Skip to content
NexusBook a demo

Enterprise use case · NetSuite + Snowflake + OpenAI

Demand forecasts with human override

Planning teams rebuild forecasts from historical extracts and undocumented judgment, making it difficult to explain why a plan changed. Planners receive versioned item-level forecasts they can challenge and adjust without losing the original model proposal.

Discuss this use case Enterprise overview

Systems involved

What each system contributes

NetSuite

38 actions

  • Items10
  • Base8
  • Invoices10
Snowflake

35 actions

  • Tables6
  • Views6
  • Warehouse3
  • Tasks & Scheduling7
OpenAI

25 actions

  • Chat Completions8

The model step runs against the AI provider you choose, including Anthropic, Google Gemini or a model you host yourself. Anything outside a pre-built action runs through an API, protocol or database call, or as Rules + Code in Python, Bash, Go, Java, .NET, PHP, PowerShell or JavaScript/TypeScript, with the same approvals and audit.

Architecture

One governed run

Trigger: schedule / planner requestIntegration: NetSuite + SnowflakeRules + CodeAI step: selected forecasting modelHuman ApprovalSystem Action: SnowflakeAudit

How it starts: A planning cycle opens or a planner releases an updated item dataset for forecasting.

Step by step

How the process runs

  1. 01

    Assemble the planning data

    Read permitted NetSuite item and transaction history through the configured interfaces and align it with the curated Snowflake dataset.

  2. 02

    Check the forecast basis

    Rules validate item keys, periods, units and missing data before the forecasting model receives the selected history and approved drivers.

  3. 03

    Generate the proposal

    The selected forecasting model produces item-level demand estimates for the agreed horizon with its assumptions and applicable uncertainty information.

  4. 04

    Prepare planner explanations

    The model-assisted workflow relates changes to available seasonality, item history and supplied drivers without presenting unsupported causes as facts.

  5. 05

    Capture the override

    The named planner accepts or adjusts the proposal with a recorded reason before an approved step publishes the selected planning version to Snowflake.

Control

Human decision and exception path

Human decision

The named planner approves the released forecast version, and any resulting purchase, production or financial commitment requires its own designated approval. The forecast does not create a NetSuite order or change a posted document.

Exception path

  • Sparse history or a new item requires a documented fallback or planner estimate rather than fabricated seasonal evidence.
  • Invalid or implausible outputs hold the forecast for review against the configured checks.
  • Changed source data or drivers create a new proposal version without erasing the approved override history.

Evidence

What gets audited

  • NetSuite references, Snowflake snapshot and forecast horizon.
  • Model configuration, supplied drivers and validation results.
  • Original proposal, planner override and stated reason.
  • Approved planning version and publication response.

Nexus runs on premises, in private or sovereign cloud, or air-gapped. Credentials stay in the vault, roles decide who can approve, and the execution record stays inside your deployment boundary.

Sovereignty and deployment

Related

Patterns, connectors and next steps

FAQ

Questions about this use case

Does Nexus replace any of these systems?
No. Each system keeps its own records. Nexus runs the process between them and records each decision.
How does the run start?
A planning cycle opens or a planner releases an updated item dataset for forecasting.
What if a step is not covered by a pre-built action?
The model step runs against the AI provider you choose, including Anthropic, Google Gemini or a model you host yourself. Anything outside a pre-built action runs through an API, protocol or database call, or as Rules + Code in Python, Bash, Go, Java, .NET, PHP, PowerShell or JavaScript/TypeScript, with the same approvals and audit.
Is a general-purpose language model assumed to be a validated forecaster?
No; the forecasting model must be selected and evaluated for the available history and planning task, while a language model may support explanation and interaction.
Can a planner disagree with the forecast?
Yes; the released plan retains both the original proposal and the authorized override with its reason.

Use case review

Discuss this use case

Bring your version of the process. We will map the systems, approvals and evidence it needs.

Discuss this use case Book a demo