AI Inside Maximo: How MAS 9.2 Turns Predictive Maintenance Into Daily Practice

Artificial intelligence in enterprise asset management has been promised for a decade. For most of that decade, AI in Maximo meant a separate analytics module that required a data science team, a pilot project, and a lot of patience. MAS 9.2, announced in June 2026, changes that equation. AI is no longer a separate module you opt into. It is embedded in the daily workflows that reliability, maintenance, field service, safety, and operations teams already use.

This is not a marketing claim. MAS 9.2 expands AI across reliability insights, field execution, safety and compliance workflows, document-based information extraction, and orchestration across systems. It introduces agentic workflows designed to help guide decisions and move work forward across these processes in practical, operationally grounded ways. The AI is built for asset management, not bolted on from a generic LLM.

As IBM stated in its announcement: "IBM is announcing Maximo Application Suite 9.2, a release that brings asset-first AI directly into the workflows that reliability, maintenance, field service, safety and operations teams use every day." The emphasis is on operationalizing AI, not experimenting with it. This article breaks down what each AI capability in MAS 9.2 actually does, how it works with your existing Maximo data, and what you need to do to prepare for it.

Maximo Condition Insight: Pattern Recognition for Asset Behavior

Maximo Condition Insight is the centerpiece of MAS 9.2's AI strategy. It brings together work orders, inspections, meter readings, and reliability strategies to identify patterns in asset behavior and recommend what to do next. This helps teams diagnose issues earlier, make more consistent decisions, and reduce dependence on fragmented systems or individual expertise alone.

The practical problem Condition Insight solves is this: maintenance teams are drowning in data but starved for insight. A typical Maximo instance has years of work order history, thousands of meter readings, hundreds of inspection forms, and dozens of reliability strategies. No human can synthesize all of that data for a single asset, let alone across hundreds or thousands of assets. Condition Insight does the synthesis automatically.

How it works:

Work Order History  ---|
Meter Readings       ---|---> Condition Insight Engine ---> Pattern Identification
Inspection Results   ---|                                    |
Reliability Strategy ---|                                    +--> Recommendation
                                                                    |
                                                                    +--> Work Order Generation
                                                                    +--> Alert Notification
                                                                    +--> Strategy Adjustment

Condition Insight analyzes the correlation between different data sources. For example, it might identify that a particular pump model fails within 30 days of a specific vibration pattern appearing in meter readings, combined with a work order history showing recurring seal replacements. That pattern, once identified, becomes a predictive signal. The next time the vibration pattern appears, Condition Insight flags the asset and recommends a seal replacement before the failure occurs.

The recommendations are not just alerts. They are actionable suggestions tied to Maximo work flows. A recommendation might suggest creating a work order for inspection, adjusting the preventive maintenance schedule, or escalating the asset to a reliability engineer for deeper analysis. The maintenance team can accept, modify, or dismiss the recommendation, and the system learns from that feedback. This feedback loop is critical: it means the AI becomes more accurate over time as your team uses it, adapting to your specific operational context rather than relying on generic models.

To prepare for Condition Insight:

• Clean your work order history. Incomplete or inaccurate work order records will produce unreliable patterns. Ensure that failure codes, asset hierarchies, and completion notes are accurate and consistent. This means auditing your last 24 months of work orders at minimum.
• Activate meter readings. If you have sensors but are not recording readings in Maximo, Condition Insight cannot use them. Connect your IoT data to Maximo meter readings before enabling the feature. Even manual meter readings are valuable if they are recorded consistently.
• Standardize inspection forms. Consistent inspection data with structured responses (not free text) gives Condition Insight more reliable signals to work with. Convert free-text inspection fields to pick lists where possible.
• Review your reliability strategies. Condition Insight uses your existing reliability strategies as context. If your strategies are outdated or do not reflect actual failure patterns, the AI recommendations will be built on a flawed foundation.

Maximo Assistant on Mobile: Natural Language for Field Technicians

Maximo Assistant on Mobile helps technicians use natural language to find asset information, review history, and complete work efficiently in the field. Instead of navigating through multiple screens to find an asset's maintenance history, a technician can simply ask: "What work orders were completed on pump P-101 in the last 30 days?"

The Assistant is powered by a large language model (LLM) that has been tuned for Maximo's domain. It understands asset management terminology, Maximo's data model, and the context of maintenance work. It is not a generic chatbot. It is a domain-specific assistant that knows the difference between a work order and a purchase order, and it understands the relationships between assets, locations, and work types.

Practical use cases for the Assistant include:

• Asset lookup: "Show me the specs for motor M-204 at the north plant." The Assistant retrieves the asset record, specifications, and recent maintenance history in a single response.
• Work order context: "What were the findings from the last inspection on this asset?" The Assistant pulls inspection results and notes from prior work orders, saving the technician from navigating through multiple screens.
• Procedure guidance: "What are the safety procedures for working on this equipment?" The Assistant retrieves the relevant safety plans and job plans associated with the asset.
• Failure history: "How many times has this pump failed in the past year and what were the failure codes?" The Assistant compiles failure history and identifies recurring patterns.
• Parts availability: "Do we have replacement seals in stock for this pump?" The Assistant checks inventory records and storeroom availability.

