Overview

This is for:

Developer System Administrator

Coveo Machine Learning (Coveo ML) is a cloud-based machine learning service that helps you deliver relevant search results, grounded generated answers, and proactive recommendations through the use of advanced machine learning models.

In addition to its machine learning models, Coveo also offers AI-powered capabilities that build on these models to deliver conversational and agent-driven experiences. For more information on these capabilities, see Beyond machine learning models.

The Coveo suite of machine learning models and AI-powered capabilities drive better outcomes, such as greater customer self-service success, higher conversion and revenue, increased case deflection, and improved agent efficiency.

Coveo Machine Learning models

Coveo ML offers the following models:

Note

For Coveo Machine Learning (Coveo ML) models specific to Coveo for Commerce, see Machine learning for Commerce.

  • Automatic Relevance Tuning (ART): Learns from user search behavior to automatically boost the relevance of search results.

  • Case Classification (CC): Learns from your historical case data to predict field values for new cases, helping complete case forms faster.

  • Content Recommendations (CR): Analyzes user browsing and search behavior to recommend content that’s most likely to be relevant to the current user.

  • Dynamic Navigation Experience (DNE): Learns from user behavior to order the most relevant facets and facet values at the top for a given query.

  • IPX Recommendations (IPXRECS): Recommends relevant content to users in a Coveo In-Product Experience (IPX) search interface, based on previous search and click activity.

  • Passage Retrieval (CPR): Uses embeddings and vector search to retrieve the most relevant passages from your indexed content to ground Retrieval-Augmented Generation (RAG) systems and large language models (LLMs).

  • Query Suggestions (QS): Recommends relevant queries to users as they type in the search box, based on past successful queries.

  • Relevance Generative Answering (RGA): Uses embeddings and vector search to retrieve the most relevant information from your indexed content and generates a grounded answer to a query.

  • Semantic Encoder (SE): Uses vector search to retrieve items from your indexed content based on semantic similarity with the query, improving relevance for complex queries and feeding results to RGA and CPR for answer generation and passage retrieval.

  • Smart Snippets: Displays a snippet of the most relevant search result item directly on the results page.

Model learning, processing, and refreshing

Depending on the model type, Coveo ML models either learn from search behavior patterns and historical case data, or create embeddings from your indexed content, to surface the results, content, and generated answers most helpful to users.

  • Usage-based models: Train on Coveo Analytics data to predict and recommend which content is most helpful to users.

    A usage-based model is trained on Coveo Analytics data from a specific period and is regularly retrained using the latest data to learn from recent behavior and keep the model up-to-date.

    In general, the more data available for training, the more relevant the recommendations. As a guideline, a Coveo Analytics dataset containing at least 10,000 queries typically provides enough data for a Coveo ML model to generate highly relevant recommendations.

    The usage-based models are:

    • Automatic Relevance Tuning (ART)

    • Content Recommendations (CR)

    • Dynamic Navigation Experience (DNE)

    • IPX Recommendations (IPXRECS)

    • Query Suggestions (QS)

  • Content-based models: Leverage large language models (LLMs) to create embeddings or snippets from your indexed content to improve the relevance of content retrieval and answer generation.

    Content-based models don’t train on usage data, so their freshness comes from regular rebuilds to reprocess your indexed content to ensure the most up-to-date content is available.

    The content-based models are:

    • Passage Retrieval (CPR)

    • Relevance Generative Answering (RGA)

    • Semantic Encoder (SE)

    • Smart Snippets

  • Case Classification (CC) model: Learns from the historical case data from your indexed content to predict field values for new cases. Regular rebuilds ensure the model stays up-to-date with the latest case data.

Note

A model can take anywhere from several minutes to several hours to build, depending on the size of the model dataset and the number of model build requests in the queue (system load).

Beyond machine learning models

In addition to its machine learning models, Coveo offers AI-powered capabilities that build on those models to deliver conversational and agent-driven experiences.

Coveo Search Agent

The Coveo Search Agent adds an agentic, conversational search experience with generative answering to your Coveo-powered search interface. Using agentic AI capabilities such as reasoning and decision-making, it orchestrates multiple rounds of content retrieval and answer generation based on follow-up questions, enabling users to find answers to complex questions on their own.

It builds on content-based models, such as Passage Retrieval (CPR) and Semantic Encoder (SE), to ground its answers in your indexed content.

Hosted MCP Server

The Coveo Hosted Model Context Protocol (MCP) Server enables large language model (LLM) apps and AI agent frameworks to access Coveo’s enterprise search and retrieval capabilities through standardized Coveo tools, such as Search, Fetch, Answer, and Passage Retrieval.

It acts as an intermediary between LLM apps and Coveo’s APIs, so these apps and frameworks can obtain secure, relevant, and context-aware responses grounded in your enterprise data without the need for custom integrations.

Conversational Product Discovery

Conversational Product Discovery is a search-native conversational experience for commerce that blends products, content, and concise explanations. It lets shoppers describe what they need in their own words and receive curated product results directly within your existing search experience, then refine, compare, and bundle those products through follow-up queries.

The experience is powered by AI agents grounded in your catalog data, constrained by enterprise-defined layouts and guardrails, so conversation and search stay unified without fragmenting the shopper journey.

Note

For Coveo Machine Learning (Coveo ML) models specific to Coveo for Commerce, see Machine learning for Commerce.