keyword expansion rate and addressable audience reach

Customer identity supportedSource describes one deploymentMaturity supported

Primary read

Use case focus

Showing 2 of 2

1Customer targeting
2Search modernization

Legacy keyword expansion struggled with current vocabulary, semantic understanding, and sometimes produced no expansions.
Yahoo needed broader yet semantically relevant audience targeting for search retargeting.

Yahoo prototyped multiple LLMs in Amazon SageMaker Studio and selected Anthropic Claude 3.5 Sonnet v2 on Amazon Bedrock.
Expanded keywords are verified with embeddings and similarity scoring, and sensitive keywords are filtered before and after inference.
Updated keywords are stored in Amazon OpenSearch Service for batch scoring against user search histories.

Technologies

Amazon Bedrock Amazon SageMaker Studio Amazon OpenSearch Service Anthropic Claude 3.5 Sonnet Embeddings

Up to 600x increase in keyword expansion rates.
Up to 5x growth in addressable audience reach.
Improved campaign performance and conversions.

Architecture

Advertisers define seed keywords in Yahoo DSP. A backend workflow expands those keywords using LLMs in Amazon Bedrock, with models prototyped in Amazon SageMaker Studio. Expanded terms are embedded and similarity-scored against the originals to remove low-similarity outputs, sensitive terms are filtered before and after inference, and the resulting keyword sets are stored in Amazon OpenSearch Service for batch scoring against user search histories.

Sources & evidence1

Evidence: Medium50/100Evidence strength