AI in Maximo: From Condition Insight to Agentic Workflows in MAS 9.2

Artificial intelligence in enterprise asset management has moved from theoretical promise to practical reality. With the release of Maximo Application Suite 9.2 in June 2026, IBM has embedded AI directly into the workflows that reliability, maintenance, field service, safety, and operations teams use every day. This is not AI as a separate dashboard or an add-on module -- it is AI woven into the fabric of asset management operations.

The shift is significant because it addresses the fundamental challenge of AI adoption in industrial settings: getting AI out of the lab and into the hands of the people who need it. Previous approaches required data scientists to build custom models, IT teams to deploy them, and maintenance teams to check a separate dashboard. MAS 9.2 eliminates these barriers by embedding AI directly into the tools that maintenance teams already use.

This article provides a deep technical look at the AI capabilities in MAS 9.2, covering how each capability works, what problems it solves, and how organizations can prepare to take advantage of it. We will examine Condition Insight for predictive analytics, the Maximo Assistant for natural language interaction, agentic workflows via the MCP Server, document intelligence with retrieval-augmented generation (RAG), and AI-powered visual inspection.

Maximo Condition Insight: Actionable Asset Performance Management

The centerpiece of the MAS 9.2 AI capabilities is Maximo Condition Insight, a new AI-powered capability within the Maximo Asset Performance Management (APM) portfolio. Condition Insight provides instant, explainable insights to support condition-based maintenance (CBM), helping organizations move from reactive to prescriptive maintenance. It was first released as part of the AI Service Component feature in December 2025 and has been significantly enhanced in the June 2026 GA release.

Condition Insight works by aggregating and analyzing data from multiple sources across the Maximo ecosystem. It evaluates work orders, inspection results, meter readings, time-series sensor data, Failure Mode and Effects Analysis (FMEA) records, and alerts to assess each asset's condition. The analysis is powered by IBM watsonx AI, which applies machine learning models trained on the organization's own asset data. This is critical -- the models learn the specific failure patterns of your assets, not generic industry patterns.

The key innovation in Condition Insight is that it does not just flag anomalies -- it explains them. When the system identifies a potential issue with an asset, it returns a plain-language summary of the condition, the evidence supporting the assessment, the trend over time, and recommended corrective actions. This makes AI practical for every maintenance team, not just data scientists. A maintenance planner with 20 years of experience but no data science background can understand why the system is flagging an asset and decide whether to act on the recommendation.

// Example: Condition Insight API response for a pump asset
GET /maximo/api/apm/conditioninsight/v1/assets/PUMP-4421/condition

{
  "assetId": "PUMP-4421",
  "overallCondition": "FAIR",
  "conditionScore": 72,
  "confidence": 0.89,
  "analysis": {
    "summary": "Vibration levels on the drive-end bearing have increased 35% over the past 30 days. Current reading of 4.2 mm/s exceeds the warning threshold of 3.5 mm/s but remains below the critical threshold of 6.0 mm/s. Temperature readings are within normal range.",
    "trend": "DETERIORATING",
    "contributingFactors": [
      {
        "factor": "Bearing Vibration",
        "severity": "WARNING",
        "trend": "INCREASING",
        "threshold": 3.5,
        "currentValue": 4.2,
        "unit": "mm/s"
      },
      {
        "factor": "Bearing Temperature",
        "severity": "NORMAL",
        "trend": "STABLE",
        "currentValue": 68,
        "unit": "C"
      }
    ],
    "recommendedActions": [
      {
        "priority": "HIGH",
        "action": "Schedule bearing inspection within 14 days",
        "workType": "CM",
        "estimatedDuration": 4,
        "estimatedCost": 850
      },
      {
        "priority": "MEDIUM",
        "action": "Review lubrication schedule and verify grease type matches specification",
        "workType": "PM",
        "estimatedDuration": 1,
        "estimatedCost": 150
      }
    ]
  },
  "lastUpdated": "2026-07-14T08:30:00Z"
}

The system supports multiple analysis modes. In advisory mode, Condition Insight presents recommendations to the maintenance planner, who reviews and approves them before work orders are created. In automated mode, the system creates work orders directly, with the recommended actions and priority levels. Organizations can start in advisory mode and transition to automated mode as they build confidence in the system. The transition typically takes three to six months as the maintenance team validates the system's recommendations against actual outcomes.