The Assistant is being updated with Granite 4.0, IBM's foundation model, which will act as a conductor orchestrating specialized AI agents. This means the Assistant will increasingly be able to not just retrieve information but coordinate actions across the Maximo platform. For example, a technician could ask the Assistant to create a follow-up work order for a defect found during an inspection, and the Assistant would route that request to the appropriate agent.

The impact on field maintenance is significant. Technicians spend less time navigating the system and more time doing the actual maintenance work. First-time fix rates improve because technicians have the context they need before they start the job. And the knowledge of experienced technicians is captured in the system through the Assistant's natural language interface, making it accessible to less experienced team members who might not know which screens to navigate or which reports to run.

Maximo Visual Inspection: AI-Powered Visual Quality Checks

Maximo Visual Inspection enables AI-based visual inspection with local inference directly on the device. This is not cloud-dependent AI. The model runs on the mobile device, which means it works in environments with poor connectivity and produces results in real time. The latency advantage of local inference is critical in field environments where waiting for a cloud API response could mean the difference between completing an inspection during a maintenance window or missing it entirely.

The use cases for visual inspection in asset management are extensive:

• Corrosion detection: A technician photographs a pipe or tank, and the AI model identifies corrosion levels and classifies them by severity. The model can distinguish between surface rust and structural corrosion, prioritizing follow-up actions accordingly.
• Wear pattern analysis: Photograph a bearing or gear, and the model compares the wear pattern against a library of known failure modes. This is particularly valuable for equipment where visual inspection is the primary diagnostic tool.
• Safety equipment verification: Photograph a lockout/tagout setup, and the model verifies that all required safety equipment is in place before work begins. This creates a photographic record of compliance that can be stored with the work order.
• Inspection form automation: Photograph an asset nameplate, and the model extracts serial number, model, and manufacturer data directly into the inspection form using optical character recognition combined with AI classification.

The local inference capability is what makes this practical for field use. A technician in a remote substation or an offshore platform cannot wait for a cloud API call to classify an image. With local inference, the result is available in seconds, not minutes. And the model continues to learn: images classified by technicians are fed back into the training pipeline to improve model accuracy over time.

To implement Visual Inspection effectively:

• Build a labeled image library for your most common inspection types. The model needs training data specific to your assets and inspection standards. Start with 200-500 labeled images per inspection type for initial training.
• Start with a single inspection type (e.g., corrosion on storage tanks) and expand from there. Trying to train models for every inspection type simultaneously will overwhelm your team and produce lower-quality models.
• Validate model predictions against human inspections for the first 90 days. This builds confidence in the model and identifies edge cases where the model needs retraining. Track the false positive and false negative rates to measure model quality.
• Integrate with inspection forms so that model predictions automatically populate inspection fields, reducing manual data entry. This is where the productivity gains compound: the technician photographs the asset, the model classifies the condition, and the form is pre-populated.
• Retrain quarterly with new images from the field. As your inspection program grows, the model should incorporate new examples, especially edge cases and unusual conditions that were not in the initial training set.

Predictive Maintenance with Maximo Predict: The Foundation

Maximo Predict is the application that started the AI journey in Maximo, and it remains the foundation for everything MAS 9.2 builds on. Predict uses AI and machine learning to analyze historical maintenance records, operational data, inspection reports, and environmental data to predict downtime, degradation, and failures.

The workflow for implementing Predict has been well-documented:

• Create asset groups based on asset type, criticality, or operational context. Predict works best when analyzing groups of similar assets rather than individual assets in isolation. A group might be "all centrifugal pumps at the north plant" or "all transformers rated above 500 kVA."
• Train predictive models using the default notebooks provided with Predict. Your data scientist works with the group ID and uses Jupyter notebooks to build and train model instances. The notebooks handle the common algorithms; your data scientist focuses on feature engineering and model selection.
• Deploy the trained model so that it generates predictions for each asset in the group. Predictions include current failure probability and estimated failure date.
• Monitor predictions through the Predictions section in Maximo, which shows failure probability, estimated failure date, and recommended actions for each asset.
• Use work queues to track assets with high failure probability or assets predicted to fail before the next scheduled PM work order. These work queues become the daily planning tool for reliability engineers.

The predictive models in MAS 9.2 benefit from continuous learning. Every action, alert, or failure makes the system smarter and more accurate over time. This is a significant advancement over the static models in earlier versions, which required manual retraining to incorporate new data. With continuous learning, the model adapts to changes in operating conditions, maintenance practices, and asset populations without requiring a manual retraining cycle.

