SiteWise + ERP + private LLM
AI for AWS IoT SiteWise: Asset Models, Property Values and ERP Context in One Question
Short answer
AI for AWS IoT SiteWise reads asset models and property values through the SiteWise APIs, joins them with ERP orders and maintenance records, and answers plant questions in plain English with the queries shown. The model can be hosted inside the customer's AWS account or on-prem, so the choice of a managed AWS model service versus an open-weight model stays with the customer.
- ERP
- AWS IoT SiteWise
- Industries
- Manufacturing, Food and Beverage, Energy, Electronics
- Written for
- Industrial Data Engineer
AWS IoT SiteWise models equipment as assets built from asset models, with properties of type attribute, measurement, transform or metric, arranged in hierarchies. Data arrives from OPC UA servers through a SiteWise Edge gateway or from other AWS services, and is stored in hot and cold tiers. For plants already standardised on AWS, it is a natural historian replacement.
What it does not provide is business meaning. A measurement property named Motor_Amps on an asset in a hierarchy tells you nothing about which product was running or whether a work order was late. That meaning sits in the ERP, and the join is almost always done in a spreadsheet or a one-off Athena query.
Teams on AWS have options: Amazon managed model services, or an open-weight model they host themselves. The right answer depends on data classification and cost. Plants with customer-restricted or export-controlled data often prefer a model they host in their own VPC or on-prem, and SiteWise Edge makes local processing possible.
This page describes how to put a language layer over SiteWise and ERP data, what to bound and log, and when native AWS tools are enough.
What usually gets in the way
The problems we hear most from industrial data engineer teams running AWS IoT SiteWise.
Asset models without business context
SiteWise hierarchies reflect the physical plant. Products, orders, customers and quality results live elsewhere, so cross-domain questions need a custom query.
Query skills sit with a few engineers
Property history APIs, SQL-style queries and S3 exports are comfortable for data engineers and opaque to supervisors and quality staff.
Hot and cold data complexity
Recent values and older exports sit in different tiers with different access paths, so questions spanning months need care.
Data classification limits
Process parameters for controlled programs may not be allowed into a shared model service, forcing an AI design decision early.
Cost surprises from broad queries
Unbounded history pulls and large model prompts both cost money. Without limits, an assistant can become an expensive way to ask simple questions.
Where AI earns its place in AWS IoT SiteWise
Each use case names the ERP objects it reads or writes, so your ERP team can judge the integration effort before anyone commits budget.
Cross-domain Pareto of downtime
Combine downtime-related properties with ERP production orders and product families to rank causes.
Touches: SiteWise assets and measurements; ERP production orders, operation confirmations
Outcome: Rankings by product and line that were previously assembled by hand in a notebook.
Natural-language access to asset hierarchies
Ask for all pumps in a hierarchy with their latest values and alarm status, without knowing property IDs.
Touches: Asset model hierarchies, properties; ERP equipment master
Outcome: Fewer requests to the data engineering team for lookups.
Process drift before quality escapes
Correlate parameter drift with lot results and scrap to identify candidates for review.
Touches: Lot, inspection and scrap records; SiteWise measurements
Outcome: Candidate causes surfaced for an engineer to confirm; correlation is not proof of cause.
Energy and utilities by product
Allocate energy readings to production runs and report energy per unit.
Touches: Production order times and quantities; energy metrics
Outcome: Per-product energy views usable in cost discussions, with alignment assumptions stated.
Maintenance prioritisation
Show condition metrics beside open work orders and spare stock to support scheduling.
Touches: Maintenance orders, equipment, spares; SiteWise properties
Outcome: Better-informed scheduling; planners approve any order changes.
Asset model quality checks
Find properties with missing units, inconsistent naming, or stale data, and propose cleanup lists.
Touches: Asset models, property aliases, data quality indicators
Outcome: A cleanup backlog for the data team rather than silent inconsistency.
Shift handover summaries
Summarise notable events, orders run and exceptions per line for the next shift.
