From Reactive to Prescriptive: The APM Stack in MAS 9.2

From Reactive to Prescriptive: The APM Stack in MAS 9.2For years, many organizations used Maximo primarily as an Enterprise Asset Management platform focused on work execution, preventive maintenance, and operational control. MAS 9.2 makes it clear that IBM is no longer positioning Manage, Monitor, Health, Predict, and Reliability Strategies as separate products. They are connected operational capabilities, and the release that brings them together is defined by one feature: Maximo Condition Insight. This AI-powered capability, embedded directly in the APM layer, analyzes work orders, inspections, meter readings, failure modes, and alerts to deliver plain-language recommendations on what to do next. This article covers the full APM stack in MAS 9.2, how the pieces connect, the data flow that makes Condition Insight possible, and the practical steps to build a condition-based maintenance program on top of it.

The APM Stack: Five Connected Capabilities

The APM stack in MAS 9.2 consists of five connected capabilities, each with a specific role in the flow from data to action. Understanding what each does and how they connect is the foundation for building a condition-based maintenance program.

Maximo Manage is the EAM foundation. It holds the asset registry, work order history, failure codes, PM schedules, inspection results, and meter readings. Every other APM capability reads from and writes to Manage. Without clean data in Manage, the rest of the stack cannot function effectively. Manage is where the asset hierarchy lives, where work orders are created and completed, where failure coding captures what went wrong, and where meter readings record operating context. It is the system of record, and every recommendation from the AI layer traces back to data that originated here.

Maximo Monitor is the IoT and sensor data layer. It ingests time-series data from sensors, devices, and industrial systems through multiple connectivity options: direct MQTT, gateway-to-cloud, PI-to-Monitor bridge, and OPC-UA. Monitor processes this data through analytics pipelines and anomaly detection models to drive alerts, dashboards, and integrations. In MAS 9.2, the Monitor architecture was rebuilt to remove the Kafka dependency, which simplifies deployment and reduces operational overhead. Monitor has over 1,600 pre-configured device partners, which means many common industrial sensors can connect with minimal configuration. For organizations already running OSIsoft PI or other historian platforms, the PI-to-Monitor bridge replicates selected tags into Monitor's time-series store without replacing existing infrastructure.

Maximo Health provides the bridge between raw sensor data and maintenance decision-making. It calculates health scores for assets on a 0 to 100 scale, aggregating multiple input dimensions: work order history, open high-priority work orders, meter readings versus design limits, age versus design life, and Monitor anomaly alerts. Health's fleet view ranks all assets in a selected classification by their condition score, so a maintenance manager looking at 500 monitored assets sees a ranked list rather than 500 individual dashboards. Health also includes an investment optimization module that uses condition scores and historical maintenance cost trends to model the cost of maintaining versus replacing each asset over a 5 to 10 year planning horizon. This module is particularly valuable for capital planning cycles where asset replacement decisions need to be justified with data rather than gut feel.

Maximo Predict uses AI and machine learning to forecast asset performance and maintenance needs. It builds models that predict days to failure, probability of failure within defined time horizons, anomaly detection, and remaining useful life. Predict requires labeled training data: examples of sensor readings and maintenance histories from periods that preceded confirmed failure events. The models are trained using classification algorithms including Random Forest, Gradient Boosting, and neural networks, depending on data characteristics. Once deployed, Predict evaluates each monitored asset's current sensor readings against the model at a configurable interval and outputs failure probability scores for 7-day, 14-day, and 30-day horizons.

Reliability Strategies connects failure analysis to maintenance action. It uses FMEA and RCM methodologies, backed by a library of 800 asset types and 58,000 failure modes, to identify failure modes and assign mitigation actions based on Risk Priority Numbers. The output is not a document; it is job plans and PMs written directly into the Maximo work objects that your team executes. In 9.1, AI was added to suggest boundary conditions and generate components and failure mechanisms, accelerating the slowest part of FMEA authoring.

The flow across these five capabilities is a closed loop: failure analysis defines the maintenance strategy, sensors and meters feed monitoring data, health scoring aggregates condition, predictive models forecast failures, alerts trigger work orders, work execution generates new history, and that history feeds back into reliability improvement. This closed loop is what makes the APM stack self-improving over time, assuming data quality is maintained.

Condition Insight: The AI Layer That Ties It All Together

Maximo Condition Insight is the headline APM feature in MAS 9.2. It is an agentic AI capability, powered by IBM watsonx, that evaluates work orders, metrics, time-series data, meter readings, Failure Mode and Effects Analysis, and alerts to evaluate asset condition, uncover performance patterns, and provide actionable recommendations in plain, understandable language.