A key insight from practitioners: the quality of your historical data determines the quality of your predictions. If your work order history has inconsistent failure codes, missing asset hierarchies, or incomplete completion notes, your predictive models will be unreliable. Data quality is not a prerequisite for implementing Predict. It is the foundation. Before training any model, conduct a data quality audit covering at least the last 24 months of work orders. Standardize failure codes, verify asset hierarchies, and ensure that completion notes contain enough detail to distinguish between actual failures and routine replacements.

The shift from reactive to proactive maintenance using Predict follows a maturity curve:

Stage 1: Reactive       ---> Fix when it breaks (no AI needed)
Stage 2: Preventive     ---> Fix on a schedule (traditional Maximo PM)
Stage 3: Condition-based ---> Fix when condition indicators trigger (Maximo Health)
Stage 4: Predictive     ---> Fix before it fails (Maximo Predict)
Stage 5: Prescriptive   ---> AI recommends the optimal action (MAS 9.2 Condition Insight)

Most Maximo organizations are at Stage 2 or 3. MAS 9.2's AI capabilities are designed to move them to Stage 4 and 5 without requiring a separate data science team. The continuous learning capability in MAS 9.2 means that even organizations with imperfect data can start at Stage 4 and improve over time as the model learns from real outcomes.

Agentic Workflows: The Next Step in Operational AI

MAS 9.2 introduces agentic workflows, a concept that goes beyond predictions and recommendations. Agentic workflows are AI agents that can take actions within defined boundaries. They do not just suggest what to do; they execute steps in a process, escalating to humans when they encounter situations outside their training.

Three AI agents are being introduced to the Maximo platform:

• Maximo Assistant: The conversational agent that allows users to query data and coordinate tasks through natural language. It acts as the orchestrator for the other agents, routing requests and synthesizing responses.
• Optimizer Agent: Recommends the optimal time to replace equipment by considering performance metrics and budget constraints. It does not just predict when an asset will fail; it calculates whether it is more cost-effective to repair, replace, or defer based on financial parameters that the organization defines.
• Condition Insights Agent: Automatically collects asset data to estimate current condition and predict potential failures. This agent runs continuously in the background, monitoring asset data streams and flagging issues without requiring manual queries.

A third agent, expected later in 2026, will help plan long-term asset investments. Going beyond the current optimizer, it will enable users to define specific investment criteria such as operating costs, budget constraints, and sustainability goals to guide asset replacement decisions. This agent will be particularly valuable for capital planning teams that need to justify investment decisions with quantitative analysis.

The agentic workflow pattern in practice:

Condition Insights Agent detects anomaly
    |
    +--> Maximo Assistant notifies reliability engineer
              |
              +--> Optimizer Agent calculates replacement vs repair cost
                       |
                       +--> Assistant recommends action with cost justification
                                |
                                +--> Engineer approves
                                         |
                                         +--> Work order auto-generated
                                         +--> Parts reserved
                                         +--> Crew scheduled

This is the vision of operationalized AI in asset management: not a single AI model doing everything, but specialized agents working together to move work forward with human oversight at decision points. The human is still in the loop, but the AI agents handle the data gathering, analysis, and routine actions that previously consumed engineering time.

Practical Implications

For Maximo teams planning their AI adoption strategy, the MAS 9.2 release changes the approach. You no longer need a dedicated data science team to get value from AI in Maximo. The AI is embedded in the workflows your team already uses. But you do need clean data.

The data quality imperative cannot be overstated. Every AI capability in MAS 9.2 depends on the quality of your Maximo data. Work order history, failure codes, asset hierarchies, meter readings, and inspection results are the fuel for the AI engine. If that fuel is contaminated with incomplete records, inconsistent coding, and missing data, the AI will produce unreliable outputs. The teams that benefit most from MAS 9.2's AI capabilities will be the teams that invested in data quality before enabling them.

Start with one AI capability, not all of them. Condition Insight is the highest-value starting point for most organizations because it leverages existing work order and inspection data without requiring new IoT infrastructure. Visual Inspection is the easiest to deploy if you have a well-defined inspection use case and can build a labeled image library. The Assistant is the quickest win for field adoption because it requires no model training, just user adoption. Pick one, demonstrate value, and expand.

Budget for change management, not just technology. The AI capabilities in MAS 9.2 will change how your maintenance team works. Technicians need training on the Assistant. Reliability engineers need to understand how Condition Insight generates recommendations. Managers need to trust the predictions enough to act on them. This human side of AI adoption is where most implementations stall. Plan for it explicitly in your project budget.

Bottom Line

MAS 9.2 makes AI in Maximo a daily practice, not a pilot project. Condition Insight, the mobile Assistant, Visual Inspection, and agentic workflows bring AI into the workflows that maintenance and reliability teams already use. The technology is ready and supported. The question is whether your data and your team are prepared. Invest in data quality first, start with one capability, and build from there. The teams that do this will move from reactive maintenance to AI-driven predictive maintenance within months, not years. The teams that skip data quality will spend years explaining why their AI predictions are unreliable. The difference is preparation, not technology.