Condition Insight requires the Health and Predict applications to be deployed and configured with sensor data sources. It also requires a minimum of six months of historical data for model training. Organizations that have been running Maximo Health for at least a year will have the best results, as the models benefit from longer training periods and more failure examples.

Maximo Assistant: Natural Language for Field and Office

The Maximo Assistant brings conversational AI to both the desktop and mobile Maximo experiences. Powered by watsonx, the assistant enables users to interact with Maximo using natural language, reducing the need to navigate complex menus and remember transaction codes. This is particularly valuable for infrequent users who may not remember how to perform specific tasks in the system.

On the desktop, the Maximo Assistant appears as a side panel in the MAS user interface. Users can ask questions like "Show me all open work orders for Building A that are past due" or "What is the current status of pump PUMP-4421?" and the assistant returns the relevant data, formatted appropriately. The assistant can also perform actions: "Create a work order to replace the air filter on HVAC-003" or "Reschedule PM-4421 to next Tuesday." The assistant confirms before executing any action that modifies data, providing a safety net against accidental changes.

On mobile, the Maximo Assistant is integrated into the Maximo Mobile application. Field technicians can use voice or text input to find asset information, review work history, and complete work without navigating through multiple screens. This is particularly valuable for technicians working in challenging environments -- wearing gloves, in noisy areas, or in confined spaces where using a touchscreen is difficult. The voice input capability uses on-device speech recognition, so it works without network connectivity.

The assistant uses a combination of intent recognition, entity extraction, and API orchestration. When a user asks a question, the assistant identifies the intent (query, create, update), extracts the relevant entities (asset number, work type, date, location), and calls the appropriate Maximo REST API. The response is formatted as natural language with the relevant data. The assistant also supports follow-up questions, maintaining context across the conversation.

User: "What's the maintenance history for pump PUMP-4421?"

Assistant: "Pump PUMP-4421 (Building A, Cooling System) has the following maintenance history:

- Last PM: June 28, 2026 (Lubrication and inspection) - Completed
- Last CM: March 15, 2026 (Bearing replacement) - Completed
- Next scheduled PM: July 28, 2026
- Current condition: FAIR (Condition Insight score: 72)
- Open work orders: 1 (WO-2026-00421 - Vibration inspection, due July 21)

Would you like me to show the detailed work order history or create a new work order?"

User: "Show me the last CM details."

Assistant: "Corrective Maintenance on PUMP-4421 (March 15, 2026):

Work Order: WO-2026-00842
Description: Replace drive-end bearing - high vibration detected
Labor Hours: 3.5
Parts Used: Bearing SKF-6205 (1 unit), Grease Shell Gadus S2 V220 (0.5 kg)
Total Cost: $1,245
Root Cause: Bearing fatigue due to misalignment (corrected during repair)
Technician: M. Rodriguez"

The Maximo Assistant is available in MAS 9.2 for both desktop and mobile. It supports English initially, with additional languages planned for future releases. Organizations can also configure custom intents and responses for their specific business processes using the assistant configuration toolkit.

Agentic Workflows and the MCP Server

MAS 9.2 introduces agentic AI capabilities that go beyond simple question-answering. The MCP (Model Context Protocol) Server enables organizations to connect AI agents directly to Maximo, allowing them to orchestrate multi-step workflows across systems. This is the most architecturally significant AI capability in the release because it changes how automation is designed and deployed.