Condition Insight is not a new application in the navigator. It is an embedded layer that surfaces asset condition recommendations inside Manage. When a planner opens a work order for a critical pump, Condition Insight can flag that the same pump has had three similar failure codes in the last 18 months, that the vibration meter is trending up, and that the reliability strategy recommends a specific PM task at this run-hour threshold. The planner sees a plain-language summary with context, trends, and recommended actions, without needing to switch to Monitor, Health, or Predict.

The underlying pattern is what IBM calls "asset-first AI." Rather than bolting an analytics layer on top of Maximo, the AI operates on the same data model that Manage already uses. Condition Insight relies on the asset, work order, inspection, and meter data that already lives in Manage, plus the failure mode and strategy data from Reliability Strategies. This means the quality of Condition Insight recommendations is directly tied to the quality of your Maximo data: clean failure codes, well-maintained meter readings, and up-to-date FMEA studies produce useful recommendations; sparse or inconsistent data produces noise.

A practical example: a maintenance manager asks the system about the condition of a critical centrifugal pump. Condition Insight returns a summary like this: "Pump P-1043 at Plant A has an elevated vibration trend over the past 30 days, with readings 40 percent above the baseline. Three corrective work orders in the past 18 months relate to bearing failures. The reliability strategy for this asset class recommends a vibration analysis at 15,000 run hours, and the current reading is 14,200 hours. Recommended action: schedule a vibration analysis and bearing inspection within the next 800 run hours." This summary, generated in seconds, replaces what would normally require a reliability engineer to pull data from three different systems and write a memo.

The value of Condition Insight is not just speed. It is consistency. Every planner, every maintenance manager, and every reliability engineer sees the same recommendation for the same asset, based on the same data. This removes the dependency on individual expertise that plagues many maintenance organizations. When the senior reliability engineer retires, their knowledge does not retire with them, because the patterns they identified are now encoded in the system and surfaced automatically through Condition Insight.

Alert Insights: From Raw Alerts to Intelligent Actions

Alongside Condition Insight, MAS 9.2 introduces Alert Insights. When an alert fires from Monitor, Alert Insights reviews it in context, analyzing the asset, its location, related alerts, and historical work orders. It generates a diagnosis of the likely failure mode along with prioritized remediation recommendations.

For example, when a pressure warning alert fires on a centrifugal pump, Alert Insights can identify whether the cause is a downstream blockage, a stuck relief valve, excessive pump speed, or sensor drift. It then recommends specific inspection steps aligned to the asset's reliability strategy. This is fundamentally different from a raw alert that says "pressure high" and leaves it to the operator to figure out why.

Alert Insights reduces alert fatigue, which is one of the most common reasons condition monitoring programs fail. When operators receive hundreds of alerts per day with no context, they start ignoring them. Alert Insights transforms each alert from a binary notification into a diagnostic package with context, likely cause, and recommended action. This prioritization means operators can focus on the alerts that matter and dismiss the ones that do not.

Building a Condition-Based Maintenance Program

Building a CBM program on MAS 9.2 requires a structured approach. The most common reason predictive maintenance pilots stall is that organizations try to skip from Level 1 or 2 (reactive or preventive maintenance) directly to Level 4 (predictive) without building the sensor and data foundation of Level 3 (condition-based).

The maintenance maturity ladder provides a framework:

Level Trigger for Maintenance Tooling

1 - Reactive Failure has occurred Maximo Manage (corrective WO)

2 - Preventive Fixed time or usage interval Maximo Manage (PM, Job Plans)

3 - Condition-based Sensor reading crosses threshold MAS Monitor, Health, Manage

4 - Predictive AI-forecasted failure probability MAS Monitor + Predict + Health

5 - Prescriptive AI recommends specific action Condition Insight, Alert Insights

The first step is sensor selection. MAS Monitor cannot predict failures on equipment for which there is no sensor data. For rotating equipment (pumps, motors, fans, compressors), the highest-value sensors target the leading failure modes. Vibration (tri-axial accelerometers) detects imbalance, misalignment, bearing wear, looseness, and cavitation. Vibration is the most universally applicable sensor type for rotating machinery because virtually all failure modes produce a characteristic vibration signature. Temperature at bearing housings detects lubrication problems or bearing wear as a lagging indicator. Motor current and power analysis can detect rotor bar failure, stator winding degradation, and hydraulic problems without physical access to rotating components.

