Skip to content
The Large Behavioral Model
.md

The Large Behavioral Model

The Large Behavioral Model (LBM) is Intuizi’s flagship large quantitative model (LQM), trained on de-identified, real-world behavioral signals. Every audience you build or size, in the console, with the API, through the MCP server, or with the CLI, is a question put to this model.

An LQM, not an LLM

A large language model (LLM) is trained on words: it reads and writes text. A large quantitative model (LQM) is trained on numbers: it finds the relationships between different kinds of data and answers quantitative questions about them. How many devices visited a coffee shop last week? Which of them came back three times or more?

The two work well together. An AI agent connected to the MCP server takes your question in plain words and puts it to the LBM as API calls.

What it learns from

The LBM learns from de-identified signals of real-world behavior, such as the places devices visit, the apps they use, and what they watch on connected TV. Every signal comes from a permissioned partner and is de-identified before the model sees it. The model keeps updating as new signals arrive, and each audience is built from the date window you choose.

Each kind of signal is a dataset. An audience is built from one dataset, or two combined, and filtered with ids from the reference catalogs.

How you ask it

Your questionAPICLIMCP tool
How many devices match?Estimate Audience Sizeaudiences estimate createestimate_audience_size
Which devices are they?Create Audienceaudiences createcreate_audience
Who else behaves like them?Create Lookalike Audienceaudiences lookalike createcreate_lookalike_audience
How many were seen three times or more?Preview Activationactivations previewpreview_activation
Deliver them to a partnerCreate Activationactivations createcreate_activation

Lookalike Models, and the frequency analysis a preview reads, require additional permissions which need to be approved by your Account Manager.

Most answers take a while: a create returns at once, and the model finishes the work in the background. The Async Model covers how to wait for it.