Verdantix Says Maximo Users Run Predictive Maintenance at Nearly 40 Percent While the Market Sits at 15: Why the Gap Keeps Widening

New research presented at MaximoWorld 2026 in Nashville put a number on something Maximo practitioners have suspected for years: organizations running IBM Maximo Application Suite are meaningfully further along on predictive maintenance than the broader asset-management market. According to the Verdantix findings announced on August 18, 2026, roughly 40 percent of Maximo users apply predictive or remote monitoring techniques to critical assets, compared with about 15 percent across the wider market. That is close to a 2.7x maturity advantage, and it is not a rounding artifact or a marketing claim. It comes from an independent analyst firm whose credibility depends on getting these numbers right, and it was presented on the main stage of the largest gathering of Maximo practitioners in the world.

The same research carried a second data point that deserves just as much attention: about 25 percent of Maximo users plan to deploy multi-agent AI in 2026, against roughly 15 percent in the broader market. Taken together, the two figures describe a user community that is not just adopting predictive techniques earlier, but is also moving faster toward the next phase of maintenance technology, where AI agents participate directly in workflows rather than sitting behind dashboards.

For anyone who has spent time inside asset-intensive organizations, the numbers pass the smell test. Maximo has always been disproportionately deployed where failure is expensive: utilities, oil and gas, transportation, mining, manufacturing. Those environments produced the condition-monitoring culture, the instrumentation investments, and the work-management discipline that predictive maintenance requires. A mature Maximo installation has asset hierarchies, failure codes, work order history, and meter readings that most CMMS deployments never accumulate. Predictive maintenance is not primarily a software purchase. It is the harvest of decades of data discipline, and Maximo shops have spent more decades doing that discipline than most.

But the Verdantix findings are also a warning, and this is the part of the story that most of the coverage has skipped. A 40-versus-15 gap means the majority of organizations, including many that use Maximo every day, are still not doing predictive maintenance on their critical assets at all. The research identifies the reasons, and they are unglamorous: poor data quality, integration problems, and skills gaps. Those three blockers appear in nearly every reliability survey published in the last five years, and the fact that they persist even inside a comparatively mature user base tells you how hard they are to fix with technology alone.

This article unpacks what the Verdantix research actually measured, why Maximo installations skew toward predictive maturity in the first place, what the leading organizations do differently from everyone else, and how the MAS 9.2 release changes the economics of closing the gap for the shops still on the sidelines. The goal is to turn an analyst headline into something a maintenance manager or reliability engineer can act on next quarter.

What the Verdantix Research Actually Says

The findings were announced via BusinessWire on August 18, 2026, timed to MaximoWorld 2026 in Nashville, where IBM and its partner ecosystem gather each year. Verdantix is an independent research and advisory firm focused on ESG, EHS, and operational technology markets, and its asset management benchmarking is widely cited by both vendors and end-user organizations. That matters because analyst survey data in this space is frequently abused: vendor press releases cherry-pick favorable numbers, and survey populations are often skewed toward customers of the sponsoring company. In this case the framing is notable precisely because IBM amplified research that quantifies both a strength and an unfinished agenda for its own user base.

The headline numbers are straightforward. Approximately 40 percent of Maximo users apply predictive maintenance techniques to their assets, versus roughly 15 percent marketwide. Verdantix characterizes Maximo users as about 2.7 times more mature in predictive and remote asset performance management for critical assets. On the forward-looking side, about a quarter of Maximo users plan multi-agent AI deployments in 2026, compared with around 15 percent of the broader market. Multi-agent AI, in this context, means orchestrated AI systems where multiple specialized agents collaborate, for example one agent triaging sensor anomalies, another drafting corrective work orders, and a third coordinating parts availability and scheduling.

Reading these numbers honestly requires understanding what they do and do not claim. The 40 percent figure describes application of predictive techniques to some portion of the asset base, not comprehensive coverage of every rotating asset in the fleet. In practice, even advanced organizations run predictive programs on their most critical assets and leave long-tail equipment on time-based or reactive strategies. That is rational prioritization, not failure. The comparison that matters is the delta between populations: if 40 percent of Maximo users have crossed the threshold of running predictive programs on critical assets while only 15 percent of the market has, something structural about the Maximo ecosystem is contributing to that difference.

The 2.7x maturity framing also implies more than tool adoption. Maturity in the Verdantix sense includes the organizational machinery around the techniques: defined criticality models, condition-based maintenance strategies tied to actual sensor data, planners who trust and act on predictive alerts, and metrics that connect avoided failures to business outcomes. Organizations that score high on maturity have typically been running reliability-centered maintenance programs for years, with Maximo serving as the system of record that makes those programs auditable.

