Major

NetSuite AI Agents: APIs, Use Cases and Governance

Explore NetSuite's native AI, eight verified SuiteTalk endpoints, and finance automation on Major with custom API connections, durable review queues, and human approval.

Ritvij GautamUpdated
Major x NetSuite logo lockup

NetSuite AI agents can investigate ERP exceptions, propose coding or customer matches, and prepare finance briefs with human control over consequential changes. This guide covers NetSuite's native AI, verified SuiteTalk endpoints, and how Major combines judgment with deterministic apps, durable review queues, and governed workflows across finance systems.

What NetSuite does for finance and operations

NetSuite holds the financial and operational records that controllers, ecommerce finance teams, and production planners need to keep consistent. A customer identity affects order matching; an item mapping affects fulfillment; a transaction's classification affects reporting. Those connections make ERP automation more demanding than copying fields between systems.

Oracle's SuiteTalk overview describes record operations, account-specific metadata, and query services. Finance teams use those capabilities to retrieve transaction context, maintain customer data, and assemble reporting inputs. Operations teams use record relationships to investigate handoffs and missing information.

Major supplies the software around that work. An agent can build a deterministic app that stores source-to-ERP mappings, checks totals, and remembers unresolved exceptions. The agent then runs that software and reserves judgment for ambiguous descriptions, conflicting evidence, and prioritization.

For a controller, the useful outcome is a queue showing which expense needs review and why. For an ecommerce finance owner, it is an order mapping with an attributable approval, not another conversation someone must reconstruct during close.

NetSuite's built-in AI and where Major fits

NetSuite already includes substantive AI. Oracle's AI feature catalog lists Text Enhance, Prompt Studio, Intelligent Close Manager, transaction-matching capabilities, and other features. Availability varies; verify the features enabled in your account rather than assuming every listed capability is included.

Text Enhance generates, refines, changes the length of, and translates business text using information on the record being edited. It fits tasks such as drafting an item's sales description from existing item information.

Exception Management uses organization-specific machine-learning models to identify unusual amounts, accounts, vendor changes, and expected transactions. Resolving an exception updates its review status; it does not automatically correct the underlying transaction. Availability depends on data volume and model readiness.

The October 2026 release notes also describe Ask Oracle drafting and previewing saved searches in NetSuite Next, whose rollout is phased.

Major adds cross-system execution and durable state: compare ERP evidence with a commerce order, preserve a controller's decision, and resume the approved process tomorrow. Native NetSuite AI remains valuable inside the ERP; Major coordinates the broader business process.

How NetSuite connects to Major

NetSuite has no native or Pipedream connector in the catalog: it connects as a custom API connector, with base URL and authentication configured once by IT. Major's catalog identifies NetSuite without either connector binding, so this is not a claim of a prebuilt NetSuite integration.

Oracle documents the account-specific REST URL structure as https://<accountID>.suitetalk.api.netsuite.com/services/rest. IT should copy the correct service URL from the account's Company URLs page and configure the custom connector against that environment.

OAuth 2.0 is Oracle's preferred authentication method; token-based authentication is also supported. Configure only the operations and record access the process needs. Keep credentials centrally in Major's connector vault, never in application code, prompts, or copied review notes.

Major orchestrates NetSuite; it never replaces it. NetSuite remains the authoritative ERP. Major's app stores orchestration state, evidence, and approvals, while SSO, RBAC, and audit logs govern the apps, agents, and workflows using the connector.

NetSuite API guide for apps, agents, and workflows

Authentication and operating constraints

Oracle supports authorization-code and client-credentials OAuth flows. The authorization request documentation names the REST scope rest_webservices. The integration record must enable REST Web Services, and the role still needs appropriate record permissions plus Log in using OAuth 2.0 Access Tokens. A service scope is not blanket permission to every record.

Collection paging uses limit and offset, with a default page size of 1,000 and response navigation links. Follow the next link and persist progress. SuiteQL requires the Prefer: transient header and returns at most 100,000 results.

