Ecommerce ops · Production

How Instacart Uses Embeddings to Improve Search Relevance

The problem

Instacart's catalog spans over 1 billion products from 900+ retailers, and search queries follow a highly skewed distribution where fewer than 1,000 popular queries account for more than half of traffic while a long tail of queries lacks reliable engagement or categorical signals for determining relevance.

Workflow diagram · grounded in source
1
Customer issues search query
trigger
“identify the intent of customers' search queries, retrieve the most relevant products from the catalog, and rank them”
2
ITEMS encodes query and products
ai_action
“ITEMS takes search queries and product information as inputs, and projects them into the same vector space so that queries and products can be directly compared against each other for relevance”
3
ANN retrieval via FAISS index
integration
“The product embeddings are organized into indices using the FAISS library and served in an approximate nearest neighbor (ANN) service with daily updates to the indices”
4
Embedding-based product retrieval
ai_action
“We use the embedding-based retrieval (EBR) system to retrieve the most relevant products for search queries from the ANN service based on their semantic similarity. Complementing the existing keyword-based and category-based retrieval, E…”
5
Ranking with embedding score
ai_action
“embedding score plays an important role as a feature in ranking functions, which balances multiple objectives including relevance, popularity, and personal preference”
6
Ranked products surfaced to customer
output
“surfaces products that best fits the customers' needs”
Reported outcome

ITEMS improved mean reciprocal rank by +1.2%, cart adds per search by +4.1%, and produced a substantial increase in gross merchandise value, while achieving 100% search query coverage with under 8ms latency.

Reported metrics
mean reciprocal rank (MRR) of first converted item+1.2%
cart adds per search (CAPS)+4.1%
gross merchandise value (GMV)substantial increase
Search query coverage100%
Show all 5 reported metrics
mean reciprocal rank (MRR) of first converted item+1.2%
cart adds per search (CAPS)+4.1%
gross merchandise value (GMV)substantial increase
search query coverage100%
query embedding latency<8ms
Reported stack
ITEMSSentence TransformersFAISSGriffinFeatureStore
Source
https://tech.instacart.com/how-instacart-uses-embeddings-to-improve-search-relevance-e569839c3c36
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

ITEMS improved mean reciprocal rank by +1.2%, cart adds per search by +4.1%, and produced a substantial increase in gross merchandise value, while achieving 100% search query coverage with under 8ms latency.

What tools did this team use?

ITEMS, Sentence Transformers, FAISS, Griffin, FeatureStore.

What results were reported?

mean reciprocal rank (MRR) of first converted item: +1.2%; cart adds per search (CAPS): +4.1%; gross merchandise value (GMV): substantial increase; Search query coverage: 100% (source-reported, not independently verified).

How is this ecommerce ops AI workflow structured?

Customer issues search query → ITEMS encodes query and products → ANN retrieval via FAISS index → Embedding-based product retrieval → Ranking with embedding score → Ranked products surfaced to customer.

Related ecommerce ops cases