Touches: Production orders, shift calendars; alarm and metric values
Outcome: Consistent handovers, with each statement linked to a source.
Reference architecture
An orchestrator calls SiteWise APIs with a least-privilege IAM role and reads the ERP read-only. A model hosted in the customer's VPC or on-prem plans bounded calls, with logs retained in the customer's account.
- 1
SiteWise access
IAM role limited to describe and read property values and history; VPC endpoints used where available; time windows and aggregate resolutions chosen per question.
- 2
Asset catalogue
Indexed asset models, hierarchies, units and aliases so the model resolves names before querying.
- 3
ERP read path
Read-only access to orders, routings, lots and maintenance, linked to assets through an equipment key.
- 4
Model hosting
Open-weight model on GPU instances in the customer's account, or on-prem, with an optional managed model service if policy allows.
- 5
Logging and approval
Calls, prompts and answers stored in customer-controlled logging; state changes need human approval.
Integration notes for your ERP team
- Create a dedicated IAM role with read-only SiteWise permissions and no wildcard resource access.
- Use asset property aliases and the describe APIs to build a catalogue; refresh it on a schedule.
- Choose aggregate resolution deliberately when requesting history to avoid huge raw pulls.
- For long ranges, query exports in S3 using a controlled query service rather than live API paging.
- Align timestamps to ERP time zones and shift calendars explicitly.
- Check current SiteWise API names and limits in AWS documentation, since the service evolves.
Deployment options
Inside the customer's AWS account
AWS-standard organisations
Model and orchestrator in a private VPC with endpoints to SiteWise; data stays in the account.
SiteWise Edge with on-prem inference
Plants needing local processing
Edge gateway keeps data local; the model runs on a plant server and queries local data and the ERP.
Hybrid
Mixed estates
Recent data local, historical exports in the cloud, with a single question layer over both.
Compliance and data control
How the architecture supports your obligations. Certification and accountability stay with your organisation; the design keeps the evidence straightforward.
IEC 62443
The AI layer reads data from a defined zone with no write path to the control network.
ITAR / EAR (customer obligation)
Controlled data stays inside infrastructure the customer designates; model choice is made with the classification in mind.
SOC 2 / ISO 27001 customer controls
IAM least privilege, CloudTrail-style audit, and customer-managed keys.
GDPR
Personal data such as operator IDs minimised before prompts; retention configured by the customer.
How an engagement runs
Phase 1 . 1-2 weeks
Assessment
- -Asset model review
- -Data classification decision
- -ERP join key plan
- -Cost guardrails
Phase 2 . 2-4 weeks
Build
- -IAM roles and endpoints
- -Asset catalogue
- -Bounded query tools
- -ERP read views
Phase 3 . 3-5 weeks
Pilot
- -Selected lines live
- -Evaluation set
- -Source-shown answers
- -Logging
Phase 4 . 2-4 weeks
Scale
- -More assets and sites
- -Cost monitoring
- -Runbook
- -Model sizing review
Questions to ask any vendor, including us
A short list that separates real AWS IoT SiteWise AI work from a chatbot demo.
- Which IAM permissions does the assistant need, and are they read-only?
- Does any data go to a managed model service, and can we choose?
- How are hot and cold tiers queried and bounded?
- How do we cap query and inference cost?
- How is the asset-to-ERP join maintained?
- Can it run with SiteWise Edge on-prem?
- Where are logs stored and who can read them?
- When is a QuickSight or Athena dashboard enough?
Frequently asked questions
Can I use Amazon managed models instead of a self-hosted one?
Yes, if your data classification and contracts allow it. Managed services reduce operations effort, while self-hosted open-weight models give more control over data location and cost at high volume. The orchestration, bounded queries and ERP integration are the same either way, so the model choice can be revisited later.
How does the assistant avoid expensive queries?
Each tool enforces time windows, aggregate resolution and row limits, and the orchestrator applies per-user budgets. Long-range questions are routed to exports rather than live API paging. Cost reporting by user and question type helps tune limits, which matters because both data scans and model tokens cost money.