Limits concern concurrency rather than a universal requests-per-minute allowance. Oracle documents base limits of 5 for Standard, 15 for Premium, and 20 for Enterprise or Ultimate under newer contracts, with 10 additional concurrent requests per SuiteCloud Plus license. REST, SOAP, and RESTlets share account capacity. Watch HTTP 429 and CONCURRENCYLIMITEXCEEDED; quota response headers are not documented on these pages.

The reviewed REST guide does not document a general outbound record-webhook subscription service, signing header, or delivery-retry contract. Oracle does document asynchronous Event Subscriber scripts for supported record events, deployed through SDF. Any external callback built around scripts needs its own authenticated delivery and replay controls; checkpointed polling is the simpler starting point.

Use sandbox and Release Preview accounts for testing. Production OAuth authorizations are not copied, and sandbox refreshes require reauthorization.

GET /services/rest/record/v1/customer

Customer collection filtering retrieves customer references using supported filters.

  • App: Maintain a customer-matching queue and source identifiers.
  • Agent: Assess ambiguous legal names after candidate retrieval.
  • Workflow: Page through a bounded customer refresh and checkpoint completion.

GET /services/rest/record/v1/customer/107

Getting a record instance returns one customer; 107 is Oracle's example internal ID, not a fixed production identifier.

  • App: Display the current customer beside a proposed match.
  • Agent: Compare address and company context before recommending a mapping.
  • Workflow: Re-read the selected customer before an approved downstream operation.

POST /services/rest/record/v1/customer

Creating a record instance creates a customer from a schema-valid body with mandatory fields.

  • App: Collect the proposed identity and required subsidiary information.
  • Agent: Explain why existing candidates do not represent the customer.
  • Workflow: Create only after approval and retain the returned Location header.

PATCH /services/rest/record/v1/customer/107

Updating a record instance changes selected customer fields; substitute the actual internal ID.

  • App: Present the old value and proposed change as a reviewable diff.
  • Agent: Evaluate contradictory source evidence and recommend a correction.
  • Workflow: Validate approval freshness, apply permitted fields, and verify the result.

PUT /services/rest/record/v1/customer/eid:CID002

Upsert by external ID creates or updates the customer identified by an external ID; CID002 is Oracle's example.

  • App: Keep the source identity and external-ID mapping durable.
  • Agent: Resolve uncertain identity links before any upsert becomes eligible.
  • Workflow: Reuse the approved identifier rather than issue an untracked create.

GET /services/rest/record/v1/metadata-catalog

Getting metadata exposes the account's available record definitions and links to schemas.

  • App: Store a validated configuration version for required fields.
  • Agent: Explain a newly encountered field using retrieved schema context.
  • Workflow: Detect schema drift and pause affected writes for administrator review.

POST /services/rest/query/v1/suiteql

Executing SuiteQL queries retrieves reporting and investigation context from permitted records.

  • App: Calculate repeatable comparison totals from approved query results.
  • Agent: Interpret mismatches and draft an evidence-linked explanation.
  • Workflow: Execute bounded queries with approved parameters and store the reporting window.

GET /services/rest/query/v1/dataset/{datasetID}/result

Dataset execution returns results from a defined SuiteAnalytics dataset.

  • App: Preserve approved dataset outputs and reporting timestamps.
  • Agent: Explain period-over-period changes using those results.
  • Workflow: Run the selected dataset on schedule and route incomplete runs for review.

These are building blocks, not permission to expose unrestricted CRUD or arbitrary queries to an agent. Keep query definitions, field allowlists, and approved write operations in deterministic software.

NetSuite automation use cases on Major

Shopify order exceptions and customer identity

Paid orders need consistent customer, item, discount, currency, and subsidiary mappings before they enter the ERP.

  • App: Hold order mappings, duplicate checks, and unresolved exceptions.
  • Agent: Judge ambiguous customer identities and explain item or discount mismatches.
  • Workflow: Read Shopify context, retrieve NetSuite candidates, and require approval for new identities or material exceptions.
  • Endpoints: Customer collection and customer-instance reads above, plus SuiteQL for approved ERP lookups. Customer creation or upsert belongs behind an explicit identity approval.