The multi-agent AI figure is the most speculative of the three, because planned deployment is not deployed capability. Still, the direction is consistent with everything else observable in the Maximo community since MAS 9.2 shipped: the MCP Server for agentic access to Maximo data, the decoupling of Predict from the IoT service layer, and Condition Insight's agentic recommendations all point the same direction. IBM is building the platform plumbing for agentic maintenance, and a quarter of its user base is signaling intent to use it. Whether those plans convert into production systems in 2026 or slip into 2027, the intent gap itself is informative: Maximo users are nearly twice as likely as the market to believe agentic AI is ready for their operations.

One caveat applies to all survey research of this type: respondents self-select and self-report, and Maximo users skew toward large, asset-intensive enterprises where maintenance maturity is often higher for reasons unrelated to software. The honest interpretation is not that Maximo causes predictive maturity, but that Maximo tends to live inside organizations that have the scale, instrumentation, and governance to support it. The rest of this article is about what those organizations actually do, because that is the transferable part.

Why Maximo Installations Skew Toward Predictive Maturity

The most common mistake in reading a statistic like this is attributing the gap to the software itself. Maximo does not predict failures. It is not a machine-learning engine, and no amount of licensing changes physics. What Maximo provides is the data substrate that predictive maintenance depends on, and after decades of deployment in exactly the industries where that substrate matters most, the correlation is entirely predictable.

Start with asset data. Predictive maintenance requires knowing what you own, where it is, what it does, how it fails, and what its failure costs. A Maximo installation that has been operated with any discipline accumulates exactly this: hierarchical asset records with classification, locations and systems, nameplate and specification data, failure codes mapped to work order history, and operating context linking assets to production impact. This is the unglamorous foundation that data scientists call labeled failure history, and it is the single scarcest resource in predictive maintenance. Most organizations attempting to build predictive models from scratch discover that their historical failure data is too sparse, too inconsistent, or too poorly categorized to train anything useful. Maximo shops that maintained failure code discipline over years sidestep that problem, because the labeled history already exists in the work order tables.

Then there is condition data. The industries where Maximo dominates were also the earliest adopters of instrumentation: vibration sensors on rotating equipment in petrochemical, temperature and dissolved-gas monitoring in transformers across utilities, telemetry in rail fleets. When SCADA, historians, and condition-monitoring systems feed readings into Maximo as measurements and meter records, the organization builds a longitudinal condition history alongside its failure history. That pairing, failures labeled against condition trajectories, is precisely what condition-based and predictive models need. Organizations that never wired instrumentation into their maintenance system have to reconstruct that history from exports and spreadsheets, and most never do.

Work-management discipline is the third pillar. Predictive alerts are worthless if the organization cannot convert them into planned, scheduled, executed, and documented work. Maximo's strength has always been the full work-management lifecycle: work identification through planning, scheduling, execution, and closure with actuals captured. An organization that runs that lifecycle with high PM compliance and strong planner-scheduler roles already has the operational muscle to absorb predictive work orders. In shops without that discipline, predictive alerts pile up in queues, get dismissed as noise, and the program dies within a year. This is why predictive maintenance adoption correlates so strongly with basic CMMS hygiene, a relationship that reliability consultants have been preaching for two decades.

Scale and criticality complete the picture. Maximo's economics favor large asset portfolios where the cost of failure justifies the cost of the platform. Those same economics justify predictive programs: when a single hour of unplanned downtime on a critical unit costs six figures, the business case for sensors and models closes itself. Small organizations with cheap downtime do not buy enterprise EAM platforms and do not run predictive programs, so the populations being compared differ in ways that go far beyond software choice.

Finally, there is the ecosystem effect. Maximo's partner network, user groups, and annual conference have been transmitting reliability-engineering practice, not just product knowledge, for twenty-five years. The conversations at MaximoWorld are as much about failure modes, criticality analysis, and work management culture as they are about features. A community that talks shop at that level reproduces its best practices. The Verdantix numbers are, in part, a measurement of that twenty-five-year flywheel.

What the Leaders Do Differently

The gap between the 40 percent and the 15 percent is not explained by budget alone, though budget helps. Across the organizations that have successfully operationalized predictive maintenance on Maximo, a consistent pattern of practices shows up, and none of them are secret.

The first is a data quality program with ownership. Leading organizations treat asset and failure data as a governed asset in its own right, with named stewardship, periodic audits, and cleanup sprints tied to business events like asset commissioning or fleet overhauls. They enforce failure code discipline through application design rather than exhortation: mandatory fields, classification-driven code lists, and closure reviews that reject work orders with useless failure documentation. The payoff compounds, because every clean work order closure improves both the reliability statistics and the training data for predictive models. Laggards, by contrast, treat data cleanup as a one-time project before a go-live, after which entropy resumes. Five years later their failure history is again unreliable, and any predictive initiative stalls at the data-preparation stage.