Does it replace Athena or QuickSight dashboards?
No. Dashboards remain best for stable, shared views. The assistant handles ad hoc questions that cross SiteWise and ERP data. If a question recurs weekly, the right outcome is often a dashboard built from the assistant's validated query.
Can it run on-prem with SiteWise Edge?
SiteWise Edge supports local data processing, and a plant-hosted model can query local data and a read-only ERP replica. Cloud-only features and long history remain in AWS, so a hybrid design is common. The trade-off is operating GPU hardware locally.
How reliable is the ERP join?
As reliable as the mapping between asset IDs and ERP equipment or work centers, and the timestamp alignment. Both need an owner. The assistant can flag unmapped assets and ambiguous matches so the data team can fix them rather than guessing silently.
Will it change data in SiteWise or the ERP?
Not by default. The IAM role is read-only, and any proposed action, such as a maintenance request, is drafted for a person to approve in the ERP. Controller-facing writes are out of scope.
Is this worth it for a small AWS footprint?
If only a few assets stream and the standard console views answer the questions, probably not. The case strengthens when many sites or lines have varied models, and engineers regularly spend hours combining SiteWise history with ERP records.
Related guides
AI Agents for ERP, Running On-Prem
A practical guide to on-prem AI agents for ERP: what they can safely automate, where human approval belongs, and how to design the guardrails.
RAG + SQL + permissionsA Private LLM Grounded on Your ERP Data
How a private LLM answers questions on your ERP data: RAG plus text-to-SQL, role-based permissions inherited from the ERP, and where each fits.
Forecasting plus a reason whyAI Demand Forecasting for Manufacturers: Better Numbers and an Explanation Planners Can Trust
AI demand forecasting explains ERP forecast changes in plain language, flags exceptions for planners, and handles short-history SKUs without Excel overrides.
SAP PM/EAM + on-prem AIAI for SAP Plant Maintenance: Notifications, Work Orders, and Predictive Signals
AI grounded on SAP PM data that triages notifications, drafts work orders, standardizes failure codes, and flags predictive maintenance signals, on-prem.
Faster paperwork, same standardAI for NCR, CAPA, and 8D Drafting in AS9100 and ISO 9001 Quality Systems
AI drafts NCR, CAPA, and 8D reports from ERP quality data for engineer review, cutting drafting time under AS9100 and ISO 9001 without changing sign-off.
Opcenter Execution + private AIAI for Siemens Opcenter Execution MES, Running Inside Your Own Network
Add private, grounded AI to Siemens Opcenter Execution MES: ask questions over lots, work orders, nonconformances and genealogy without leaving your network.
Plan it with numbers
OEE Calculator (Overall Equipment Effectiveness)
Break your Overall Equipment Effectiveness into availability, performance, and quality, and see how far you sit from the 85% world-class benchmark.
Free ToolOn-Prem vs Cloud AI Cost Comparison Calculator
Compare your projected cloud LLM API bill against an amortized on-prem GPU deployment and find the breakeven month.
Free ToolLLM API vs Self-Hosted Cost Calculator
Model your API bill from requests and token mix, compare it against an all-in self-hosted monthly cost, and see monthly and annual savings.
GuideIoT Sensor to ERP Data Integration for Manufacturing
Build reliable IoT sensor to ERP data pipelines for manufacturing. Covers MQTT, edge processing, data normalization, and ERP API ingestion patterns.
GuideEdge Computing for Manufacturing ERP Integration
Deploy edge computing for manufacturing ERP integration. Covers edge gateway selection, local processing, latency reduction, and offline-capable ERP operations.
GuideOn-Prem LLM Deployment Architecture: Reference Guide
Reference architecture for on-prem LLM deployment: inference servers, GPU sizing, RAG pipelines, and security zones for regulated manufacturers.
Talk it through with an engineer who knows AWS IoT SiteWise
Bring one real question your team cannot answer from the ERP today. We will map the data path, the model, and where it runs, and tell you honestly if AI is the wrong tool for it.