AI Inside Maximo: From Predictive Models to Agentic Workflows in MAS 9.2

AI Inside Maximo: From Predictive Models to Agentic Workflows in MAS 9.2

Artificial intelligence in IBM Maximo has evolved from an optional add-on to a core platform capability. With the release of Maximo Application Suite 9.2 in June 2026, AI is no longer a separate layer that sits on top of the EAM system. It is embedded directly into the workflows that reliability, maintenance, field service, safety, and operations teams use every day. This is a fundamental shift from AI as an observer that offers recommendations to AI as a participant that helps guide decisions and move work forward.

The journey from Maximo Predict's early machine learning models to MAS 9.2's agentic workflows represents three distinct phases of AI maturity in asset management. Understanding where we have been, where we are now, and where IBM is headed will help you make informed decisions about which AI capabilities to adopt and when.

This article traces the evolution of AI in Maximo, breaks down each AI capability available in MAS 9.2, explains what the agentic workflow architecture means in practice, and provides a maturity model for organizations planning their AI adoption roadmap. Whether you are on Maximo 7.6 evaluating a migration to MAS or already running MAS 9.1 and considering an upgrade to 9.2, this guide will help you understand the AI landscape and where to invest.

The Evolution: From Predict to Agentic Workflows

AI in Maximo has gone through three distinct phases, each building on the previous one. Each phase has expanded the scope of what AI can do within the platform, from analyzing historical data to participating in real-time operational decisions.

Phase 1: Predictive Models (MAS 8.x to MAS 9.0)

Maximo Predict was the first significant AI capability in the Maximo Application Suite. It uses AI and machine learning to predict asset performance and maintenance needs by analyzing time-series data from Maximo Monitor and failure data from Maximo Manage. Data scientists build and train predictive models using Python notebooks integrated with the platform.

The typical workflow involves creating groups of assets, selecting relevant data sources, training a model on historical failure data, and deploying the model to generate predictions. The predictions populate a Predictions section in Maximo Manage for each asset, showing failure probability, estimated days to failure, and recommended actions.

Maximo Predict supports several model types:

• Failure probability models: Predict the probability that an asset will fail within a specified time window (e.g., 30, 60, 90 days). These models use features like age, loading history, maintenance history, and condition data. The output is a probability score that can be used to prioritize maintenance activities and capital budget allocations.
• Remaining useful life models: Estimate how many days of useful life remain before a failure is expected. These are more granular than probability models and help with maintenance scheduling by providing a specific time horizon for planning. Planners can use this information to sequence maintenance activities and coordinate with operations to find optimal maintenance windows.
• Anomaly detection models: Identify unusual patterns in sensor data that may indicate developing failures. These models do not require historical failure data; they learn what normal operation looks like and flag deviations. This is particularly useful for new assets or assets with limited failure history where supervised learning models cannot be trained.

A key limitation of Phase 1 AI is that it requires data science expertise. Building, training, and deploying predictive models requires Python knowledge and an understanding of machine learning concepts. This limited adoption to organizations with data science resources, which meant that many Maximo customers used Predict only for a small number of critical assets rather than across their full asset base.

Phase 2: AI-Enhanced Workflows (MAS 9.0 to 9.1)

MAS 9.0 introduced Maximo Work Order Intelligence, which uses IBM watsonx generative AI capabilities to speed up work order approval, improve data quality, and provide failure code recommendations. This was the first time AI was applied to the work order process itself, not just to asset condition prediction.

Work Order Intelligence addresses a real pain point: maintenance managers often receive work orders with vague descriptions and missing failure codes. The AI model, trained on work order descriptions, provides recommendations for the most likely problem code. This helps maintenance managers quickly identify the issue, assign the right problem code, allocate technicians, and reduce troubleshooting time.

The generative AI approach is significant because it can enhance insufficient data. Many organizations have years of work order history with incomplete descriptions, missing failure codes, and inconsistent terminology. Work Order Intelligence can analyze this messy data and provide structured recommendations, effectively retroactively improving the quality of historical work order data.

MAS 9.0 also introduced Maximo Reliability Strategies, which provides the ability to analyze failure modes and access a comprehensive library of asset-specific failure details and mitigation activities. This includes the ability to create, import, and modify Failure Mode and Effects Analysis (FMEAs), making it easier to build customized maintenance reliability strategies tailored to specific operational contexts.

Phase 3: Agentic Workflows (MAS 9.2)

MAS 9.2 is where AI in Maximo makes the jump from recommendation to participation. The release introduces agentic workflows designed to help guide decisions and move work forward across reliability, field execution, safety, and compliance processes. IBM's announcement describes this as "asset-first AI" that is built for asset management and keeps operations running.

