Calling something a gold standard is a claim about what practitioners ought to do, not about what they do. In maintenance the gap between those two things is unusually wide and unusually well documented. In the 2025 State of Industrial Maintenance survey, predictive maintenance was the primary strategy for 27% of respondents, down from 30% the year before, while preventive maintenance sat at 71% and running equipment to failure at 38%. The reference standard is losing ground on the shop floor while gaining it in the literature.
Key takeaways
- ISO 17359:2018 covers all machines and directs monitoring at root cause failure modes.
- ISO 13374 spans four parts: general guidelines, data processing, communication, presentation.
- Predictive maintenance was the primary strategy for 27% of teams in 2025, against 30% in 2024.
- Deloitte’s range: uptime up 10% to 20%, planning time down 20% to 50%.
The standards are what make it a standard
Strip away the vendor material and the claim rests on two documents that any engineer can buy and read.
ISO 17359:2018, condition monitoring and diagnostics of machines, general guidelines, sets out the generic procedure for implementing a condition monitoring programme and is explicitly applicable to all machines. It covers the parameters usually associated with performance and condition criteria: vibration, temperature, tribology, flow rate, contamination, power and speed. Two features of the 2018 edition, which replaced the 2011 version, matter more than the parameter list. It directs monitoring activity at root cause failure modes rather than at whatever is convenient to instrument, and it describes a generic approach to setting alarm criteria, which is the step most programmes improvise and then regret.
The ISO 13374 series handles the part that turns readings into a system. Part 1, published in 2003, gives general guidelines for the software that processes, communicates and presents condition monitoring information. Part 2 details the data processing methodology. Part 3, from 2012, specifies data communication for an open reference architecture. Part 4, from 2015, covers presentation for technical analysis and decision support. The problem it was written to solve is entirely familiar to anyone running more than one monitoring vendor: systems that cannot exchange data without bespoke integration, and therefore no unified view of machine condition.
That is the honest basis for the gold standard label. Not superior results in every case, but a documented method with defined interfaces, which is what an auditable engineering practice looks like.
What the adoption numbers actually show
The picture from the field is not one of steady conversion. It is one of a plateau, with a rhetorical consensus sitting on top of it.
| Indicator, 2025 survey | Figure | Reading |
|---|---|---|
| Preventive as primary strategy | 71% | The calendar still governs most plants |
| Run to failure | 38% | Reactive work is not a legacy category |
| Predictive as primary strategy | 27%, from 30% in 2024 | Movement is sideways, not upward |
| AI fully or partly implemented | 32% | Tooling is ahead of practice |
| Expect to adopt AI within 12 months | 65% | Intention, which is not the same as budget |
Respondents could report more than one strategy, which is why the shares exceed 100%, and that overlap is itself informative: predictive work is usually layered onto a preventive programme rather than replacing it. The stated obstacles are equally prosaic. Budget was named by 25%, lack of expertise by 24% and cybersecurity concerns by 22%. Nobody in that list is arguing that condition-based intervention does not work.
The benefit ranges, stated as ranges
Deloitte’s position paper on predictive maintenance gives the figures most often quoted in board papers, and they are worth reproducing with their original width rather than as a single headline number. Predictive maintenance increases equipment uptime by 10% to 20%, reduces overall maintenance costs by 5% to 10%, and cuts maintenance planning time by 20% to 50%.
The planning figure is the one we would put in front of a sceptical operations director, because it is the least dependent on asset criticality. Uptime gains scale with how expensive your downtime already is, so a plant with cheap stoppages sees very little. Planning time, by contrast, improves for structural reasons: intervening on evidence removes the guesswork about what parts and skills a job needs, which is where scheduling waste accumulates regardless of sector.
Second-hand versions of these figures circulate with much larger numbers attached, sometimes cost reductions of 40% or downtime cuts of 50%. Those do not match the source document and we would not put them in a business case.
The ladder most plants are actually climbing
Reading the standards alongside the survey data, the progression that survives contact with a real site looks like this. Each rung is a prerequisite for the next, and skipping one is the most common reason a programme stalls at rung three.
- An asset register that reflects reality. Not the one in the finance system. The one that says which machines exist, in what duty cycle, with what redundancy behind them.
- A criticality ranking based on consequence. ISO 17359’s logic starts here, because monitoring effort should follow failure modes that matter rather than sensors that are easy to fit.
- Routine condition monitoring on the ranked assets. Vibration routes, thermography and oil analysis on a schedule deliver a large share of the benefit before any continuous instrumentation is bought.
- Alarm criteria set deliberately. This is the step the 2018 standard added detail to, and it is where a programme either becomes trusted or becomes noise the operators mute.
- An interoperable data layer. The ISO 13374 architecture matters only once you have more than one source, at which point it stops being theoretical and starts being the difference between one view and five dashboards.
Why the label has held anyway
Something has to explain a reference standard that keeps its status while its adoption share drifts down. Our reading is that the label describes the direction of every adjacent development rather than the current installed base. Instrumentation is getting cheaper and more self-aware, which is the argument behind the case for smart sensors in quality control. Regulatory and insurance expectations increasingly assume evidence of asset condition rather than evidence of a completed inspection round. And the workforce arithmetic points one way: fewer experienced maintainers means more reliance on instruments that can flag what a veteran would once have heard.
None of that guarantees the adoption curve turns upward next year. It does mean the standards work is already done, the interfaces are specified, and the constraint sits squarely in budgets and skills. That is a considerably better position than most industrial technologies occupy, and it is the honest version of why this one earned the label in the first place.
Want the deployment detail rather than the strategic case?
Where these programmes actually stall, what an hour of stoppage costs, and how we would scope a first rollout.
Sources: ISO 17359:2018, Condition monitoring and diagnostics of machines, General guidelines, for the generic procedure for setting up a condition monitoring programme, its applicability to all machines, the monitored parameters (vibration, temperature, tribology, flow rate, contamination, power, speed), the direction of monitoring activity towards root cause failure modes, the generic approach to setting alarm criteria, and its replacement of ISO 17359:2011. ISO 13374-1:2003, ISO 13374-2, ISO 13374-3:2012 and ISO 13374-4:2015, Condition monitoring and diagnostics of machines, Data processing, communication and presentation, for the four-part open architecture covering general guidelines, data processing, communication and presentation. The 2025 State of Industrial Maintenance report for predictive maintenance as primary strategy at 27% in 2025 against 30% in 2024, preventive at 71%, run to failure at 38%, AI fully or partially implemented at 32%, 26% piloting or evaluating, 65% expecting to adopt AI within twelve months, and stated barriers of budget (25%), lack of expertise (24%) and cybersecurity (22%); respondents could select more than one strategy. Deloitte Analytics Institute position paper on predictive maintenance for uptime gains of 10% to 20%, overall maintenance cost reduction of 5% to 10% and maintenance planning time reduction of 20% to 50%. Updated August 2026.
