Skip to main content

Test AI with internal teams before customers see it.

Customer Zero means running data pipelines and AI workflows inside your own operations before offering them to customers. Internal teams use them first and find the organisational problems that model testing misses.

Why internal-first

Enterprise GenAI programs usually fail when company incentives reject the workflow, even if the model performs well. People embrace tools that make them look capable, hide tools that make them feel replaceable, and rarely volunteer a saving that costs their own team.

Synthetic test environments show whether a model produces the right tokens. They cannot show whether the workflow around the model produces the right outcome on real data, with real permissions and real accountability. That takes an internal production deployment.

The three stages

Each stage produces something the next stage uses, and each can be audited on its own.

  1. Integrate

    Data governance

    Connect systems of record, isolate credentials in Google Secret Manager, and select verified business entities. The result is a catalogue of source instances, each with a credential pointer and an active-subscription grant. No plaintext secret is stored.

    • c0_raw_feeds.<source>_<instance>
    • credential_ref: projects/…/versions/N
  2. Distribute

    Internal production testing

    Scope access to active instances, enforce append-only audit trails, and run real team workflows against the platform in live operations. The result is a stress-tested grant set, proven by what the audit log shows in daily use.

    • grant: user × instance × client
    • AuditEvent: mcp.query.executed
  3. Serve

    Customer rollout

    Connect AI clients (Claude, ChatGPT, Gemini Enterprise) to the Snowflake MCP endpoint. The result is a per-instance, per-client grant set that customers can audit, rotate, and revoke without involving the vendor.

    • Snowflake MCP endpoint
    • c0_analytics.fct_campaign_performance

How the framework becomes the platform

Customer Zero started as IKMJ's internal system for running AI across its own business. The platform packages those same controls: instance grants, credential isolation, and audit logs, exposed through a Snowflake Model Context Protocol (MCP) endpoint that any supported AI client can query.

Integrate and Distribute produce IKMJ's instance catalogue, grants, and audit log. Serve adds the customer-facing MCP endpoint and the per-tenant controls operators use to grant, scope, rotate, and revoke access.

Customers can set up the same controls for their own operations, or start with Serve.

Who runs it

Customer Zero is built and operated by IKMJ (Ikigai & Majime Pty Ltd), an active Australian private company. IKMJ has written more about the framework and the team behind it on its website.

See how Customer Zero implements the framework.