AI Inside Maximo: How MAS 9.2 Operationalizes Intelligence for Asset Management

Artificial intelligence in asset management has moved from buzzword to production capability. IBM Maximo Application Suite 9.2, released in June 2026, represents the most significant AI integration in the platform's history. Rather than treating AI as a separate layer that sits on top of Maximo, version 9.2 embeds AI directly into the daily workflows that reliability, maintenance, field service, safety, and operations teams use every day.

This is not AI for its own sake. The release addresses a specific problem that asset-intensive organizations face: they already have large volumes of operational data, but using that data consistently across teams and processes remains a challenge. Reliability teams need to spot changes before they become failures. Technicians need access to information at the point of work. Safety leaders need better ways to connect reporting and corrective action. Operations and IT teams need systems that work together without constant manual coordination.

This article examines the AI capabilities in MAS 9.2, how they fit into existing maintenance workflows, and what organizations need to know to deploy them effectively. It covers Maximo Condition Insight, Maximo Predict, Maximo Visual Inspection, the mobile AI assistant, agentic workflows, and the practical requirements for getting AI to work in a production asset management environment.

Maximo Condition Insight: From Data to Action

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

Condition-based maintenance has traditionally required significant data analysis by experienced specialists. A reliability engineer would need to pull work order history, review inspection results, analyze meter readings, cross-reference failure mode and effects analysis (FMEA) data, and synthesize all of this into a condition assessment. This process might take hours per asset, which means it typically happens only for the most critical equipment. Condition Insight automates this analysis, reducing the effort required to interpret large volumes of asset data and making condition-based maintenance accessible to a broader range of assets.

Powered by IBM watsonx, Condition Insight evaluates work orders, metrics, time-series data, meter readings, FMEA, and alerts to evaluate asset condition. It then produces a clear, explainable summary of condition, trends, and recommended actions. The output is communicated in plain language, not in model scores or probability percentages that require interpretation. This explainability is critical for adoption: maintenance teams are more likely to act on recommendations they can understand than on opaque model outputs.

How Condition Insight Works in Practice

The workflow begins with data that Maximo already collects. No additional sensors, data sources, or integration projects are required for the base functionality. Condition Insight analyzes:

• Work order history: Failure codes, root causes, repair frequency, and mean time between failures (MTBF) patterns
• Inspection results: Checklist responses, measurement readings, and pass/fail outcomes
• Meter readings: Operating hours, cycles, production counts, or any numeric meter configured on the asset
• Reliability strategies: FMEA data, failure modes, and established asset criticality ratings
• Alerts: Condition monitoring thresholds, anomaly detections, and external system alerts

The AI evaluates this data and produces a condition assessment that includes: current asset condition summary, emerging trends or anomalies, mapped failure modes with likelihood assessment, and recommended corrective actions. The recommendation is specific: it might suggest creating a work order for a particular type of inspection, adjusting a PM frequency, or monitoring a specific parameter more closely.

Conversational Queries via Maximo AI Assistant

Condition Insight integrates with the Maximo AI Assistant, allowing users to query asset condition using natural language. A reliability engineer can ask: "What is the condition of Pump 1001?" and receive a plain-language response summarizing the asset's health, recent trends, and recommended actions. This conversational interface lowers the barrier to using condition data, because users do not need to navigate through multiple screens or build complex queries to get answers.

The conversational interface also supports follow-up questions. After asking about Pump 1001's condition, the engineer can ask: "What about other pumps in the same facility?" The assistant can compare condition scores across the asset class, identify patterns, and highlight outliers. This turns condition monitoring from a one-asset-at-a-time activity into a fleet-level analysis that would be impractical to perform manually.

Maximo Predict: Two Paths to Failure Prediction

Maximo Predict works along two complementary paths, each suited to different data scenarios. Understanding both paths is essential for setting realistic expectations about what Predict can deliver.