The key innovation in Phase 3 is the MCP Server, which enables organizations to bring their own AI agents and integrate them directly with Maximo Manage APIs. This means AI agents can participate in operational processes without manual coordination between disconnected tools. The AI is no longer just suggesting what to do; it is doing parts of the work, within controlled boundaries and with full audit logging.

IBM Research has described the vision: agents that consolidate data silos into a holistic assessment of asset health, shift maintenance from regular check-ups to condition-based interventions, and ultimately extend the useful life of expensive multi-year investments. The research team has also indicated that a third agent focused on asset investment planning is on the roadmap, which will go beyond the default optimizer to allow users to set conditions for replacement based on operating costs, budgeting constraints, or sustainability targets.

Maximo Condition Insight: Simplifying Condition-Based Maintenance

One of the headline AI capabilities in MAS 9.2 is Maximo Condition Insight. This brings together work orders, inspections, meter readings, and reliability strategies to identify patterns in asset behavior and recommend what to do next.

Condition-based maintenance has traditionally required significant data analysis by experienced specialists. These specialists need to understand failure modes, interpret condition data, and translate findings into maintenance actions. Maximo Condition Insight simplifies and scales this analysis, reducing complexity and the effort required to interpret large volumes of asset data.

The capability helps teams:

• Diagnose issues earlier: By correlating data across work orders, inspections, and meter readings, Condition Insight can identify developing issues before they become apparent to individual technicians or planners. What a specialist might catch after hours of analysis, the system can flag automatically.
• Make more consistent decisions: Instead of relying on individual expertise, which varies by person and by site, Condition Insight applies the same analytical framework across all assets. This produces more consistent recommendations and reduces the variability that plagues maintenance operations across distributed facilities.
• Reduce dependence on fragmented systems: Rather than checking work order history in one system, inspection results in another, and meter readings in a third, Condition Insight consolidates this data into a single analytical view. The condition assessment is based on all available data, not just what one system can see.

IBM has indicated that the Condition Insights agent will estimate each asset's current condition and projected replacement date as part of a broader move from regular check-ups to condition-based interventions. Only when Maximo detects signs of trouble will the system flag the asset for attention, shifting the maintenance model from time-based to condition-based. This has the potential to significantly reduce unnecessary maintenance activities while ensuring that developing issues are not missed.

Maximo Assistant on Mobile: Natural Language in the Field

MAS 9.2 introduces Maximo Assistant on Mobile, which lets technicians use natural language to find asset information, review history, and complete work efficiently in the field. This is powered by IBM Granite models and represents a significant usability improvement for field technicians.

The mobile assistant addresses a long-standing problem in Maximo implementations: technicians often struggle to navigate the system on mobile devices, and the complexity of the interface discourages adoption. By enabling natural language queries, the assistant lowers the barrier to system use and increases the quality of data captured during maintenance work.

A technician can ask: "What is the work order history for Pump 101?" or "Show me the safety procedures for working on high-voltage equipment." The assistant retrieves the information through the MCP Server and presents it in a conversational format. This is faster than navigating through Maximo's mobile interface and more accessible to technicians who are not Maximo power users.

IBM has confirmed that the agent underpinning Maximo Assistant will be upgraded to a Granite 4.0 model in the near future, which should improve both the accuracy of responses and the range of queries the assistant can handle.

AI-Enabled Visual Inspection with Local Inference

Maximo Visual Inspection has been available for several releases, but MAS 9.2 enhances it with AI-based visual inspection and local inference directly on the device. This means technicians can use their mobile device camera to inspect equipment, and the AI model running on the device can identify defects, wear patterns, or anomalies without requiring a network connection.

Local inference is a significant capability for field operations in remote locations or facilities with poor network connectivity. Previous versions of visual inspection required sending images to a server for processing, which was not practical in environments with intermittent connectivity. With local inference, the AI model runs on the mobile device, providing instant results regardless of network conditions.

Common use cases for AI-based visual inspection include:

• Corrosion detection: Identifying rust or corrosion on pipes, tanks, and structural components before it becomes a safety concern
• Leak identification: Detecting fluid leaks from valves, pumps, and fittings that might be missed during visual inspection
• Wear pattern analysis: Identifying abnormal wear on belts, bearings, and moving parts that could indicate misalignment or lubrication issues
• Safety equipment verification: Confirming that safety barriers, signage, and PPE are in place and compliant with regulatory requirements
• Asset condition documentation: Automatically capturing and categorizing photos for compliance records, reducing manual documentation effort

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. This capability is distinct from the Maximo Assistant on Mobile because it is designed for planning and scheduling workflows, not field execution.

A planner can ask: "What if we increase maintenance capacity by 20% next week?" The AI agent queries the current work order backlog, resource availability, and capacity constraints, then simulates the scenario and returns a comparison of outcomes. This helps planners make informed decisions about resource allocation, overtime, and scheduling priorities.