For heat exchangers, pressure vessels, and piping systems, differential pressure detects fouling and filter loading. Acoustic emission sensors detect leaks, cavitation, and crack propagation in pressure boundaries at frequencies above the range of standard accelerometers. Connectivity options include direct MQTT, gateway-to-cloud via PLC or edge gateway, PI-to-Monitor bridge for sites already running OSIsoft PI, and OPC-UA for industrial control systems.

The second step is configuring Health score cards. Health's condition score aggregates multiple input dimensions into a single 0-100 score per asset. Organizations configure score cards that define data sources and weights, normalization rules, and weighting of each dimension. The fleet view then ranks all assets by their condition score, with red zone assets (score below 30) requiring immediate attention, yellow zone assets (30 to 60) needing monitoring, and green zone assets (60 to 100) operating within normal parameters. Start with a simple score card: 40 percent weight on recent corrective work order frequency, 30 percent on Monitor anomaly alerts, and 30 percent on meter readings versus design limits. Refine the weights after 90 days of operational data.

The third step is training Predict models. This requires labeled training data: failure events from Manage work orders, corresponding sensor time series from Monitor, and sufficient history to identify patterns. The training pipeline extracts work order records identifying failure dates and modes, extracts corresponding sensor time series, trains classification models, and validates performance on held-out test data. Once deployed, Predict evaluates each monitored asset's current sensor readings against the model at a configurable interval (default: hourly) and outputs failure probability scores for defined time horizons.

The fourth step is enabling Condition Insight and Alert Insights. Once the foundation is in place (sensors in Monitor, scores in Health, models in Predict, failure modes in Reliability Strategies), Condition Insight can analyze all of it and produce recommendations. Start with a single asset class, measure recommendation acceptance over 30 days, and expand from there.

Common Pitfalls and How to Avoid Them

APM implementations fail for predictable reasons. Understanding these pitfalls before starting saves time, money, and organizational credibility.

The first pitfall is sensor overdeployment. Organizations install sensors on every asset in the plant, generating massive data volumes but without a clear analysis plan for each sensor stream. The result is that the monitoring team is overwhelmed with data and cannot identify the signals that matter. The fix: start with critical assets only. Use the asset criticality ranking from Manage to identify the top 10 percent of assets by risk, and deploy sensors only on those. Expand incrementally as the monitoring team develops the capacity to act on the data.

The second pitfall is skipping the failure coding foundation. Predictive models and Condition Insight both depend on failure history from Manage. If your work orders have inconsistent or missing failure codes, the models train on bad data and produce unreliable predictions. The fix: before deploying Predict or Condition Insight, run a failure code audit. Ensure that at least 80 percent of corrective work orders on critical assets have complete failure class, problem code, and cause code. Use Work Order Intelligence (available in Manage 9.0 and later) to improve failure coding going forward.

The third pitfall is underestimating the organizational change. Moving from time-based to condition-based maintenance changes how planners, technicians, and managers work. Planners need to trust sensor data over calendar intervals. Technicians need to understand why a PM was deferred based on a health score. Managers need to accept that some assets will not receive scheduled maintenance because the data says they do not need it. The fix: involve the maintenance team in the pilot from day one, share the data openly, and let the results speak for themselves. A successful pilot on one asset class builds the trust needed for broader adoption.

Practical Implications

The APM stack in MAS 9.2 is the most integrated condition-based maintenance platform IBM has offered. Condition Insight and Alert Insights close the gap between data and action that has historically been the bottleneck in APM implementations. But the stack depends on data quality at every layer. Clean failure codes in Manage, well-placed sensors feeding Monitor, accurate health score configuration, sufficient failure history for Predict training, and up-to-date FMEA studies in Reliability Strategies are all prerequisites. Organizations that invest in the foundation will see the value; organizations that try to skip to the AI layer without building it will get noise. The closed-loop design of the stack means that every improvement in data quality compounds over time, as better work order data improves Predict models, which improves Condition Insight recommendations, which drive better maintenance decisions, which generate better work order data.

The Bottom Line

MAS 9.2 has transformed the Maximo APM story from a collection of separate applications into a connected, AI-powered platform. Condition Insight is the capability that makes the connection visible and actionable, and it is the primary reason for reliability-focused organizations to prioritize the 9.2 upgrade. The path to value is structured but not simple: fix your failure coding, deploy sensors on critical assets, configure Health score cards, train Predict models, populate Reliability Strategies, and then enable Condition Insight. Each step builds on the previous one, and the result is a maintenance program that is defensible, data-driven, and continuously improving. For organizations that have been waiting for APM to mature before investing, the wait is over. MAS 9.2 is the version where the pieces finally fit together.