Glossary

Every Neuralift term in one place, from action to workspace.

A reference for the vocabulary used throughout Neuralift and this guide, in alphabetical order. Across the docs, the first mention of a term on each page carries a hover tooltip that links back to its entry here.

Action

A strategic recommendation generated for a run: a ranked play composed from segment tactics, led by a computed value headline, with one combined audience and one measurement plan. See Actions.

Activation

Putting segments to work in the systems where you reach customers: joining segment-labeled results to your own data, building campaign audiences, and feeding your channel tools. The exporting overview lays out the routes.

Activation URL

A one-time URL generated when open-protocol Delta Share delivery is configured. You use it to register the share in Snowflake or another open-protocol reader; it is shown exactly once and cannot be retrieved later. See Delivering results.

Admin

The organization role that manages setup: data connections, members and invitations, organization settings, and result delivery. See Roles & permissions for the full capability matrix.

AI Management System (AIMS)

Neuralift’s ISO/IEC 42001-certified management system governing how AI is built and operated, covering AI risk management, system impact assessment, performance monitoring, and continual improvement. Certification materials are available in the Trust Center.

Angle

The one-sentence statement of a segment tactic’s version of an action: the audience slice that segment contributes, what separates its variant from the sibling segments’ variants, and the concrete mechanic. See Actions.

API key

An org-scoped secret that authenticates an AI assistant to Neuralift’s MCP endpoint. Shown once at creation, so treat it like a password. See Managing API keys.

Clone

A copy of an existing use case. The definition always carries over (goal, context, KPIs, segment size, configuration), with optional toggles to copy the input data and the segment results, so you can iterate without touching the original. See Cloning a use case.

Cloud-storage push

A Preview delivery method that copies a published run’s segment-labeled Parquet files to your own S3 bucket, authorized by a bucket policy you add for Neuralift’s delivery role. See Delivering results.

Custom connector

The claude.ai and Claude Desktop feature for attaching a remote MCP server to Claude by URL from the Connectors settings. See Connecting Claude.

Customer 360

The prepared table segmentation runs on: one row per customer, with cleaned, model-ready columns describing each customer. Also called a customer 360 table; see How your data is prepared.

Data connection

A link between your data warehouse (Databricks, Snowflake, BigQuery, or Redshift) and Neuralift, set up by an organization admin under Settings → Data Connections, that lets Neuralift read input data for segmentation runs. See the data connections overview.

Data dictionary

A catalog of every column in a prepared dataset, classified as a KPI, continuous, categorical, or ID field, with distribution stats and annotations. See The data dictionary.

Data preparation

The step that turns synced source tables into a prepared dataset and its data dictionary. Neuralift runs it as part of your engagement; see How your data is prepared.

Data run

The running form of a use case: one execution of the segmentation model against your prepared dataset, producing segments, insights, and actions. See The run lifecycle.

Defining features

The columns that most distinguish a segment from everyone else: the model’s explanation of what makes the segment distinct. Each segment’s detail panel lists its top defining features; see Inside a segment.

Delivery

Pushing segment-labeled results back into your own data platform via Delta Share, cloud storage, or sFTP, so you can activate them where your data lives. See Delivering results.

An open protocol for secure data sharing between platforms. Neuralift uses it both to read your source tables and to deliver results back to you. See Connecting Databricks.

Export

A downloadable file of segment-labeled results: a CSV of customers with their segment assignments. See the exporting overview.

Feature attribution

An explainability output that quantifies how much each feature contributes to a segment or an individual assignment. Every Neuralift segment ships with cross-checked feature-level attributions. See How the segmenter works.

Feature engineering

Turning raw history (orders, sessions, transactions) into per-customer columns that describe behavior, such as windowed spend, recency, cadence, and diversity features. See Feature engineering examples.

Generative orchestration

The Copilot Studio agent mode that decides at runtime which tools to call. Microsoft requires it to be enabled before an agent can use MCP tools. See Connecting Microsoft Copilot.

Insight

An auto-generated observation about a segment: a notable pattern in its behavior or KPIs, grounded in the segment’s data. See Inside a segment.

Invitation

An email offer to join an organization, sent by an admin from Settings → Invites and accepted or rejected from the invitee’s Invitations page. See Inviting people.

KPI

A key performance indicator: a column you care about moving, such as revenue or visits. Segments are profiled and compared against your KPIs. You rank them in the use case definition.

KPI lens

A view of the segment landscape recolored by one KPI, showing how that metric’s values distribute inside every segment so you can spot where it concentrates. See KPI lenses & lift.

Lift

How much a segment over- or under-indexes on a KPI compared to the overall population. It is the quickest read on where the value sits. A segment with 3× revenue lift contributes three times its population share of revenue.

Lift potential

An estimated value a segment or run could yield under a what-if scenario where a share of customers is lifted by a set number of incremental events. Currently a Preview capability. See KPI lenses & lift.

