Manufacturing SystemsERP Platform

IBM Maximo + private AI

AI for IBM Maximo That Reads Your Work Orders and Failure History Without Leaving Your Boundary

Short answer

AI for IBM Maximo connects a private language model to Maximo's REST and integration framework so planners, reliability engineers and technicians can ask plain-English questions about WORKORDER, ASSET and failure data and get answers grounded in live records. It runs on-prem or in a sovereign cloud next to Maximo Manage or a 7.6.1.x install, and any write-back is drafted for human approval.

ERP: IBM Maximo
Industries: Manufacturing, Aerospace, Utilities, Defense
Written for: Maintenance and Reliability Manager

Discuss your IBM Maximo environment See use cases

Maximo is usually the system of record for the physical side of the business: every asset with its location hierarchy, every preventive maintenance schedule, every work order from request to close, every failure class, problem, cause and remedy that a technician ever picked from a list. After ten or fifteen years of use, that is an extraordinary body of knowledge, and most of it is only reachable through list screens, saved queries and reports built for questions someone asked years ago.

The people who need the answers are not database people. A reliability engineer wants to know which pump models on the east side keep failing within a year of a rebuild. A planner wants to know which approved work orders are blocked by a part that is not in the storeroom. A technician wants the last three times this fault appeared on this asset and what fixed it. Each of these is a join across WORKORDER, ASSET, LOCATIONS, FAILUREREPORT, MATUSETRAN and INVBALANCES, and each currently becomes a request to the one person who knows how to build the report.

IBM is adding its own AI features to Maximo Application Suite, and for many sites those are the right place to start, particularly if you are already moving to MAS on OpenShift. A private layer earns its place when you run Maximo 7.6.1.x on-prem for years to come, when your data is subject to defense or customer restrictions on cloud AI, or when you want questions that cross Maximo, the ERP and documents like OEM manuals and SOPs.

This page describes how such a layer works against Maximo's data model and APIs, which use cases are realistic first, where the risks are, and what to ask any vendor before they touch a production maintenance system.

What usually gets in the way

The problems we hear most from maintenance and reliability manager teams running IBM Maximo.

Failure data is rich but hard to query

Failure class, problem, cause and remedy codes live in a hierarchy that technicians choose from under time pressure. Turning that into a trend for one asset class means joining several tables, so most sites look at it once a year, if at all.

Work order backlog is hard to explain

A backlog number is easy to report. Why those work orders are waiting (parts, labor craft, permits, outage windows, approval) takes a planner hours to assemble from WORKORDER status, WPMATERIAL, WPLABOR and inventory balances.

Heavy customization limits upgrades

Years of automation scripts, custom MBOs, domains and application designer changes make a move from 7.6.1.x to MAS Manage a project in itself, and generic documentation no longer describes the instance you run.

Knowledge walks out with senior technicians

The real fix for a recurring fault is often in long descriptions, attached PDFs and the memory of one person. New technicians cannot search what they do not know exists.

Maintenance data meets restricted-data rules

Aerospace, defense and utility sites may hold asset configurations or location data under contract or regulatory limits, which rules out pasting records into a public chatbot.

Where AI earns its place in IBM Maximo

Each use case names the ERP objects it reads or writes, so your ERP team can judge the integration effort before anyone commits budget.

Natural-language backlog and planning questions

Planners ask which approved work orders are waiting on material, labor or a shutdown window, and get a grouped list built from a live query with the filter shown.

Touches: WORKORDER, WPMATERIAL, WPLABOR, INVBALANCES, ITEM

Outcome: Shifts backlog explanation from an hours-long manual assembly to a question answered in seconds, leaving planners time for exceptions.

Technician fault history assistant

From a phone or tablet, a technician asks what has happened on this asset or similar assets with this symptom, and gets prior work orders, causes and fixes with links back to the records.

Touches: ASSET, WORKORDER, FAILUREREPORT, WORKLOG, DOCLINKS

Outcome: Reduces repeat diagnostic time on recurring faults, most noticeably for newer technicians.

Failure pattern analysis for reliability engineers

The assistant summarizes failure codes, repair intervals and part consumption by asset class or manufacturer and flags candidates for a PM change or a root cause study.

Touches: FAILURECODE, FAILUREREPORT, ASSET, MATUSETRAN, LABTRANS

Outcome: Moves failure review from an annual spreadsheet exercise to a routine question, with the engineer still deciding what to change.

