Ecommerce ops · Production

Amazon uses LLMs for catalogue-scale product listing quality control

The problem

Amazon's traditional specialized ML models, each optimized for an independent product category, handled simple structured attributes well but could not scale to products with complex or nuanced attributes, which required specially trained models or costly manual review.

Workflow diagram · grounded in source

Catalogue summarization

integration

“That process starts with summarizing and organizing the entire catalogue by product type and attribute value”

Attribute statistics analysis

ai_action

“These statistics are fairly good indicators of a value's correctness. If a higher number of products in a category uses a certain attribute value, for instance, or if products with a certain attribute value are more frequently viewed by …”

Iterative prompt tuning

validation

“an iterative process called prompt tuning, wherein general-purpose LLMs are exposed to particular schemas, rules, and terms that appear in the environment where they will be used”

LLM quality control tasks

ai_action

“it performs three main tasks: recognizing standard attribute values, to establish correctness; collecting alternative representations of standard values, or synonyms; and detecting erroneous or nonsensical data entries”

Catalogue attribute update

output

“the latest seller values are included in the catalogue more quickly (within days)”

Reported outcome

LLM-based quality control now corrects and updates product attributes at the scale of Amazon Stores, includes the latest seller values within days, saves thousands of hours in human reviews, and extends coverage to more languages.

Reported metrics

Time to include latest seller valueswithin days

Human review time savedthousands of hours

Reported stack

large language models

Source

https://www.amazon.science/blog/using-llms-to-improve-amazon-product-listings

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

LLM-based quality control now corrects and updates product attributes at the scale of Amazon Stores, includes the latest seller values within days, saves thousands of hours in human reviews, and extends coverage to mo…

What tools did this team use?

large language models.

What results were reported?

Time to include latest seller values: within days; Human review time saved: thousands of hours (source-reported, not independently verified).

How is this ecommerce ops AI workflow structured?

Catalogue summarization → Attribute statistics analysis → Iterative prompt tuning → LLM quality control tasks → Catalogue attribute update.