The second practice is starting narrow and proving the loop. Mature programs do not attempt to predict failures across the entire asset registry. They select one asset class, usually rotating equipment with clear failure signatures and high consequence, instrument it adequately, run anomaly detection and remaining-useful-life models, and crucially, close the loop: alert to inspection to work order to outcome, with the outcome recorded back into Maximo. That closed loop is what turns a pilot into a program, because it produces the evidence, in work order history, that predictions are being acted upon and failures avoided. Organizations that stall at the pilot stage almost always skipped the loop-closing step, leaving their data science team producing insights that operations never consumes.

The third is skills investment in the maintenance organization itself. The leaders do not hire a data science team and hope synergy happens. They train reliability engineers to interpret model outputs, teach planners how predictive work orders differ from calendar PMs, and build translator roles, often called reliability data analysts, who sit between the analytics platform and the work-management process. The Verdantix research lists skills gaps among the top blockers marketwide, and the leaders solve it internally rather than waiting for the labor market to deliver maintenance-savvy data scientists, a candidate population that barely exists.

The fourth is measurement that survives contact with finance. Leading programs track MTBF improvement, unplanned downtime hours, MTTR, PM compliance, and the ratio of predictive to reactive work, and they tie avoided-failure estimates to production or revenue impact in terms finance accepts. This matters more than it sounds: predictive programs live or die on budget renewal, and budget renewal depends on a credible value narrative. Programs that cannot articulate avoided cost in the CFO's language get cut in the first downturn, regardless of their technical elegance.

The fifth practice is platform pragmatism. Leaders use what their EAM already provides before bolting on new platforms: Maximo's condition monitoring and measurement history, its work queues for high failure-probability assets, and in recent releases, Predict and Monitor capabilities that run inside the same governance model as the rest of maintenance. Integration projects are scoped to sensor data ingestion rather than wholesale workflow replacement. The contrast with laggard organizations, which often stall in extended platform evaluations while their data decays, is stark. The leaders are distinguished less by which tools they chose and more by how quickly they put any tool into the closed alert-to-work-order loop.

The Blockers Holding the Majority Back

If the practices above are well known, why does the broader market sit at 15 percent and why do even many Maximo organizations lag? The Verdantix research names the culprits, and each deserves honest treatment rather than hand-waving.

Poor data quality is the first and most fundamental blocker. Predictive models trained on mislabeled, sparse, or inconsistent failure history produce predictions nobody trusts, and untrusted predictions poison the program. The deeper problem is that data quality is a moving target: assets get installed without records, work orders get closed with garbage failure codes under schedule pressure, and reorganizations orphan location hierarchies. Organizations consistently underestimate that building predictive capability requires a sustained data operations function, not a cleanup sprint. The realistic path for most shops is to fix data quality for a single critical asset class first, accepting imperfection elsewhere, and expand as the program proves value. Attempting registry-wide data perfection before any predictive work begins is the most common form of program suicide.

Integration issues are the second blocker, and in the Maximo world they have a specific texture. Condition data lives in historians, SCADA systems, vibration-analysis platforms, and increasingly cloud IoT services, while the work-management truth lives in Maximo. Bridging those worlds has historically meant custom middleware, fragile point-to-point integrations, and the organizational friction of OT and IT teams with different priorities and security postures. Many predictive initiatives die not because the models fail but because the sensor-to-alert-to-work-order pipeline is never made reliable in production. This is precisely the problem IBM targeted in MAS 9.2 by decoupling Predict from the IoT service layer and re-architecting Monitor to reduce mandatory dependencies, lowering both the integration surface and the resource footprint required to run condition-based programs. Organizations blocked on integration for years may find the 9.2 architecture removes enough friction to finally justify a restart.

Skills gaps form the third blocker, and they operate at two levels. At the specialist level, the market lacks people who combine reliability engineering judgment with data science literacy. At the workforce level, planners and technicians must trust and act on predictive outputs, and that trust is earned through experience, not training slides. The pragmatic answer, which the leading quartile has quietly adopted, is to grow capability internally: take strong reliability engineers, give them the analytics tools and time, and accept a slower ramp in exchange for durable institutional knowledge. Organizations that outsource the entire capability to a vendor often find that when the contract ends, so does the predictive program.

Budget and organizational attention deserve mention as the meta-blocker. Predictive maintenance competes for capital against initiatives with clearer, faster payoffs, and its benefits accrue as avoided events that are inherently counterfactual and hard to claim credit for. The organizations that sustain funding are the ones that started with the highest-consequence assets, where a single avoided failure pays for years of program cost, and that reported those saves with work-order-level evidence.

