IBM has introduced Maximo Condition Insight, a new AI-powered capability within the Maximo Application Suite that reads asset data and explains, in plain language, what is happening to an asset and what should be done about it. IBM announced it on 5 December 2025 as part of the broader Maximo Asset Performance Management (APM) capability set, and it is the clearest sign yet that agentic AI is moving from the MAS roadmap into features that functional teams will actually be asked to configure. For anyone responsible for condition-based maintenance, this is a development worth understanding before the first internal demo lands.
What IBM Announced
IBM positions Maximo Condition Insight as an agentic AI capability inside Maximo APM. Rather than a new standalone application, it operates across the data already flowing through MAS: work orders, metrics, time-series data, meter readings, Failure Mode and Effects Analysis records, and alerts. The engine is powered by IBM watsonx, and the stated aim is to compress the distance between an unusual signal and a defensible maintenance decision.
The feature has four headline behaviours, as described by IBM:
- Summarising asset condition from meters, KPIs, and alerts in seconds, without a separate data-modelling exercise.
- Mapping detected conditions to failure modes so the recommended activity aligns with the organisation’s reliability strategy.
- A conversational interface through the Maximo AI Assistant, so a planner or reliability engineer can ask “what is happening with this chiller” and get a readable answer.
- Automated or semi-automated work order creation against prescribed strategies, described as a near-term capability rather than a finished one.
IBM Research has published a parallel account describing how the Condition Insights agent is being piloted against a real chiller in IBM’s own estate, using time-series foundation models to spot the kind of low-grade bearing vibration that human operators cannot hear. That pairing of product announcement and research disclosure is why this is worth taking seriously. It is not a slide-ware feature.
Where It Sits in the Suite
A useful mental model: Maximo Manage remains the transactional spine (work orders, assets, inventory, procurement). Maximo APM sits above it as the lens for asset health, reliability analytics and investment decisions. Monitor handles real-time IoT ingestion. Predict targets probabilistic failure models against historical data. Health scores criticality and condition.
Condition Insight is not a replacement for any of those. It is an agent that reasons across the data already in place, produces an explanation, and hands back a recommendation. In practice that means the quality of the output is bounded by the quality of the inputs. A beautiful natural-language summary of bad asset data is still bad asset data, now served with more conviction.
Why This Matters for MAS Customers
Two things make this announcement different from previous AI feature drops in the suite.
It targets the explanation problem, not just the prediction problem
Most predictive maintenance programmes do not fail because the maths is wrong. They fail because the output is a probability score that nobody trusts, from a model nobody can interrogate, presented on a dashboard that sits outside the work management tool. Condition Insight aims squarely at that gap. If an agent can say “vibration amplitude on this asset has drifted outside its normal band over the last fourteen days, the closest FMEA entry is bearing wear, the recommended job plan is X,” the maintenance planner has something to act on and an auditor has something traceable. That is a different conversation from “the model says 0.73.”
If your organisation has struggled to move beyond pilots, the pattern is familiar and the fix is usually organisational, not algorithmic. Our note on why predictive maintenance programmes fail covers that ground and applies directly to how you should approach Condition Insight.
It assumes your data foundations are in order
IBM is explicit that Condition Insight reads what MAS already knows. That sounds inclusive until you examine the implications. If asset hierarchies are inconsistent, if failure coding is optional in practice, if meters are configured but readings are sporadic, or if FMEAs are a Word document owned by a reliability engineer who left last year, the agent has little to reason over.
In the estates we look after, the most common gaps are:
- Failure coding enforced unevenly across sites, so history looks thinner than it is.
- Meter reading groups defined but not consistently populated, which hides slow drift.
- FMEA data captured outside Maximo, so failure modes are not linked to the asset records the AI agent will query.
- Duplicate asset records caused by historical migrations, which split condition evidence across ghost twins.
None of these are new problems. Condition Insight simply makes them more expensive to ignore.
What Functional Teams Should Check Now
You do not need to wait for a vendor presentation to do useful preparation. For teams on MAS 9.x or planning a move to it, the following are all worth working through in the next quarter.
- Rationalise your failure coding. Confirm that failure classes, problem codes, and failure codes are defined, required on closure, and actually being entered. Run a quick query against the last six months of closed work orders and look at how much content sits in the code structure versus the long description.
- Get FMEAs into Maximo. Agentic AI mapping condition to failure modes only works if the failure modes are present as structured data. Even a modest catalogue of critical assets with their top failure modes tied to job plans is a meaningful starting point.
- Clean up asset identity. Duplicate or orphaned assets will mislead any AI agent. A structured deduplication exercise, with ownership, is overdue in most estates.
- Validate meter coverage on critical assets. Condition Insight reads meter readings. If your critical assets have the right meter groups defined, and readings are coming in from the historian or inspection process, you are much closer to a useful pilot.
- Agree who owns the output. An AI recommendation is still a recommendation. Decide now whether a Condition Insight suggestion becomes a work request automatically, a triage item for a reliability engineer, or an alert that a planner reviews. That is a governance choice, not a product setting.
For teams already thinking about sequencing, our earlier piece on sequencing Monitor and Predict after Manage argues that advanced suite capabilities pay back faster when Manage is trusted first. The same logic applies to Condition Insight.
A Grounded View
Agentic AI in MAS is being marketed enthusiastically, and that is fair enough: this is the direction of travel IBM has signalled since the introduction of the Maximo AI Assistant. The useful posture for asset-management teams is neither cynicism nor early adoption for its own sake. It is to treat Condition Insight as a demanding new consumer of the data you already have, to close the gaps that consumer exposes, and to decide which assets genuinely warrant a condition-based approach before you ask an agent to recommend work against them.
Used that way, it is a practical accelerant for the move from reactive and calendar-based maintenance toward a condition-based posture. Used without that preparation, it becomes another dashboard nobody opens. The difference is not the technology.