The MCP Server provides a standardized interface for AI agents to discover and interact with Maximo's capabilities. An agent can look up available object structures, inspect their schemas, query live data, and execute actions -- all through a governed, auditable interface. This enables patterns that were previously impossible without custom development:

• Automated root cause analysis: When a critical asset fails, an AI agent queries Condition Insight for the failure analysis, checks the work order history for similar failures, reviews the OEM manual for known issues, and generates a root cause report with recommended corrective actions. This process, which might take a human analyst several hours, is completed in seconds.
• Cross-system orchestration: An AI agent monitors asset health across Maximo, SAP, and an IoT platform. When it detects a deteriorating condition, it creates a work order in Maximo, checks inventory in SAP for required parts, and schedules the technician through the field service management module. The agent handles the coordination between systems that would otherwise require manual intervention.
• Compliance monitoring: An AI agent continuously monitors regulatory compliance requirements, checks asset inspection records against those requirements, and alerts the compliance team when gaps are identified. The agent can also generate compliance reports and submit them to regulatory systems.

The governed MCP Server framework includes user and role management, controlled access to tools, and audit logging. This ensures that AI agents operate within the same security boundaries as human users. Every action taken by an AI agent is logged and attributable, which is essential for regulated industries.

// Example: AI agent using MCP Server to investigate an asset failure
// This is a conceptual example of how an agent orchestrates multiple tools

// Step 1: Get the asset details
const asset = await mcp.callTool("get_asset", { assetnum: "PUMP-4421" });

// Step 2: Check Condition Insight for recent analysis
const condition = await mcp.callTool("get_condition_insight", { 
  assetId: "PUMP-4421" 
});

// Step 3: Query work order history for similar failures
const woHistory = await mcp.callTool("query_work_orders", {
  query: `assetnum="PUMP-4421" and worktype="CM" and status="COMP"`,
  orderby: "statusdate DESC",
  limit: 10
});

// Step 4: Generate root cause analysis
const analysis = await mcp.callTool("analyze_failure_pattern", {
  assetId: "PUMP-4421",
  conditionData: condition,
  workOrderHistory: woHistory
});

// Step 5: Create corrective work order
const newWO = await mcp.callTool("create_work_order", {
  description: analysis.recommendedAction,
  assetnum: "PUMP-4421",
  worktype: "CM",
  priority: analysis.priority,
  estimatedDuration: analysis.estimatedHours
});

The MCP Server is available as an open-source npm package (maximo-mcp-server) and as a governed enterprise deployment on OpenShift. Organizations can start with the open-source version for development and testing, then deploy the governed version for production. The open-source version was created by Markus van Kempen and is available on GitHub and npm.

Document Intelligence with RAG

One of the most practical AI capabilities in MAS 9.2 is document intelligence using retrieval-augmented generation (RAG). Maintenance teams spend a significant amount of time searching through OEM manuals, inspection procedures, safety documents, and technical specifications. RAG makes this information instantly accessible through natural language queries, eliminating the need to search through file shares or paper binders.

The document intelligence capability works by ingesting technical documents into a vector database, where they are indexed by semantic content. When a user asks a question, the system retrieves the most relevant document passages and uses a large language model to generate a concise answer with citations to the source documents. This approach has two key advantages over traditional search: it understands the meaning of the question, not just the keywords, and it provides a synthesized answer rather than a list of document links.

This is particularly valuable for field technicians who need quick access to technical information. Instead of carrying paper manuals or searching through file shares, they can ask the Maximo Assistant: "What is the torque specification for the mounting bolts on pump PUMP-4421?" and receive the answer with a reference to the specific page in the OEM manual. The system can also provide step-by-step procedures, safety warnings, and troubleshooting guides.

User: "What are the safety procedures for replacing the bearing on pump PUMP-4421?"

Assistant: "Based on the OEM manual (Grundfos CR 45-3, Rev 4, Section 5.2):

Safety procedures for bearing replacement on PUMP-4421:

1. Lockout/tagout the electrical disconnect switch (located at panel P-12)
2. Close isolation valves V-101 and V-102
3. Depressurize the system using drain valve DV-03
4. Verify zero energy with a multimeter at terminals T1-T3
5. Allow pump housing to cool below 40C before opening
6. Use PPE: safety glasses, cut-resistant gloves, steel-toed boots
7. Have spill containment kit available (coolant capacity: 3 gallons)

Would you like me to show the step-by-step replacement procedure?"