The conversational scheduling capability optimizes assignments based on real-time conditions, constraints, and resource availability. It considers factors like technician skills, travel time, parts availability, and asset criticality to recommend the most efficient assignment of work. This is faster and more comprehensive than manual scheduling, which typically optimizes for one or two factors at most.

For organizations with complex scheduling requirements (multiple sites, varied skill sets, competing priorities), conversational scheduling can reduce the time spent on schedule creation from hours to minutes. The AI can evaluate thousands of possible combinations and recommend the optimal schedule, which a human planner then reviews and approves.

The AI Maturity Model for Maximo Organizations

Based on the evolution of AI capabilities in Maximo, organizations can assess their current AI maturity and plan their next steps using the following model:

Level 0: Reactive - No AI capabilities in use. Maintenance is purely reactive or time-based. This is the starting point for many Maximo 7.6 organizations that have not yet migrated to MAS.

Level 1: Predictive - Maximo Predict deployed for critical assets. Data scientists building and training models. Predictions integrated into work order backlog prioritization. Requires MAS 9.0 or later and data science resources.

Level 2: AI-Enhanced - Work Order Intelligence active. Failure code recommendations improving data quality. Reliability Strategies and FMEAs standardized across asset classes. Requires MAS 9.0 or later.

Level 3: Condition-Based - Maximo Condition Insight deployed. Condition-based maintenance replacing time-based maintenance for critical assets. Condition assessments automated rather than manual. Requires MAS 9.2.

Level 4: Agentic - MCP Server active with external AI agents. Repair assistants, conversational scheduling, and cross-system orchestration in production. AI agents participating in workflow execution, not just recommendation. Requires MAS 9.2 and AI agent development expertise.

Level 5: Autonomous - Future state where AI agents handle routine maintenance decisions autonomously, with human oversight for exceptions. Asset investment planning agent makes replacement recommendations based on operating costs, budgeting constraints, and sustainability targets. Not yet available but on IBM's roadmap.

Most organizations today are at Level 0 or Level 1. A small number have reached Level 2. MAS 9.2 enables organizations to move to Level 3 and Level 4, but doing so requires investment in data quality, AI agent development, and change management.

Practical Implications

The progression of AI in Maximo from predictive models to agentic workflows has significant implications for how organizations should approach their AI adoption strategy.

First, do not skip the fundamentals. Organizations that want to deploy agentic workflows in MAS 9.2 need to have their data house in order. AI agents that query Maximo through MCP will encounter the same data quality issues that plague every other integration. If your work order failure codes are inconsistent, your asset hierarchy is incomplete, or your meter readings are irregular, AI agents will produce unreliable results. The path to Level 4 starts with the data quality work required for Level 1.

Second, start with one use case. The Maximo Guys recommend starting with the use case that scores highest across four dimensions: data availability, failure frequency, detectable patterns, and actionable outcomes. This advice applies equally to agentic workflows. The repair assistant pattern is a strong starting point because it delivers immediate value to technicians while exercising the full MCP capability set: resource retrieval, tool invocation, and contextual prompts.

Third, plan for skills development. AI in Maximo is no longer just a data science exercise. With agentic workflows, you need AI agent development skills, MCP integration knowledge, and prompt engineering expertise. These are different skill sets from traditional Maximo administration, and most organizations will need to invest in training or hire new talent.

Fourth, governance matters more than ever. When AI agents can create work orders, update statuses, and modify data through MCP, the governance framework needs to keep pace. Define clear policies for what AI agents are allowed to do, what requires human approval, and what is prohibited. Start with read-only agents in production, and enable write capabilities only after thorough testing.

Bottom Line

MAS 9.2 represents the most significant AI advancement in Maximo's history. By embedding AI across reliability insights, field execution, safety workflows, document processing, and system orchestration, IBM has moved Maximo from an EAM platform with AI features to an AI-native asset management platform. The introduction of agentic workflows through the MCP Server opens possibilities that were not practical with previous releases.

The organizations that will benefit most from these capabilities are not necessarily the ones with the largest AI budgets. They are the ones with the cleanest data, the clearest use cases, and the willingness to start small and scale from proven results. AI in Maximo is a journey, and MAS 9.2 is a significant milestone on that journey, but it is not the final destination. IBM's roadmap includes additional agents for asset investment planning and continued model upgrades, which means the platform will continue to evolve.

The recommendation is to evaluate MAS 9.2's AI capabilities against your most pressing maintenance challenges, prototype one or two use cases in a non-production environment, and build the internal expertise needed to scale AI adoption over time. The cost of starting is modest; the cost of waiting is growing as the gap between AI-enabled and traditional asset management widens.