Path 1: Historical Failure Prediction

The first path is prediction based on historical failure data. By analyzing corrective maintenance work orders, failure codes, root causes, repair frequency, and MTBF patterns, Predict learns how assets have failed in the past. This works especially well for repetitive assets and known failure modes. It helps answer questions such as: which assets are statistically most likely to fail next, and which failure patterns tend to repeat under similar conditions.

This approach is powerful, but only when failure data is consistent and meaningful. Organizations with poor failure coding practices (generic codes like "broken" or "failed" instead of specific codes like "bearing failure" or "seal leak") will get poor predictions because the AI has nothing meaningful to learn from. Data quality is the foundation.

The historical prediction path also works best with a minimum volume of failure data. An asset class that has experienced 30+ failures with consistent failure coding provides enough data for the model to identify patterns. An asset class with only 3 or 4 failures in the historical record is too sparse for reliable prediction.

Path 2: Condition-Based Prediction

The second path is prediction based on condition and sensor data. Here, Predict learns from vibration, temperature, pressure, runtime, and other condition indicators, combined with degradation trends derived from Maximo Health. Instead of asking what failed before, this approach asks what behavior usually precedes a failure. This is where early detection becomes possible, long before thresholds are crossed.

The condition-based path requires integration with sensor data sources. This typically means connecting Maximo Monitor to IoT sensors, process historians (OSIsoft PI, Honeywell PHD), or SCADA systems. The data pipeline must be reliable and consistent, because gaps in sensor data reduce the model's accuracy.

The combination of both paths provides the most robust prediction. Historical failure data tells the model what to expect. Condition data tells the model what is actually happening. Together, they produce remaining useful life (RUL) estimates that maintenance planners can act on with confidence.

# Conceptual example: Accessing prediction data via MAS REST API
import requests

MAS_BASE_URL = "https://mas-host.company.com/maximo/oslc/os" API_KEY = "your-api-key-here" headers = {"apikey": API_KEY, "Content-Type": "application/json"}

if response.status_code == 200: data = response.json() asset = data.get('member', [{}])[0] print(f"Asset: {asset.get('assetnum')}") print(f"Health Score: {asset.get('healthscore')}") print(f"Prediction Score: {asset.get('predictedscore')}") print(f"Predicted Failure: {asset.get('predictedfailuredate')}") # Use prediction data to prioritize maintenance scheduling if asset.get('predictedscore', 0) > 0.7: print("ACTION: High failure probability - schedule inspection immediately") elif asset.get('predictedscore', 0) > 0.4: print("ACTION: Moderate risk - review at next planning cycle") ```

Maximo Visual Inspection: AI at the Edge

Maximo Visual Inspection (MVI) enables AI-based visual inspection with local inference directly on the device. This is a significant advancement for field operations where network connectivity is unreliable or where inspection images need immediate analysis.

MVI uses computer vision models trained on images of asset conditions: corrosion, cracks, leaks, misalignments, or any visual defect that can be captured with a camera. The models run on the mobile device itself, meaning the inspection result is available in real time without sending images to a cloud service for processing. This edge inference capability eliminates the latency and connectivity requirements that previously limited AI-based inspection in remote locations.

The workflow is straightforward:

• A technician captures a photo of an asset using a mobile device
• The MVI model running locally analyzes the image
• The model identifies defects and classifies severity
• Results are stored in the work order inspection record
• If defects are found, a follow-up work order can be automatically generated

A documented case study from the nybl and IBM partnership showed a 50% reduction in inspection costs and safety incidents, a 20% decline in outage and emergency repair costs, and a 20% improvement in grid uptime. These results come from pilots that integrated nVision AI inspection with Maximo Visual Inspection and watsonx.governance, demonstrating that visual inspection AI is ready for production deployment.

Training Visual Inspection Models

MVI models are trained through the Maximo Visual Inspection application, which provides a no-code interface for uploading training images, labeling defects, and training models. The platform supports multiple model architectures and automatically selects the appropriate one based on the training data. Organizations can start with as few as 50-100 labeled images per defect type, though 200+ images produce more reliable models.