MCP

The Model Context Protocol: an open standard that lets AI assistants like Claude query your published Neuralift segments directly, read-only, using an org-scoped API key. Currently in Preview; see the MCP overview.

MCP server

The secure, read-only Neuralift endpoint that AI assistants connect to. It returns only your organization’s published results, authorized by an API key. See the MCP overview.

Measurement plan

The experiment an action commits to before you run it: a design (typically a holdout percentage), the success metric the play is accountable to, and the readout window. See Actions.

Member

The organization role for consumers of results: members explore published use cases, segments, and actions, and export what they need. See Roles & permissions.

Membership

A user’s place in an organization, carrying their role of admin or member. Managed under Settings → Users; see Members & roles.

Organization

The workspace where your team collaborates. It holds your data connections, use cases, members, and delivery settings. See the organizations overview.

Persona

A plain-English portrait of a segment’s typical customer, generated from the segment’s defining features and KPI profile. See Inside a segment.

Prepared dataset

Your source data after preparation: one row per customer ID with cleaned, numeric, model-ready columns and no PII. Every run starts from a prepared dataset; see How your data is prepared.

Preview

A release stage: the capability is functional and supported, but not yet enabled for every organization and may still change. Contact Neuralift to enable Preview features. See Release stages.

Publishing

Making a completed run visible to every member of your organization. Runs start private; publishing flips the switch, unpublishing hides them again. Unpublishing deletes nothing: the run returns to unpublished and can be published again later; see Publishing & unpublishing.

Reach

The audience size of an action or of one contributing segment tactic: the number of customers it targets, shown with its percentage of the run’s dataset rows. See Actions.

Segment

A distinct group of customers discovered during a run, with its size, KPI profile, insights, persona, and defining features. Start with the segments overview.

Segment landscape

The visual map of all segments in a run, positioned and sized by population, with KPI lenses recoloring the view; position doesn’t encode lift. See The segment landscape.

Segment tactic

An evidence-backed recommendation for a single segment, derived from its insights: the activation atom actions are composed from, contributing one segment’s audience slice, angle, and evidence to a play. See Segment tactics.

Segment-labeled data

A run’s results in row form: your prepared dataset with each customer’s segment assignment (segment ID and name) attached. It is the payload of CSV exports and warehouse delivery.

Neuralift Segmentation Engine

Neuralift’s patented segmentation model (US 12,321,960 B2), also called the segmenter in Neuralift’s governance documents: it groups a client’s records into distinct, meaningful segments from that client’s first-party data and produces feature attributions explaining what defines each segment. See How the segmenter works.

sFTP gateway

A private, per-organization file gateway provisioned by Neuralift on request. You authenticate with an SSH key pair; no password is stored. Results are delivered as Parquet files under outbound/, and Parquet datasets dropped under inbound/ can be copied in as source tables. See Delivering results and Syncing source tables.

Neuralift’s sharing identifier, copyable from the Add data connection dialog, that you use to add Neuralift as a recipient of your Databricks Delta Share. See Connecting Databricks.

SOC 2 Type II

An independent audit attestation that Neuralift’s security controls operate effectively over time. Reports are available via the Neuralift Trust Center.

Source table

A table copied into Neuralift from a data connection, or from the sFTP gateway’s inbound folder. Source tables are snapshots, refreshed on demand with Copy again, and are the inputs data preparation turns into run-ready datasets. See Syncing source tables.

Tier chips

The plain-language chips summarizing an action’s standing — Impact, Effort, Confidence, and Margin — each with a one-line reason behind it. Impact and Effort are relative to the run’s other actions. See Actions.

Transient processing

Neuralift’s handling model for raw client data during segmentation: data is processed in the pipeline, not durably retained beyond producing results, and intermediate working data is ephemeral. See Data handling & security.

Trust Center

Neuralift’s security and compliance portal at trust.neuralift.ai, hosting the SOC 2 Type II report, security policies, and ISO/IEC 42001 materials.

Unclassified

Customers a run could not confidently place into any segment. They are counted separately rather than forced into a poor-fitting group. See The segment landscape.

Use case

A single segmentation project: a business question you want to answer with segments, pointed at a prepared dataset. “Which customers are about to churn?” is a use case; see Creating a use case.

Use case definition

A use case’s goal, desired segment count, business context, and prioritized KPIs, captured on the Definition tab (called the brief in some earlier materials). Neuralift uses it to frame the engagement and to shape how results are named, explained, and prioritized. See Defining the use case.

Value headline

The computed figure an action leads with: what the play could be worth, in dollars, customers, a rate move, or KPI units, derived from measured audience counts and stated assumptions. Assumption-based figures carry an estimate badge. See Actions.

Warehouse delivery

Pushing a published run’s segment-labeled results into your own data platform (a shared table via Delta Share, or Parquet files via cloud storage or sFTP) so you can join and activate them where your data lives. See Delivering results.

Workspace

Your company’s account with Neuralift, sitting above your organizations. A workspace can contain one or more organizations, and its name appears in your organization’s URL. See the organizations overview.