NetSuite AI readiness

Make NetSuite AI answer correctly.

At SuiteWorld 2026, Oracle presented AI as a new way into NetSuite: Ask Oracle, the AI Connector, AI Descriptions, SuiteSkills and SuiteAgents. In the technical sessions, Oracle was also clear that the results depend on your data descriptions, your permissions and your setup.

That setup is what we do. We get NetSuite AI ready, review what it can see and do, and test it against your real questions.

The new features

What the new NetSuite AI features change.

What Oracle presented, and what each feature does.

Diagram, The road to AI-ready ERP: switching AI on is the on‑ramp, not the destination; the road to reliable daily use is held up by five supports: business meaning, permissions, tested tools, repeat testing and a named owner for each workflow.
Ask OracleQuestions in plain languageYour team asks NetSuite questions in their own words and gets answers inside NetSuite. Attach a document and it can prepare a draft transaction too.
AI ConnectorConnect outside AI assistantsClaude, ChatGPT and other assistants can work with your NetSuite data through Oracle’s AI Connector, using the permissions of the role they connect as.
AI DescriptionsYour business terms, written downShort notes on your records and fields tell the AI what they mean in your business, so answers use your terms.
SuiteSkillsProcedures the AI can followA skill teaches the AI a task step by step, and which tools to use for it.
SuiteAgentsRoutine tasks, automatedAn agent handles a specific task when something triggers it or on a schedule, once an administrator turns it on.
Custom ToolsYour scripts as AI actionsYour existing scripts and calculations become actions the AI can call, under the user’s own role.

Announced vs. out of the box

What was announced, and what it takes in practice.

Keynote demos show what is possible. The technical sessions described the setup behind it, and several of the newest features need NetSuite Next.

Diagram, Same question, two answers: without an AI Description the AI sees only the field name and ID and misses the customers; with a short description it names the three customers to follow up. Both runs are checked against an agreed reference answer.
AI can answer questions about custom records.What it takes in practice: Clear descriptions help it understand what a record is for. In an Oracle session, a follow-up question about a custom record failed until short descriptions were added to the record and its fields.
AI follows NetSuite roles.What it takes in practice: The role of the person asking, and the run-as role chosen for scheduled agents, must match the access you intend. The AI Connector does not support Administrator roles, so outside assistants connect with a non-Administrator role.
Agents automate defined tasks.What it takes in practice: An administrator must activate each agent, grant access and set up its triggers before it does anything. At SuiteWorld, the tools agents get out of the box could read but not create records; an agent that needs to write uses a custom tool. Creating draft transactions from a document is an Ask Oracle feature, shown below.
Skills and tools extend what AI can do.What it takes in practice: Each tool declares the permissions it needs, and skills keep the toolset organized. Oracle cautioned that too many tools can confuse the model.
AI Descriptions add business context.What it takes in practice: Descriptions are short, up to 280 characters, so each one must be specific and tested against real questions. Descriptions are not used until they pass Oracle’s validation, and until then the AI sees only names and IDs.
Agents can run on a schedule.What it takes in practice: Scheduled runs need a deliberately chosen run-as role, supervision and evaluation. Ask Oracle and SuiteAgents run on NetSuite’s built-in AI and use AI units, with no per-agent cap today. Assistants connected over MCP run on your own AI model, so they do not use AI units.
Agent performance can be evaluated.What it takes in practice: NetSuite has no built-in evaluation tool yet. Oracle’s advice is to run the same questions again and again against answers you know are right, because the output isn’t deterministic. Plan for extensive testing and the AI units those tests consume.
Drafts still need a person to review them.What it takes in practice: A person reviews and completes the draft in the normal form, and your business rules and approval workflows still apply.

How we help you prepare

We set it up, review it and test it against your real questions.