The training process involves uploading images, drawing bounding boxes around defects, assigning labels (corrosion-light, corrosion-moderate, corrosion-severe), and training the model. The platform handles data augmentation, model training, and validation automatically. Once trained, the model can be deployed to mobile devices for field use. Models can be retrained as new defect types are identified or as image quality improves.

Maximo Assistant on Mobile: Natural Language in the Field

Maximo Assistant on Mobile helps technicians use natural language to find asset information, review history, and complete work efficiently in the field. This capability puts the knowledge of your best technicians, your documentation, and your asset history in the palm of every worker's hand.

A technician arriving at a pump that has failed can ask the assistant: "What was the last repair done on this asset?" The assistant queries Maximo's work order history and responds with a summary of recent maintenance, including failure codes, parts replaced, and labor hours. The technician can then ask: "What are the common failure modes for this type of pump?" and receive a response based on the asset's failure history and FMEA data.

This conversational access to asset information reduces the time technicians spend navigating through Maximo's interface on a mobile device. Instead of drilling through asset records, work order history tabs, and failure code lookups, the technician asks a question and gets an answer. The result is improved first-time fix rates, better SLA performance, reduced assignment time, and increased technician productivity.

The assistant also supports voice input, which is critical for technicians wearing gloves or working in environments where typing on a phone is impractical. A technician on a catwalk above a refinery can verbally ask: "Show me the last inspection results for this vessel" and receive a spoken summary without needing to type anything.

AI-Enabled Conversational Scheduling and What-If Analysis

MAS 9.2 introduces AI-enabled conversational scheduling and what-if analysis that empowers planners, schedulers, and field service managers to explore changes using plain language. A scheduler can ask: "What happens if I add two more technicians to next week's schedule?" or "What if I prioritize all critical work orders first?"

The AI analyzes the current schedule, resource availability, and work order constraints to model the impact of the proposed change. It provides a data-driven response that includes projected completion dates, resource utilization rates, and potential scheduling conflicts. This allows schedulers to make informed decisions quickly, without manually reworking the entire schedule to test a scenario.

This capability is particularly valuable in dynamic environments where priorities shift frequently. A storm response scenario might require reassigning 50 crews from planned PMs to emergency restoration work. The conversational scheduling AI can model this reallocation in seconds, showing the scheduler exactly which PMs will be delayed and by how much. The scheduler can then make an informed decision about which PMs to defer and which to protect.

Agentic Workflows and the MCP Server

MAS 9.2 introduces agentic workflows designed to help guide decisions and move work forward across processes in practical, operationally-grounded ways. This is where AI moves from providing recommendations to taking action.

The MCP Server (Model Context Protocol) is a key enabler. It allows organizations to bring their own agents and integrate them directly with Maximo Manage APIs, enabling AI to participate more seamlessly in existing operational processes without relying on manual coordination between disconnected tools.

An agentic workflow in Maximo might look like this:

• Condition Insight detects an emerging anomaly on a critical transformer
• The agent evaluates the anomaly against reliability strategies and historical patterns
• The agent creates a work order with appropriate priority, labor estimates, and parts
• The agent schedules the work order considering crew availability and asset criticality
• The agent notifies the reliability engineer with a summary of the decision and its rationale

Each step is auditable and reversible. The agent does not act in a black box. Every decision is logged with the data that informed it, and human reviewers can override any agent decision at any point. This governance model is essential for organizations that need to maintain compliance and audit trails for regulatory purposes.

AI for Safety and Compliance

MAS 9.2 also brings AI into safety and compliance workflows. Expanded capabilities across asset-centric waste management, contractor safety oversight, and mobile-first safety workflows make it easier to manage compliance, track risk, and act in real time. Field teams can capture incidents, complete inspections, and initiate permit-to-work processes directly on a mobile device, improving data accuracy and responsiveness.