Ramp coding review without duplicate posting

The published Ramp guide uses a review layer before Ramp's existing NetSuite sync, rather than creating a second journal-entry path.

  • App: Store uncoded transactions, proposed accounts, review status, and sync evidence.
  • Agent: Propose coding from merchant and memo context, explaining uncertainty.
  • Workflow: Route the controller's review and leave posting ownership with the existing sync.
  • Endpoints: SuiteQL or dataset execution can supply approved NetSuite reference and reconciliation context. Do not interpret this as authority for direct journal creation.

NetSuite cash-position briefs in Slack

A weekly brief should distinguish observed balances from interpretation and missing inputs.

  • App: Store reporting windows, balance snapshots, and unresolved reconciliation items.
  • Agent: Explain changes, likely drivers, and evidence gaps without declaring an unaudited number final.
  • Workflow: Run approved SuiteQL queries or a finance-owned dataset, require finance review when checks fail, and deliver the approved brief to Slack.
  • Endpoints: SuiteQL and dataset-result execution above, with bounded retrieval windows.

Salesforce bookings compared with ERP reporting

A closed-won opportunity and recognized revenue are different accounting concepts. Sales orders alone do not prove revenue recognition.

  • App: Maintain opportunity-to-ERP mappings and finance-approved comparison definitions.
  • Agent: Classify timing differences, identity mismatches, and missing evidence.
  • Workflow: Retrieve Salesforce opportunities and approved ERP reporting results, then route unresolved variances to Finance and RevOps.
  • Endpoints: Customer reads establish identity; SuiteQL or dataset results supply finance-defined measures. The agent must not invent a revenue-recognition policy from order status.

Build the NetSuite architecture on Major

NetSuite and Shopify connect through a Major workflow to agent judgment, approval, and a stateful review app

NetSuite context becomes a reviewable proposal before any consequential action.

Build the deterministic app layer first. It is both the operator's control surface and the execution layer the agent runs, with managed database tables for source IDs, retrieval windows, evidence versions, approvals, and confirmed outcomes.

  1. Connect: IT configures the NetSuite custom API connector and the other systems involved in the process.
  2. Retrieve: A workflow runs on schedule or a supported source-system trigger, collects bounded inputs, and handles paging.
  3. Validate: App code checks required fields, approved mappings, totals, and duplicate keys.
  4. Judge: The agent receives unresolved cases through scoped connectors and recommends a match, classification, or next action.
  5. Review: The app shows evidence and the proposal; the workflow routes consequential changes to a person.
  6. Confirm: Approved execution records the result and read-back evidence before marking the item complete.

When a matching rule becomes stable, move it into app code. The next run reuses that software instead of asking the agent to reason through the same normalization again. Judgment stays with the agent; state stays in the app.

A worked example with illustrative numbers

For example, consider an ecommerce finance team reviewing 400 paid orders each week. Assume staff currently spend 3 minutes per order assembling source and ERP context: 400 × 3 = 1,200 minutes, or 20 hours. These are illustrative assumptions, not a customer result or a promised saving.

A Major app stores approved mappings and runs repeatable checks. Suppose 320 orders satisfy those checks and 80 require judgment. The agent proposes resolutions for those 80, while finance retains approval over customer creation and material discrepancies.

If each exception needs 4 minutes of human review, that is 80 × 4 = 320 minutes. Add a 40-minute weekly control check across the complete run, giving 360 minutes, or 6 hours of modeled human effort. The arithmetic suggests a 14-hour difference against the assumed baseline, before implementation, monitoring, or rework costs.

Measure the exception rate, reviewer corrections, duplicate prevention, and reconciliation completeness during a pilot. If ambiguous discounts affect half the orders, this example's assumptions do not hold. The first goal is a trustworthy record of what happened, not a time-saving claim.

Governance, approvals, and failure handling

NetSuite financial proposals pass policy checks and a controller queue before restricted writes and audit confirmation

Read-only reporting and financial writes follow different approval paths.

Separate read-only reporting from operations that change financial records. An agent may recommend a classification; it must never approve its own posting, change bank details alone, override a closed-period control, or expand its permissions. Human review is a workflow gate, not a sentence in the prompt.