PM optimization candidates

Compares PM work orders and findings to identify tasks that never find anything or intervals that are consistently too long, as a review list for engineers.

Touches: PM, JOBPLAN, WORKORDER, WOACTIVITY

Outcome: Produces an evidence-based shortlist for PM review rather than a gut-feel one.

Work order drafting from requests

Reads incoming service requests, proposes an asset, job plan, priority and failure class, and drafts the work order for the planner to confirm.

Touches: SR, WORKORDER, ASSET, JOBPLAN, CLASSSTRUCTURE

Outcome: Cuts triage time per request, while a planner remains the approver of every record created.

Spares and storeroom questions

Answers which critical spares are below reorder, which have not moved in years, and which assets depend on a single-source part.

Touches: INVENTORY, INVBALANCES, INVRESERVE, ASSETSPAREPART, PO, POLINE

Outcome: Surfaces slow-moving and at-risk stock without a custom report per storeroom.

Manual and SOP retrieval beside the work order

Retrieval over OEM manuals, lockout procedures and internal SOPs, answered with citations and tied to the asset being worked on.

Touches: DOCLINKS, ASSET, LOCATIONS, JOBPLAN, document stores

Outcome: Lowers time spent hunting for the right procedure page, with a citation the technician can open.

Reference architecture

The AI layer sits beside Maximo and reads through its supported APIs and an optional read replica, so security groups and site-level access are respected. A private model is served locally, and anything that writes goes through a human-approved action.

1
Maximo connector
Uses the REST API (/maximo/api/os/ object structures) or OSLC endpoints with a dedicated API key and a security group scoped to the sites and applications the assistant needs. Integration framework object structures can expose exactly the fields required.
2
Semantic layer
Maps WORKORDER, ASSET, LOCATIONS, PM, INVENTORY and failure tables to the terms maintenance staff use, including site, organization and status synonyms (WAPPR, APPR, INPRG, COMP, CLOSE) and your own domains and custom attributes.
3
Model serving
Open-weight models served with vLLM or Ollama on customer GPUs or a sovereign cloud, sized to the number of concurrent planners and technicians rather than to the size of the asset register.
4
Retrieval and agents
Live API queries for structured questions, vector retrieval over long descriptions, work logs and manuals for unstructured ones, and agents that draft work orders or PM changes held for approval.
5
Governance and audit
Each question, generated query and answer is logged against the requesting Maximo user and mapped to existing security groups; no record is written without an approver.

Integration notes for your ERP team

Authenticate with a Maximo API key tied to a dedicated user in a limited security group, not a maxadmin account.
Query through REST object structures such as MXWO, MXASSET and MXPO, or custom object structures built in the Object Structures application to limit fields and relationships.
Respect site and organization security so the assistant only sees the sites the requesting user can see.
For heavy analytics such as multi-year failure trends, use a read replica of the Db2, Oracle or SQL Server database rather than hammering the application server.
Index automation script and domain definitions so the assistant understands custom statuses and attributes after a change.
Write-back uses a queued, human-approved action that calls the standard API, never a direct SQL insert, so workflows and escalations still fire.
Test against your exact release, including 7.6.1.x fix packs and MAS Manage versions, since API paths and fields differ.

Deployment options

Air-gapped on-prem

Defense, aerospace and critical infrastructure sites running Maximo 7.6.1.x or MAS on their own infrastructure.

Model, retrieval index and connector run inside the same network as Maximo with no outbound path; documents are indexed locally.

Private or sovereign cloud

Organizations on MAS in a customer-managed cloud who want AI under their own tenancy.

Deployed in the customer's own cloud account or region, on separate nodes from the Maximo cluster so GPU workloads do not compete with it.

Hybrid

Groups with some sites on legacy Maximo and others already on MAS.

A central governance layer connects to each instance through its own API key, keeping data residency per site.

Compliance and data control

How the architecture supports your obligations. Certification and accountability stay with your organisation; the design keeps the evidence straightforward.

Aviation maintenance records (FAA 14 CFR Part 43 and 145, EASA Part-145)

The assistant is read-only against maintenance records and drafts nothing into a regulated record without a certifying person; the customer keeps its existing signoff process.

ITAR and export-controlled technical data

A private deployment inside the customer's boundary lets the customer keep controlled asset and design data on infrastructure it already governs; the obligation to restrict access to authorized persons remains the customer's.

