Why Predictive Maintenance Isn’t the Right Answer for Every Machine

Predictive maintenance has well-documented returns. McKinsey and the U.S. Department of Energy have documented 10:1 to 30:1 ROI on predictive maintenance investment within 12–18 months for high-criticality assets. But the same facilities chasing that return often apply predictive monitoring uniformly across every asset, including ones where it doesn’t make financial sense. The strategy isn’t the problem. Applying one strategy to every machine is.

What predictive maintenance requires

The upfront cost gets underweighted in most pitches for the technology. Implementing condition monitoring equipment can be expensive up front. Predictive programs require sophisticated data management tools and skilled staff capable of interpreting what the data is reporting.

The full benefits of a predictive program typically take 24–36 months to materialize. Integration with legacy equipment that wasn’t designed for condition monitoring adds complexity that’s easy to underestimate during planning.

Where predictive maintenance earns its cost

The asset profile where the investment pays off shares a few characteristics. Unplanned downtime in process industries can reach $2 million per hour, which is the kind of cost that makes early warning worth almost any monitoring investment.

High-criticality, high-consequence assets are where best-in-class operations concentrate predictive maintenance, rather than spreading it evenly across the asset list. The strongest candidates share these traits:

• Failure stops production rather than just degrading output
• Replacement parts are expensive or slow to source
• Failure creates a safety risk, not just a production risk
• The equipment has a well-understood degradation pattern that sensors can reliably detect

Where predictive maintenance isn’t the right fit

Reactive maintenance is the correct strategy for non-critical assets where the failure consequence is genuinely low, replacement is fast and inexpensive, and there’s no safety or compliance exposure, such as office lighting, simple tools, and redundant non-production components. Low-criticality equipment with cost-prohibitive preventive maintenance economics is a legitimate candidate for run-to-failure, not a maintenance shortfall.

Equipment without a well-understood failure mode that sensors can reliably detect is another poor candidate. Monitoring something without a clear degradation signature produces data without diagnostic value. Redundant systems, where a failure shifts load elsewhere rather than stopping production, often don’t justify the investment either.

Building the right mix

Top-performing manufacturing plants target roughly 25–35% predictive maintenance, 50–60% preventive maintenance, and less than 20% reactive maintenance across their asset portfolio. Most facilities today are inverted from that target, with reactive maintenance dominating despite having more critical assets than that approach reflects.

The starting point is a criticality ranking — production impact, replacement cost, and safety exposure — rather than a technology decision made in isolation. The mix should shift over time as assets age, production needs change, and maintenance data matures. It’s an ongoing classification, not a one-time setup.

Predictive maintenance is one tool among three

Predictive maintenance is the right tool for a specific category of equipment. Applying it everywhere wastes money on low-stakes assets. Applying it nowhere leaves high-stakes assets exposed to the kind of failure predictive monitoring is built to catch early. The right question isn’t whether to adopt predictive maintenance but which machines justify it.