1AI readiness checkWe take your most-used records and fields and a short list of real business questions, test them in Ask Oracle and the AI Connector, and show you where the answers fail and why.
2Descriptions and metadataWe write and test short AI Descriptions for the records and fields behind your top questions, and keep them under version control.
3Read-only access designWe review which roles the AI can act as, what each role can reach, and what is logged. We design AI access to read, then test that the AI identity cannot change records within the scope we checked.
4Safe writes through drafts NetSuite NextAttach a customer PO, a vendor bill or a journal spreadsheet to Ask Oracle and ask for the transaction. It saves a draft under your own NetSuite role, and you review it in the normal form and submit. Best experience in the NetSuite Next interface.
5Custom skills and toolsWe build the few skills and tools your team needs, scoped to roles, and keep the toolset small. Custom tools reach outside AI assistants through a custom connector URL, not the standard one; we set that up.
6Agents tested for consistencyWe write test questions with known right answers for your agents and skills, plus trap cases they should never fall for, like answering outside their role or inventing a figure. We run both on every change and tune the instructions, skills and tools until the answers match far more consistently. AI output is never fully deterministic, so we measure the match rate and show it to you instead of promising perfection. We do one narrow agent at a time, each with a deliberately chosen run-as role.
Diagram, NetSuite Next: prepare a draft, then review and submit. All three steps run under your own NetSuite role: AI reads what it needs with your access; in Ask Oracle you attach a document such as a customer PO, a vendor bill or an Excel journal and ask for the transaction, and it saves a draft; you review it in the normal form and submit, and your approval rules and scripts still apply. Best experience in the NetSuite Next interface.

Our SuiteApps

Clean data for AI to work from.

Our SuiteApps already hold much of what AI needs to answer correctly: verified schema, governed definitions and exact cost data.

They already work with NetSuite AI. Our SuiteApps ship as NetSuite custom tools: 17 tools across CraftQL, Lexicon, CostLens and ReportBridge, available in Ask Oracle and to your AI assistant over MCP, running under each user’s own role.

CraftQLAI tools: 5Which fields each role can readCraftQL knows which tables and fields exist in your account and which ones each role can actually read. That is the starting point for descriptions and access design.“Which fields on vendor bills can the AP role read?”Explore CraftQL ↗LexiconAI tools: 8Agreed definitions, with ownersLexicon gives each report and metric an owner and a plain definition. We write descriptions and skills from those definitions.“What does net margin mean here? Run last quarter’s figure.”Explore Lexicon ↗CostLensAI tools: 2Exact cost and marginCostLens rebuilds landed cost and inventory age from the general ledger, so questions about margin get exact figures with the detail behind them.“Which SKUs have stock older than 365 days, and at what cost?”Explore CostLens ↗
ReportBridgeAI tools: 2Ask for your reports by nameReportBridge brings your existing NetSuite reports to AI assistants over MCP, so a question comes back with the report’s own figures. Available now.“Give me the September balance sheet.”
Diagram, The answer needs a definition: one question about last quarter’s margin has four defensible meanings, each giving a different illustrative number, until a person approves one meaning and it becomes one agreed calculation, kept in YTBE Lexicon. Figures are illustrative.

Common questions

Before you turn it on.

Diagram, One workflow through the test gate: one finance workflow is checked for the right answer, respected roles, the same result on repeat and owner sign-off, and comes out as use, fix or defer. A defer is a result.
Can the AI change our data?

It depends on the roles and tools you allow. We design AI access to read, then test that the AI identity cannot change records within the scope we checked. Where AI should create something, it prepares a draft under your own NetSuite role, and you review and submit it.

Do we need NetSuite Next?

Not for everything. Some features work today, and several of the newest were presented as part of NetSuite Next. A readiness check tells you which apply to your account.

Where do we start?

With your questions. Bring three to five things your team needs to know from NetSuite, and we will test them, show where the answers fall short, and fix the setup behind them.

Get ready for NetSuite Next

Switching it on?

NetSuite Next is rolling out now. Here is how to turn it on safely.

Prepare for NetSuite Next ↗

Is your NetSuite ready for AI?

Bring your real questions. We'll show you where the answers fall short and what it takes to fix them.

Talk to us