Safety procedures (OSHA 1910.147 lockout/tagout, site permit-to-work)

Retrieval cites the current approved procedure and never invents steps; the assistant does not replace the permit or isolation process.

GDPR and regional data protection

Labor and technician records stay in the customer's environment, and the semantic layer can exclude personal fields from retrieval.

Where Netray fits

Custom build

Maximo is highly configurable and sites differ widely, so a bespoke connector and semantic layer built around your object structures, failure hierarchy and custom attributes is usually the right starting point.

DataRay

Useful when the same questions must cover Maximo, historian exports, manuals and shared drives, with chat over mixed sources on-prem.

How an engagement runs

Phase 1 . 2-4 weeks

Discovery

-Review of sites, object structures, customizations and security groups
-Ranked use cases by planner and reliability pain
-Data quality check of failure codes and asset hierarchy
-Deployment recommendation and sizing

Phase 2 . 6-8 weeks

Pilot

-Read-only connector against a test or replica environment
-3-4 use cases live for a pilot site
-Accuracy review against answers the team already trusts
-Logging validated against security groups

Phase 3 . 4-6 weeks

Production

-Production service account with scoped access
-Audit logging and monitoring
-Runbook for adding use cases
-Training for planners and technicians

Phase 4 . Ongoing

Scale

-Additional sites and organizations
-Approved write-back flows for draft work orders
-Model refresh as open-weight options improve
-Quarterly accuracy and usage review

Questions to ask any vendor, including us

A short list that separates real IBM Maximo AI work from a chatbot demo.

Does the connector use Maximo's supported REST or integration APIs under a scoped security group, or direct database credentials?
How are site and organization restrictions enforced for each question?
How does the assistant handle our custom statuses, domains and automation scripts?
What happens after we move from 7.6.1.x to MAS Manage?
Can you show the query behind every answer so a planner can verify it?
Where does the model run, and what is the network path for our data?
Can we start read-only before any work order drafting is enabled?
How does this relate to IBM's own AI features in MAS, and when would we choose those instead?

Frequently asked questions

Should I use IBM's AI features or a private layer for Maximo?

If you are on or moving to Maximo Application Suite and are comfortable with IBM's cloud services, start with IBM's own features, since they are integrated and supported. A private layer makes sense when you stay on 7.6.1.x, face restrictions on cloud AI, or need questions across Maximo, ERP and documents.

Does the AI work with Maximo 7.6.1.x as well as MAS?

Yes. Both expose REST and integration framework endpoints, though paths, fields and authentication details differ by release. The connector is built and tested against your exact version, and the semantic layer reflects your own customizations.

Can it create or close work orders?

It can draft work orders and PM changes. Anything that writes goes through a human approval step and the standard Maximo API, so workflows, escalations and signoffs still apply. Closing or certifying regulated work stays with the responsible person.

How good is the answer if our failure codes are messy?

Answers are only as good as the data. Discovery includes a failure code and asset hierarchy check, and the assistant can use work log text to supplement weak codes, but it cannot recover information that was never recorded.

Will it slow down Maximo?

Structured queries are scoped and rate limited, and heavier analytical questions can use a read replica. The model runs on separate hardware from the Maximo application servers, so GPU load does not compete with maintenance users.

Is predictive maintenance part of this?

Not by itself. Sensor-based prediction is a different problem handled by tools like Maximo Monitor and Predict or a historian. The assistant helps interpret maintenance history and can bring condition data into context, but it is not a replacement for a validated predictive model.

What does it need from our team?

A read-scoped API user, a database replica if needed, a named reliability or planning lead to judge answers, and access to a sample of documents. Most of the effort is agreeing the vocabulary and verifying answers, not infrastructure.

Related guides

SAP PM/EAM + on-prem AI

AI 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.

Maintenix + on-prem AI

AI for IFS Maintenix in aviation MRO

AI over IFS Maintenix maintenance records, AD/SB compliance, and task cards for MRO, deployed on-prem so tech records never leave your network.

MRO technical records + on-prem AI

AI for aviation MRO maintenance records and technical documentation

On-prem AI reads AD and SB packets, cross-checks logbooks and work orders against your MRO ERP, and drafts compliance narratives, without technical records leaving your network.

RAG + SQL + permissions

A 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.

Agents + approval gates

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.

Opcenter Execution + private AI

AI 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

Free Tool

Talk it through with an engineer who knows IBM Maximo

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.

Start the conversation See ERPray

All AI for ERP guides