How Neuralift works

Neuralift sits alongside your existing data platform. It securely reads the prepared first-party tables you choose to share, discovers the distinct segments that latently exist in that data, explains what defines each one, and delivers the results back to your stack, with segmentation processing kept transient (see Data handling & security). This page walks the whole path once, so every other section of the docs has a place on the map.

The journey at a glance

Each stage below links to the section of the docs that covers it in depth. Who performs each stage (Neuralift’s team or people in your organization) is laid out in Roles & permissions.

Your data platform stays the source of truth

Neuralift connects to your warehouse or lakehouse through a secure connection that your organization configures and controls. The connection is used for metadata and governed reads of the source tables you choose to sync. You decide exactly which tables Neuralift can read, and access is revocable on your side at any time.

Start with the data connections overview, then see syncing source tables for how tables become available to a use case.

Data preparation

Before segmentation, the source data is shaped into a single analysis-ready table with one row per customer: typically behavioral and transactional events plus enrichment, formulas, and measures about each customer. The table is profiled into a data dictionary that classifies every column, and the model operates on de-identified, normalized numerical features. Neuralift does not accept any data with PII.

The data preparation overview explains the process, and data requirements lists what makes a table ready.

The segmentation model

The prepared table feeds the Neuralift Segmentation Engine, the patented deep learning model at the core of the platform, covered by granted patent US 12,321,960 B2. It groups your customers into distinct, meaningful segments and produces feature attributions that explain what defines each segment. Every client’s model is tailored to that client, learned from that client’s data alone, and refit as new data arrives.

For a module-by-module view of the method, read Inside the Neuralift Segmentation Engine.

Results in the Neuralift app

Finished runs surface in the app as a landscape of segments, each with the features that define it, its KPI profile, plain-English insights, and recommended actions broken into action plans. Runs are reviewed and then published so your whole team can explore them.

See the segments overview for exploring results, and publishing for how results become visible to your organization.

Export and delivery

Insight only matters once it reaches the systems where you act on it. From a published run you can export segment-labeled results and have them delivered back to your data platform, closing the loop between discovery and activation.

The exporting overview covers the formats and paths, and delivering results covers getting them back into your stack.

AI assistants via MCP (Preview)

Optionally, your organization can connect AI assistants to Neuralift over MCP (Preview), so you can ask questions about your published segments from the tools you already work in. See the MCP overview and release stages for what Preview means.

Where your data lives

Data is encrypted in transit and at rest, segmentation processing is transient, one client’s data is never used to train another client’s model, and production runs in the United States (AWS US West, Oregon) under Neuralift’s ISO/IEC 42001-certified AI Management System. The full picture, including what is retained and for how long, is in Data handling & security.

Keep reading

• Inside the Neuralift Segmentation Engine: the method, module by module.
• Data handling & security: the commitments behind the flow above.
• Core concepts: the vocabulary the rest of the docs build on.