  • Stale approvals: Bind approval to the record version, proposed fields, evidence, and amount. Material changes return the item to review.
  • Capacity failures: Reduce concurrency and retry transient failures with increasing delays and jitter. Oracle's retry guidance recommends waiting, increasing delays, and avoiding synchronized retries.
  • Uncertain writes: A timeout is not proof of failure. Reconcile the source mapping and ERP state before repeating a create.
  • Duplicate events: Enforce a durable source-and-operation key, whether inputs arrive through polling or a custom callback.
  • Permanent errors: Quarantine permission, schema, and validation failures rather than retrying forever.

Retain the source references, agent rationale, reviewer identity, permitted action, request outcome, and read-back result. Apply retention and access rules to sensitive finance evidence as carefully as to connector credentials.

NetSuite AI Apps, Agents and Workflow Automations

Every published Major use case that works with NetSuite, newest first. Each one names the NetSuite API objects involved and shows the app, agent and workflow split.

Frequently asked questions

What are NetSuite AI agents?

They are agents that interpret NetSuite context and recommend or carry out permitted work. On Major, deterministic apps hold mappings, evidence, and review queues, agents handle judgment, and workflows control execution and approvals.

Does NetSuite have built-in AI?

Yes. Oracle documents Text Enhance, Exception Management, Prompt Studio, and other features. Availability varies by feature and account. Major adds cross-system orchestration and durable review state around that native functionality.

Does Major have a native NetSuite connector?

No. The current catalog lists neither a native nor a Pipedream connector for NetSuite. IT configures a custom API connector using the account-specific SuiteTalk base URL and supported authentication.

Can a NetSuite agent post financial changes without review?

A controlled implementation should require human approval for consequential financial changes. Limit write permissions, validate the proposed action in code, and keep approval separate from the agent's recommendation.

Can I test NetSuite automation in a sandbox?

Yes. Oracle documents sandbox OAuth authorization separately from production, with reauthorization required after a sandbox refresh. Test permissions, account-specific fields, duplicate handling, and failure recovery before enabling production writes.

What limits matter for NetSuite REST automation?

Account-level concurrency is shared with SOAP and RESTlets. Handle HTTP 429, page bounded result sets, and monitor incomplete runs. SuiteQL has a documented maximum of 100,000 results, so reporting design must account for that ceiling.

Sources

Start building on Major

Build a NetSuite exception-review app with durable mappings, scoped agent judgment, and human approvals. Start with read-only evidence gathering, then add approved actions once finance and IT have validated the controls.

About the author

Ritvij Gautam is Head of Growth at Major.

Last reviewed November 19, 2026 by Ritvij Gautam.

Frequently asked questions

What are NetSuite AI agents?
They are agents that interpret NetSuite context and recommend or carry out permitted work. On Major, deterministic apps hold mappings, evidence, and review queues, agents handle judgment, and workflows control execution and approvals.
Does NetSuite have built-in AI?
Yes. Oracle documents Text Enhance, Exception Management, Prompt Studio, and other features. Availability varies by feature and account. Major adds cross-system orchestration and durable review state around that native functionality.
Does Major have a native NetSuite connector?
No. The current catalog lists neither a native nor a Pipedream connector for NetSuite. IT configures a custom API connector using the account-specific SuiteTalk base URL and supported authentication.
Can a NetSuite agent post financial changes without review?
A controlled implementation should require human approval for consequential financial changes. Limit write permissions, validate the proposed action in code, and keep approval separate from the agent's recommendation.
Can I test NetSuite automation in a sandbox?
Yes. Oracle documents sandbox OAuth authorization separately from production, with reauthorization required after a sandbox refresh. Test permissions, account-specific fields, duplicate handling, and failure recovery before enabling production writes.
What limits matter for NetSuite REST automation?
Account-level concurrency is shared with SOAP and RESTlets. Handle HTTP 429, page bounded result sets, and monitor incomplete runs. SuiteQL has a documented maximum of 100,000 results, so reporting design must account for that ceiling.