Predictive Maintenance Savings: What’s Real (and What Isn’t)

A capital request will quote “70% fewer breakdowns” and attribute it to the Department of Energy. An operations slide will put McKinsey’s 30–50% downtime band next to a hospital panel as if it were last quarter’s result. Both treat a number that belongs to someone else’s factory — or to a potential — as a finding about this building.

The savings that survive a budget meeting are narrower, named, and hedged. Deloitte’s manufacturing-oriented work on predictive technologies for asset maintenance puts PdM at 5–10% lower maintenance cost and 10–20% better uptime. DOE/FEMP estimates 8–12% versus a preventive program. PwC and Mainnovation’s 2018 European survey reported averages of −12% cost and +9% uptime among implementers. Those are the bands this note will use. Everything else is a potential, a different job, or a warning.

This field note is about what those figures are actually of — and what they are not. Circuit-level energy monitoring is the measurement job underneath one technique in this family. It is not a substitute for a maintenance program, and it is not a BMS. The job split is in circuit-level monitoring vs BMS vs EMS.

What the words mean

Three labels get stacked as if they were one purchase.

Condition-based maintenance (CBM) is the decision rule: act on measured condition, not on a calendar alone. Oil analysis, vibration, thermography, and electrical signatures are techniques under that rule.

Predictive maintenance (PdM) is a CBM approach that uses condition data — and often a model — to estimate a failure window or remaining useful life, so work can be scheduled before the break. PdM is not a sensor. It is the program that turns a condition signal into a planned intervention.

Continuous electrical monitoring is one CBM technique: observing current, voltage, and power on a circuit over time so change is visible. It can feed CBM or PdM. It is not, by itself, a PdM program, and it is not a claim that every circuit has a failure model.

The rest of this note treats industrial Deloitte and PwC figures as analogs for what a mature PdM program has reported in manufacturing-heavy settings. They are not DeepEnergy case results.

Maintenance cost: the band that holds

Start with cost, and keep energy out of it.

Deloitte’s Industry 4.0 / smart-factory analyses of predictive technologies for asset maintenance cite 5–10% lower overall maintenance cost and 10–20% higher equipment uptime and availability. Those ranges come from Deloitte’s published industrial work, not from a circuit-level electrical product and not from a hospital CBM study. They are the conservative band we prefer when someone asks what PdM has been worth on the maintenance ledger.

DOE’s Federal Energy Management Program, in Operations & Maintenance Best Practices (Release 3.0), estimates that a properly functioning predictive maintenance program can provide 8–12% cost savings over a program that uses preventive maintenance alone. That is a vs-PM estimate from a federal O&M guide — “past studies have estimated” — not a site guarantee, and not a field study of electrical CBM in commercial buildings. FEMP also notes that a site still leaning on reactive maintenance can see larger opportunities than that 8–12% vs-PM band; that is a starting-point hedge, not a second DOE average.

PwC and Mainnovation’s 2018 Predictive Maintenance 4.0 survey covered about 268 companies in Belgium, Germany, and the Netherlands, manufacturing-heavy. Among implementers they reported averages of −12% cost and +9% uptime. Those are survey averages from EU pilots and programs already using PdM 4.0 — not a controlled US field trial, and not a DeepEnergy result.

None of these three is a promise for your panel. They are named analog bands for the maintenance conversation.

Uptime: potential is not observed

Uptime is where slides get loose.

McKinsey has published a 30–50% downtime band for predictive or analytics-driven maintenance. In this note that band is potential only. It is not an observed result you can drop onto a hospital riser or a commercial panel. The observed ranges McKinsey has also cited are more modest and should be labeled as such: +5–15% availability and −18–25% maintenance cost — still analog, still not a site guarantee, still not a DeepEnergy case.

Deloitte’s 10–20% uptime and PwC’s +9% uptime (2018 implementer average, EU, n≈268) sit in the reported column, with the hedges already attached. Use them that way.

What to use, and what not to

The comparison is a use rule, not a ranking of studies.

Energy is a different conversation

Maintenance savings and energy savings are different jobs. Mixing them is how a PdM slide inherits a building-operations number.

PECI’s 5–20% energy figure belongs here, and only here. Portland Energy Conservation, Inc. estimated that low-cost building operations and maintenance improvements — scheduling, setpoints, calibration, balancing, repairs — can save 5–20% of a building’s annual energy bill. DOE/FEMP repeats that band for O&M programs targeting energy efficiency, not as a PdM result. It is building O&M and operations. It is not predictive maintenance, and it is not a CBM energy percentage.

PwC’s 2018 survey is useful on motive, and silent on magnitude. Energy was rarely why a company started PdM — about 1% named energy reduction as the primary reason. Later, 36% of users said they saw some energy benefit. The survey does not publish an average energy-savings percentage. This note will not invent one.

If a motor is wasting energy because of a condition you can see electrically, that is a measurement finding on that circuit — see circuit-level energy monitoring — not a license to apply PECI’s 5–20% to a PdM business case.

Electrical rooms, facilities, hospitals

This is the section that refuses a percentage.

There is no hospital CBM savings percentage in this note, and no electrical-maintenance CBM percentage, because we do not have a named field study that produced one. Inventing one from the industrial analogs is the error.

NFPA 70B is a standard for electrical equipment maintenance. It sets a condition-based, risk-informed interval framework. It is a compliance and risk document. It is not an ROI percentage, and it does not underwrite Deloitte’s 5–10% or PwC’s −12%.

Nowlan and Heap’s reliability-centered maintenance work found that about 89% of failure patterns have no wear-out zone — age is a poor predictor for most items. That is why condition monitoring is a rational strategy. It is not “89% savings,” and it is not a hospital result.

Lawrence Berkeley National Laboratory’s updated value-of-service work (Sullivan, Schellenberg, and Blundell) puts a one-hour interruption for a typical medium or large C&I customer at about $17,800 in 2013 dollars. That is an outage cost, averaged across a utility-customer class. It is not a CBM ROI, and it is not a hospital clinical-loss figure.

The honest frame for a hospital or a facilities electrical program is the job: continuous observation of circuits that matter, so a change is visible before it is a work order or an outage. The industrial percentages remain analogs.

Where DeepEnergy fits

DeepEnergy is continuous electrical monitoring as a CBM technique: circuit-level hardware and the second-by-second traces that come off it. It observes. It does not replace a BMS, a CMMS, or a PdM program. The Deloitte, DOE/FEMP, and PwC bands above are industrial analogs for what mature PdM programs have reported. They are not DeepEnergy customer percentages. We do not have those to publish.

Once that observation exists, Joule AI is the natural-language interface to the measured data. VEM is the managed-outcomes layer that acts on it. This note is not about those two.

To see the measurement layer on a panel, book a demo. Product detail is on DeepEnergy. To start from a bill, use the calculator.