Why does predictive maintenance ROI so rarely survive a finance review? Because the number is an estimate, not a measurement, and the missing layer is cost attribution.
Predictive maintenance ROI is real, and it is almost never proven. The number most teams report is an estimate built on industry-average multipliers, not on their own measured costs, which is why it rarely survives a finance review. The missing layer is cost attribution: the data that ties every maintenance dollar, every hour of downtime, and every avoided failure to the specific asset that caused it. A calculator estimates ROI. Attribution proves it.
This article is written from running that attribution layer in production. Tericsoft engineers the platform behind one of the world's larger electric deployments, a 3,000-plus vehicle fleet where component failures are flagged before they strand a trip and every maintenance cost is booked against the vehicle that incurred it.
What the predictive maintenance ROI calculators actually measure
Search for predictive maintenance ROI and you find a wall of calculators. They all run the same formula: savings minus cost, divided by cost. The savings side is built from four familiar buckets. Downtime avoided by turning a breakdown into a scheduled repair. Cheaper proactive fixes, since a planned repair costs a fraction of an emergency one. Leaner parts inventory. Fewer after-hours callouts.
None of that is wrong. The evidence for the benefits of predictive maintenance is genuinely strong. Deloitte's research on predictive technologies for asset maintenance puts the potential at 10 to 20 percent higher equipment uptime and 5 to 10 percent lower maintenance costs, and the US Department of Energy puts reactive repairs at three to five times the cost of planned work.
Here is the problem. Every input in that model is an average borrowed from someone else's operation. The 5 to 10 percent is a band, not your number. The multiplier on emergency repairs is a rule of thumb. Plug those into a spreadsheet and the predictive maintenance cost savings you report describe the industry, not your fleet. That is an estimate wearing the costume of a measurement.
Why the ROI never survives the finance review
Take that estimate to a CFO and watch it come apart. The questions are always the same, and they are fair.
How do you know predictive maintenance software avoided that failure, rather than the component simply lasting longer? If it had failed, how do you know it would have cost what your model assumed? How much of this year's maintenance saving came from prediction, and how much from a newer, healthier fleet? Which of these dollars would you have saved anyway with a decent preventive schedule?
The honest answer to all of them, without attribution, is that you do not know. The ROI rests on a counterfactual, the failures that did not happen, and a counterfactual you cannot measure is a story, not a number. Finance discounts unfalsifiable numbers toward zero, and it is right to.
The problem is not that predictive maintenance lacks value. The problem is that the value is invisible to the ledger that is supposed to approve it.
Cost attribution: the missing layer
Cost attribution is the layer that makes the value visible. It is the discipline of tagging every maintenance event, the parts, the labor, the tow, the downtime, the missed trip, to the specific vehicle, component, and failure mode that produced it, and then linking each prediction to the work order it generated and the outcome that followed.
With it, the vague claim that predictive maintenance saved roughly two million dollars becomes something a CFO can audit. On this vehicle class, the system flagged a number of degrading components two to three weeks early, and a share of them were converted to scheduled service instead of roadside failure.
Each avoided roadside event carries a known cost in your own history: the tow, the expedited part, the overtime, and the trip the vehicle did not run. Net that against the program cost and you have an ROI built from your own attributed events rather than from an industry brochure.
This is also where the distinction between condition based maintenance and true prediction stops being academic. Condition monitoring tells you a part is degrading. Attribution tells you what catching it was worth.
The four links cost attribution has to close
Attribution is not a report you generate at year end. It is a set of joins that have to exist in the data from the moment a fault is detected. Four of them do the work.
- Every cost tagged to an asset and a failure. Maintenance spend cannot sit in one lump line called repair and maintenance. Each dollar has to carry the vehicle, the component, and the failure mode, so the ledger can be sliced by cause, not just by month.
- The prediction linked to the work order. When the system flags a degradation, that alert has to connect to the work order it raised and the service that closed it. Without that thread, you can never prove the catch was the catch.
- The counterfactual priced from your own data. The cost of the failure you prevented has to come from your fleet's actual history of that failure, the real roadside events on the same component, not a generic multiplier. Your averages, measured, are the only ones a review will accept.
- The revenue impact carried through. Downtime is not just a repair cost. It is the trip the vehicle was booked to run. Attribution has to follow the asset into the profit-and-loss, so an hour off the road lands against the contract it belonged to.
What running attribution on 3,000 vehicles taught us
The reason we can write this from experience rather than theory is that the fleet behind this article runs on one data model where operations and maintenance are not separate systems. Telematics events auto-generate work orders, so a fault on the road becomes a scheduled service rather than a note nobody actions. Component failures, including battery-signal anomalies, are flagged 2 to 3 weeks before breakdown and escalated automatically into workshop tickets.
Because every one of those events is booked against a vehicle and a failure mode, fleet maintenance costs stop being a monthly lump and resolve to the vehicle, the failure, and the trip they protected. The lesson is not that prediction pays back. Everyone claims that. The lesson is that a fleet with attribution can prove which repairs paid back and which did not, and a fleet without it is left arguing from averages. One version gets funded again. The other gets questioned every budget cycle.
How to build an ROI you can actually defend
The order matters. Most programs buy the sensors and the software first and try to reconstruct the savings later, which is exactly the reconstruction that fails in front of finance. Attribution has to come first.
Instrument the cost data before the pilot, so maintenance spend is already tagged to asset and failure on day one. Baseline from your own history rather than an industry table.
From there, the maintenance ROI calculation is a chain, not a lump sum: prediction, to work order, to avoided failure, to attributed cost. Net it against the full program cost, including the analyst hours nobody budgets for. Then report the return per vehicle class, because a fleet-wide average hides the classes where prediction is winning and the ones where it is not yet worth it.
There is an honest floor here. Below a certain scale, or without the data layer to attribute anything, a calculator estimate is a perfectly reasonable way to justify a pilot. What it cannot do is prove the return once the program is real. For that, the estimate has to become a measurement, and a measurement needs attribution.
Key takeaways
- The ROI you report is an estimate until it is attributed. Calculators borrow industry averages. Proof comes from your own measured costs.
- Finance is right to discount an unfalsifiable number. A saving built on failures that did not happen is a story until the data can trace it.
- Cost attribution is the missing layer. It ties every maintenance dollar, downtime hour, and avoided failure to the asset and failure mode that caused it.
- Attribution has to come first. Instrument the cost data before the pilot, not after, or the savings cannot be reconstructed in a way a review will accept.
About Tericsoft
Tericsoft does not sell a sensor or a dashboard with a monthly price per asset. We build the data layer underneath a fleet, the model where a predicted failure, the work order it raises, the cost it avoids, and the trip it protects are all the same record rather than four disconnected ones. That is the layer that turns a predictive maintenance program from a line item leadership takes on faith into a return the finance team can audit. If your maintenance program works but you cannot prove what it is worth, that gap between the value and the ledger is the problem we engineer for.
The return from catching failures early: less downtime, cheaper planned repairs, leaner inventory, fewer emergencies, net of program cost.
Savings minus program cost, over cost. Build savings as a chain: prediction, work order, avoided failure, priced from your own data.
It rests on a counterfactual: failures that did not happen. Without attribution linking predictions to avoided costs, it stays unprovable.
Tagging every maintenance cost, parts, labor, downtime, lost trip, to the specific vehicle, component, and failure mode that caused it.
It depends on scale, failure costs, and attribution data. At fleet scale with attribution you can prove which repairs actually paid back.