The document ingestion process supports PDF, Word, HTML, and plain text formats. Documents can be uploaded through the MAS UI or ingested automatically from configured file shares and document management systems. The system supports multiple languages and can handle documents up to 500 pages in length. Organizations can also configure which documents are available to which users, ensuring that sensitive technical information is only accessible to authorized personnel.

AI-Powered Visual Inspection

Maximo Visual Inspection brings computer vision to asset inspection workflows. Using AI models that run locally on mobile devices, technicians can capture images of assets during inspections and receive immediate feedback on their condition. This capability transforms inspections from subjective, paper-based processes to objective, data-driven assessments.

The visual inspection capability supports two modes. In guided inspection mode, the system shows the technician exactly where to position the camera and what to look for, based on the inspection template for that asset type. This ensures consistency across inspections and reduces the training time for new technicians. In analysis mode, the AI model analyzes the captured image for defects -- cracks, corrosion, leaks, misalignment, wear patterns -- and records the findings directly in the inspection record.

The AI models run locally on the device using on-device inference, which means inspections work even in areas with limited or no network connectivity. The results are synchronized to Maximo when connectivity is restored. This is critical for field inspections in remote locations, offshore platforms, underground facilities, and other environments where connectivity cannot be guaranteed. The on-device models are optimized for mobile hardware and can analyze an image in under two seconds.

# Example: Visual inspection result payload
{
  "inspectionId": "VI-2026-07421",
  "assetnum": "PUMP-4421",
  "inspectionDate": "2026-07-14T09:15:00Z",
  "inspector": "jdoe",
  "images": [
    {
      "imageId": "IMG-001",
      "view": "DRIVE_END_BEARING",
      "aiAnalysis": {
        "defects": [
          {
            "type": "CORROSION",
            "severity": "MINOR",
            "confidence": 0.94,
            "location": "Bearing housing, 2 o'clock position",
            "area": "12mm x 8mm"
          }
        ],
        "overallAssessment": "PASS_WITH_OBSERVATIONS",
        "recommendedAction": "Monitor corrosion spot on bearing housing. Schedule detailed inspection during next PM cycle."
      }
    }
  ]
}

The visual inspection models are pre-trained on common asset types and defect categories, and organizations can train custom models using their own image datasets. The training process uses transfer learning, so even small datasets (50-100 images per defect category) can produce usable models. IBM provides a model training tool within MAS that guides users through the process of uploading images, labeling defects, and training custom models.

Practical Implications

The AI capabilities in MAS 9.2 represent a significant step forward, but they require preparation. The most important prerequisite is data quality. AI models are only as good as the data they are trained on, and organizations with poor data quality will see poor results. Before deploying any AI capability, organizations should audit their asset data, work order history, and inspection records for completeness and accuracy. A data quality score of less than 80% will significantly degrade AI model performance.

The second prerequisite is connectivity. While some capabilities (visual inspection, mobile assistant) work offline, the core AI capabilities require connectivity to the MAS instance and the watsonx AI services. Organizations with limited connectivity at their field sites should plan for hybrid deployment models where AI processing happens on the edge and results are synchronized when connectivity is available.

The third consideration is change management. AI capabilities change how maintenance teams work, and not everyone will embrace the change. Organizations should plan for training, pilot programs, and phased rollouts. Start with one capability (Condition Insight is a good choice) in advisory mode, prove the value, and expand from there. The most successful AI deployments in Maximo are those where the maintenance team is involved from the beginning and sees the AI as a tool that makes their job easier, not a threat to their expertise.

Bottom Line

Maximo Application Suite 9.2 delivers on the promise of AI in enterprise asset management. Condition Insight makes predictive maintenance practical and explainable. The Maximo Assistant brings natural language interaction to both desktop and mobile. The MCP Server enables agentic workflows that orchestrate across systems. Document intelligence with RAG puts technical information at technicians' fingertips. And visual inspection brings computer vision to the field. These capabilities are not theoretical -- they are shipping now, and organizations that prepare their data, infrastructure, and teams will be well-positioned to take advantage of them. The era of AI in asset management is here, and it is built into Maximo.