AI-assisted incident classification helps improve the consistency and completeness of reporting by suggesting categories and identifying similar events. This makes it easier to detect patterns and strengthen overall data quality. When an incident is reported, the AI analyzes the description, compares it to historical incidents, and suggests the appropriate classification. This reduces the variability that comes from having different people categorize incidents differently.

Additionally, AI-enabled document abstraction uses retrieval-augmented generation (RAG) to extract key information from leases, OEM maintenance manuals, and technical documentation. This turns unstructured documents into structured, searchable data that Maximo can use. Instead of a technician searching through a 200-page OEM manual to find the recommended torque specification for a particular bolt, the AI extracts the relevant information and presents it in context.

Data Quality: The Foundation of AI in Maximo

Every AI capability described in this article depends on data quality. This is not a caveat or a footnote. It is the single most important factor in whether AI in Maximo will deliver value.

Maximo Predict needs consistent failure codes. If your organization uses generic failure codes like "broken" or "failed" instead of specific codes like "bearing failure" or "seal leak," the prediction models have nothing meaningful to learn from. Maximo Condition Insight needs consistent inspection data. If inspection checklists are completed inconsistently, with some technicians marking everything as "OK" without actually checking, the condition assessments will be unreliable. Maximo Visual Inspection needs well-labeled training images. A model trained on 50 images of surface corrosion will not recognize structural cracks.

Organizations deploying AI in Maximo should invest in data quality before deploying AI capabilities. This means: auditing failure code usage and standardizing failure coding practices, reviewing inspection checklist designs to ensure they capture meaningful data, cleaning asset hierarchies to ensure assets are properly categorized, and training maintenance teams on the importance of consistent data entry.

As Matt Boehne from IBM has noted in presentations on AI in Maximo: the AI is not magically figuring things out from nothing. It is working with the data we provide. Poor data in means poor insights out. One of the most effective ways to improve data quality is to mobilize your workforce with the full power of Maximo in their hands, so they record specific detail at the point of work rather than reconstructing it from memory at a desktop computer later.

Practical Implications

For organizations evaluating AI capabilities in Maximo, several factors determine time-to-value. First, Maximo Condition Insight requires no additional sensors or data sources for base functionality. It works with data Maximo already collects. This makes it the fastest AI capability to deploy, and the best starting point for organizations new to AI in asset management. Second, Maximo Predict's value is directly proportional to the quality of historical failure data. Organizations with years of well-coded work order history will see predictions within weeks. Organizations with poor failure coding will need to fix their data practices first, which can take months. Third, Maximo Visual Inspection is immediately deployable for common defect types (corrosion, cracks, leaks) with minimal training data. Custom defect types require more training images but the no-code training interface makes the process accessible to non-data-scientists. Fourth, the agentic workflows and MCP Server represent the frontier of AI in Maximo. They are powerful but require careful governance. Start with human-in-the-loop workflows where the agent recommends and a human approves before moving to fully autonomous agent actions. Finally, AI in Maximo is not a replacement for maintenance expertise. It is a multiplier. The best results come from organizations that combine AI capabilities with experienced reliability engineers who can validate and contextualize the AI's recommendations.

Bottom Line

MAS 9.2 makes AI practical for asset management. Condition Insight, Predict, Visual Inspection, and the mobile assistant are production-ready capabilities that work with data Maximo already collects. The agentic workflows and MCP Server provide a path to more autonomous AI operations for organizations ready to take that step. The technology is real and the results are documented. The question is not whether AI works in Maximo. It does. The question is whether your data is ready for it. Organizations with clean failure codes, consistent inspection data, and well-maintained asset hierarchies will see value quickly. Organizations with poor data quality will need to invest in data remediation first, but that investment pays dividends regardless of AI adoption. Start with Condition Insight. It requires the least data preparation and provides the most immediate value. Then expand to Predict, Visual Inspection, and agentic workflows as your data maturity grows.