None of these blockers are solved by software releases alone, which is exactly why the gap persists marketwide. But the blockers are now more tractable than they were five years ago: the MAS 9.2 changes lower the platform burden, and the accumulated playbook from the leading 40 percent shows what the rest should copy.

How MAS 9.2 Changes the Predictive Economics

For Maximo organizations that have watched the predictive-maturity gap from the sidelines, the timing of MAS 9.2 is fortunate, because the release attacks several of the classic blockers directly.

The most consequential change is architectural: Predict is decoupled from the IoT service layer, and Monitor and IoT Platform have been re-architected around a leaner stack with fewer mandatory dependencies. Under previous versions, running predictive workloads meant operating a substantial IoT footprint even when the use case was narrow, which inflated OpenShift resource requirements and, in practical terms, inflated the price of admission for predictive maintenance. For mid-size deployments where the IoT stack cost dominated the business case, this re-architecture changes the calculation. Teams can now stand up condition monitoring and Predict capabilities with a smaller resource envelope, which shortens the path from decision to pilot and makes the pilot cheaper to abandon if it fails, itself a feature that encourages experimentation.

Condition Insight layers agentic recommendations on top of that leaner foundation, moving the product beyond raw anomaly scores toward recommendations an engineer can evaluate and act on. This addresses the interpretation-skills blocker at its root: the gap between a statistical anomaly and a maintainable action is where most programs lose their workforce, and recommendations expressed in maintenance language narrow that gap.

The MCP Server introduced in MAS 9.2 matters for a different reason: it opens Maximo data and actions to external AI agents through a standardized protocol, rather than the bespoke API integrations each team previously had to build. In the near term this mostly benefits organizations with AI engineering capacity, but the direction is significant. The pattern that Verdantix's multi-agent statistic points toward, where triage agents, planning agents, and scheduling agents cooperate across systems, requires exactly this kind of governed programmatic access to EAM data. Shops that establish MCP-based access patterns now will be the ones that can adopt agentic workflows incrementally rather than as a rip-and-replace project.

For existing Monitor deployments, migration to the 9.2 re-architecture is a real project with real dependencies, and organizations should evaluate it against their roadmap rather than assume a free upgrade. But the strategic calculus favors movement: the direction of the platform is toward lower operating cost per predictive use case, and every release cycle spent on the old architecture is a cycle of paying the premium that the re-architecture removes.

Taken together, these changes do not eliminate the human blockers. Data quality programs, skills development, and closed-loop work management remain the actual work. What MAS 9.2 changes is the fixed cost of getting into the game, and for the majority still at zero predictive coverage, fixed cost was always the first excuse. As the platform burden falls, the remaining gap between leaders and laggards becomes a pure execution gap, and that is a gap no vendor can close on your behalf.

Practical Implications

For Maximo organizations currently below the predictive threshold, the Verdantix numbers should be read as a benchmarking exercise with three concrete takeaways. First, establish where you actually stand: pull the last two years of work order history and calculate the percentage of unplanned corrective work on critical assets, PM compliance, and whether any condition data flows into Maximo today. This baseline costs a week of analyst time and tells you which blocker dominates in your environment. Second, pick one critical asset class and commit to closing the full loop, from sensor or inspection input to prediction to work order to recorded outcome, within two quarters. The loop matters more than the sophistication of the model, because the loop is what builds both the trust and the training data. Third, verify your failure-code and asset-hierarchy discipline on that chosen asset class before any modeling starts, because a model trained on your current data quality will embarrass you in front of the operations team. Fourth, if you are planning a MAS 9.2 migration anyway, sequence the Predict and Monitor evaluation into that project rather than treating predictive capability as a separate future initiative; the architectural changes lower the incremental cost substantially. Finally, staff the program from inside: identify the reliability engineer or senior planner who already understands the assets, and make them the program owner with protected time. External expertise is useful, but the durable capability, and the workforce trust that sustains it, is built by people who will still be there in five years.

Bottom Line

The Verdantix research quantified a real and widening divide: Maximo users run predictive maintenance on critical assets at roughly 40 percent versus 15 percent marketwide, and they are roughly twice as likely to be planning multi-agent AI deployments this year. The gap exists because Maximo's traditional install base, asset-heavy industries with decades of work-management discipline, sits closer to the data foundations that prediction requires. But the more actionable finding is what holds the majority back: data quality, integration friction, and skills, none of which are primarily software problems. MAS 9.2 meaningfully lowers the platform-side barriers by decoupling Predict from the IoT stack and opening governed agentic access through the MCP Server, which shifts the challenge toward execution. The organizations that close the gap next will be the ones that start narrow, close the alert-to-work-order loop, and treat data quality as an ongoing operation rather than a project. The playbook is not secret. It is simply work, and the 40 percent have already done it.