Frequently asked questions

Quick answers to common questions about using Neuralift. For platform, security, and commercial questions, see the main site’s FAQs; for anything not answered here, see Getting help.

Getting started

Who runs segmentation, us or Neuralift?

Neuralift’s data science team creates use cases, prepares your data, starts runs, and publishes results, working with your organization at each step. Your side owns the org itself: admins manage data connections, members and invites, and delivery configuration, while members explore published results and export them. The full division of labor is in Roles & permissions, and How Neuralift works walks the end-to-end journey.

Why can’t I see any use cases?

An empty dashboard usually means nothing has been published to your organization yet. Members see only published use cases, so runs still in preparation, running, or under review are invisible to you by design. If a colleague can see results you can’t, confirm you’re signed in to the right organization (you may belong to more than one) and that you’re a member of it. See Roles & permissions for exactly what each role sees.

Why is the Segments tab grayed out?

The Segments button, along with the Discovery and Actions items in a use case’s sidebar, unlocks once a run has produced results (Discovery needs the run’s segments, Actions its generated actions). If they’re disabled, the use case hasn’t completed a run yet: check its status badge on the dashboard. A pending use case is still being defined and prepared, and a running one is in progress. See The run lifecycle for what each status means.

Who do we contact for help, and is there a status page?

Email support@neuralift.ai for anything: technical problems, but also questions about your data, your use case definition, or how to read a set of results, since the same team that runs your use cases answers them. Current availability and incident history are on the status page at status.neuralift.ai, and security and compliance documentation lives in the Trust Center. What to include in a report is covered in Getting help.

Data & connections

What data do you need from us, and can we send raw event history?

Neuralift works from first-party tabular data about the entities you want to segment (behavioral, transactional, and derived attributes) with a unique pseudonymous ID column and the KPI columns you care about, and no PII. You can provide entity-level tables that are already one row per customer, or raw transaction and event history: Neuralift aggregates raw history up to customer level during preparation. The full checklist is in Data requirements, and Feature engineering examples shows what aggregation adds.

What formats can I send data in?

The usual path is a warehouse table synced through a data connection: Databricks and Snowflake are generally available, with BigQuery and Redshift in Preview. When a warehouse connection isn’t the right fit, Parquet, CSV, and Delta files are also accepted via file hand-off, and organizations with an sFTP gateway can drop Parquet datasets into its inbound/ folder; see Syncing source tables. Whatever the format, the table itself should meet the checklist in Data requirements, and remember that Neuralift does not accept any data with PII.

How does Neuralift connect to Databricks without taking our credentials?

Databricks connects over Delta Share: you create a share in your own workspace, add Neuralift as a recipient using a Sharing ID copied from the app, and Neuralift reads only the tables you put in the share. No usernames, passwords, or tokens are handed over or stored. Sharing happens at the governance layer of your lakehouse, so you stay in control of exactly what’s shared and can revoke access on your side at any time. The full setup is in Connecting Databricks.

Is my data used to train other clients’ models?

No, never. Each client’s model is trained on that client’s data alone, and Neuralift never uses one client’s data to train another client’s model or any model outside that client’s engagement. The full set of commitments, including encryption, transient processing, and retention, is in Data handling & security.

Where does processing run, and is our data encrypted?

Production runs in the United States, in AWS US West (Oregon). Data is encrypted in transit and at rest, segmentation processing is transient: intermediate working data is ephemeral, and raw data is not durably retained beyond what’s needed to produce results. The platform operates under an ISO/IEC 42001-certified AI Management System with SOC 2 Type II attestation; see Data handling & security and the Trust Center.

Runs & results

Why can’t I start a run or publish?

Neuralift’s data science team creates use cases, starts runs, and publishes results, working with your organization’s admins at each step, so there is no start or publish button for customers. Your side of the work is the definition and the data; once both are confirmed, Neuralift starts the run and you follow its status on the dashboard. The full split is in Roles & permissions and Starting a run.

How long does a run take?

It depends on the size of your prepared dataset and the run’s configuration, so there’s no single number. The use case’s status updates in real time on the dashboard, so you don’t need to keep the page open, and Neuralift monitors every run, so if anything goes wrong it’s investigated and re-run without you doing anything. If a run seems to be taking longer than expected, ask your Neuralift contact for a read on where it is.

How many segments will we get?

That’s set by the Desired Segments size in your use case definition: Small aims for 4–12 segments, Medium for 12–20, Large for 20–40, and Extra Large for 40+. New use cases default to Small, and starting smaller is usually right: you can clone the use case and rerun at a finer size once you’ve seen the first landscape. The trade-offs between sizes are laid out in Defining the use case.

What’s the difference between an insight, a segment tactic, and an action?

They sit at increasing levels of “so what”. An insight is a confidence-ranked observation about one segment, something worth knowing about why the group coheres (see Inside a segment). A segment tactic is a concrete, evidence-backed recommendation for that segment, scored on Evidence Strength, Expected Impact, and Executability (see Segment tactics). An action is a run-level strategic play built on those tactics, broken into action plans — each owning a slice of the audience with a one-sentence angle and its own reach (see Actions & action plans).

Exports & delivery

How do I get results into my warehouse?

An admin configures a delivery destination once for your organization under Settings → Data Connections; after that, delivering any published run is one click from its Exports tab. Results arrive as a shared table over Delta Share, or as Parquet files via cloud-storage push (Preview) or sFTP (Preview). The full setup for each method is in Delivering results to your warehouse.

Can we download results as files instead?

Yes. From a published use case you can create CSV exports of one or more segments’ labeled rows, ready to load into an email or ad platform as an audience. Exports and delivery are available on contracted plans; the exporting overview compares both routes.

MCP & Preview features

What’s a Preview feature?

A Preview capability is functional and supported, but not yet switched on for every organization, and details may still change based on feedback. Preview capabilities are enabled per organization; contact your Neuralift representative or support@neuralift.ai to request one. Pages documenting a Preview capability carry a badge and banner; the full definitions, including Beta and GA, are in Release stages.

How do I get MCP?

MCP is in Preview and is available by request and under agreement only. Contact your Neuralift representative or support@neuralift.ai to request it. Once enabled, your organization’s admins see an API keys tab in Settings and can connect Claude, OpenAI, or Microsoft Copilot. Start with the MCP overview.

Which AI assistants work with MCP?

Claude is the most-verified path (Claude Code is fully verified, with Claude Desktop and claude.ai documented), followed by OpenAI’s Codex CLI and ChatGPT, and Microsoft Copilot Studio. Each has its own guide with authentication caveats: Connect Claude, Connect OpenAI, and Connect Microsoft Copilot. The support matrix is in the MCP overview.

How do we rotate or revoke an MCP API key?

In Settings → API keys, find the key by name and click Revoke. It takes effect immediately, and the revoked key stays listed as an audit record. Because the full key is shown only once at creation, rotation means revoking the old key, creating a replacement, and updating the client that used it; keys also expire 365 days after creation by default. Revoke immediately if a key may have been exposed or its holder has left your organization; details are in Managing API keys.