A NetSuite AI-readiness review that shows whether an assistant will answer reliably from your account

#The situation

NetSuite can now be connected to an AI assistant through Oracle's AI connector. The assistant can only be as good as what it reads. Free-text fields that should be lists, custom fields with no help text, several fields that mean the same thing, and roles that expose too much or too little all turn into confident wrong answers. Companies want to know where they stand before they connect an assistant.

#What we built

A review method with its own tooling, run the same way for every account.

• Twelve areas. The areas sit under six pillars (legibility, completeness, semantics, access, volume and serviceability). They cover the account profile, schema legibility, entities, items, transactions, financial structure, automation and integrations, the reporting layer, unstructured text, access and governance, volume and history, and readiness for the AI connector. Each area has checks, a scoring rubric and recommendation patterns.
• More than 120 read-only probes. Each probe is a SuiteQL or metadata question, grouped by area, with a line explaining how to read its answer. We validated the probe library against two live accounts.
• A customization snapshot. An SDF object import captures the account's configuration, and an analyser reports on help text, workflows, roles, scripts, and which custom fields saved searches depend on.
• Four questions every review answers. Which free-text fields should be lists, which lists should be segments, which fields are no longer needed, and which custom tools the connector will need.
• Delivery tooling. A report builder produces the Word deliverable, and a scaffolder creates a self-contained review workspace for each customer.

#What it does now

This is a new offering. The method and the probe library are complete and validated, and the first engagement is under way. The review can run under the customer's own AI subscription, on their own machine, because the workspace carries its own instructions, method and tools. Every probe is a read, and the review changes nothing in the account.

#Under the hood

• Evidence comes from two sources, live read-only SuiteQL for data and SDF import for configuration and metadata.
• Probes are plain SQL, JSON or JavaScript files, so they can be read and reviewed before they run.
• Each area is scored by rubric, so two reviews of the same account come out the same.
• Credentials stay with the customer and out of version control.