Your Asset Utilisation KPI Is Solving Yesterday's Problem
Squeezing maximum uptime out of every expensive machine made sense when the machine was the scarce thing. Collapse the cost of building another one and redundancy beats utilisation: cheaper, safer, and nobody has to solve fast turnaround at all.
The proudest number in most operations reviews is the asset utilisation rate: the share of time each expensive machine spends earning its keep. High utilisation reads as discipline. Idle kit reads as waste. Whole careers in capacity planning and maintenance scheduling are built on pushing that percentage up.
But the KPI smuggles in an assumption: that the asset is the scarce thing. For most of industrial history it was, so the assumption never needed stating. It is now failing across several sectors at once, and the curves are public. Lithium-ion battery packs cost well over $1,000 per kilowatt-hour in 2010; BloombergNEF's annual price survey put the 2023 average at $139. In launch, analysis by CSIS's Aerospace Security Project puts the Space Shuttle's cost to low Earth orbit at roughly $54,500 per kilogram in 2021 dollars, against about $2,600 for a Falcon 9. Batteries and launch are simply the best-documented instances of a general repricing: when the cost of building or flying another unit falls by an order of magnitude, the constraint moves. The busy-asset playbook keeps optimising a scarcity that no longer exists.
What is asset utilisation actually a proxy for?
What a capital asset is really for is availability: the ability to do the job whenever the job turns up. Availability has exactly two levers. You can turn one unit around faster, or you can have more units. When a unit costs a fortune, the second lever is welded shut, so engineering culture poured a century of ingenuity into the first: tighter maintenance windows, condition monitoring, hot-swappable everything. That effort was rational. It was also contingent on the price of the machine, and contingent truths come with expiry dates.
Cheapen the unit and the second lever swings open. Five units cycling slowly can deliver the same effective capacity as one unit cycling frantically, with slack left over for failures, demand spikes and unhurried inspection. The number that matters flips from time between uses to units built, and slow, thorough maintenance becomes something you can simply afford.
Sixty-year-old maths, freshly broken targets
To be clear about what is and isn't new here: the case for redundancy is old. Reliability engineering has priced spares this way since the 1960s, and Barlow and Proschan's 1965 Mathematical Theory of Reliability treats redundancy allocation as a solved optimisation: buy standby units until the next spare costs more than the failures it prevents. Queueing theory is blunter still. Kingman's 1961 heavy-traffic formula shows waiting times growing roughly in proportion to utilisation divided by spare capacity, which is why queues quietly explode once utilisation climbs past about 80% and why seasoned capacity planners have always left headroom.
What the textbooks assumed, reasonably for their era, was that the price of a unit stays roughly put across a planning horizon. That assumption is the casualty. The decision rule the textbooks imply is one line long: buy the spare when the unit costs less than the expected annual cost of downtime beyond your availability commitment, multiplied by the probability of the breach. Every term in that rule is slow-moving except unit cost, and when unit cost falls 80% inside a single depreciation schedule the crossover between sweating one asset and buying a fleet doesn't drift, it lurches. A target set even five years ago encodes a dead price. Run the rule on a plausible case. Take an asset that cost £600,000 when the target was set, downtime worth £5,000 an hour, and an availability commitment of 99.5%, which permits about 44 hours of downtime a year. A single well-maintained unit achieving 98% availability is down around 175 hours a year, so a breach is close to certain and the rule collapses to the cost of the excess: 131 hours over the commitment, roughly £655,000 of exposure a year. That is the crossover price for the spare. Against it, £600,000 was a genuinely marginal call, arguable either way once you admit how soft the estimates are, and most boards argued for utilisation. If the same unit now costs £120,000, nothing marginal remains: refusing the spare is a £655,000 bet to save £120,000, better than five to one, and a second unit on rotation closes the gap entirely with inspection time left over. Nothing in the reliability textbooks changed. One input fell by 80%, and the answer crossed the line.
The cleanest large-scale demonstration comes from the data centre. In the early 2000s the orthodoxy for serious computing was fault-tolerant enterprise hardware: machines engineered so that an individual unit almost never failed, priced to match, and repaired urgently when it did. Google built its infrastructure on the opposite bet. The Google File System paper, published in 2003, states plainly that component failures were treated as the norm rather than the exception, with the whole system designed around inexpensive commodity machines that fail constantly. Failed servers wait for scheduled repair rounds while software routes work to their neighbours. Notice what Google never solved: it did not build an individual server that refuses to die, and it did not make fixing one fast. Cheap, numerous units meant neither problem needed solving, and the urgency that makes hurried intervention error-prone disappeared along with the scarcity. Unit reliability and service reliability turned out to be separable objectives that the expensive-hardware era had bundled together.
When does redundancy beat high asset utilisation?
Redundancy wins when three conditions hold at once. Unit manufacturing cost is falling steeply, so a spare is cheap relative to history. The cost of failure is high relative to the unit, so an outage or an accident costs more than the spare that would have prevented it. And what the business actually sells is availability rather than the continuous output of one specific machine. Where all three hold, redundancy is simply the cheaper way to buy the thing utilisation was always a proxy for.
Compute hardware is walking into this crossover on a documented curve: Epoch AI's 2022 analysis of GPU price-performance found that the floating-point operations per second a dollar buys have doubled roughly every two to two and a half years since 2006. Teams sweating every accelerator at maximum occupancy are optimising the metric of a scarce-GPU era even as that curve commoditises the fleet. Provisioning headroom is often the cheaper way to honour a latency or delivery promise, and that arithmetic belongs in the decision before any build commitment (the working is set out in AI readiness before build). Software teams make the same mistake in miniature: environments cost almost nothing to duplicate, yet organisations still queue engineers behind a single staging server, which is utilisation thinking applied to an asset with a marginal cost of roughly zero. Getting platform economics right is mostly a matter of noticing which era's constraint you are pricing.
Now the contrarian test, because the reframe has limits. Where the unit stays genuinely scarce (airport slots, urban land, cutting-edge fabrication capacity, senior people), utilisation keeps its throne. And fleets rot: a spare that is never exercised is a spare you can't trust, so redundancy needs a rotation discipline of its own, or your insurance policy is a museum. None of this makes utilisation wrong. It makes utilisation a derived metric rather than a first principle, one that has to be re-derived every time unit cost moves an order of magnitude.
The second-order consequences land on decisions being made this quarter. Maintenance contracts priced on intervention speed lose their premium when nothing is urgent. Depreciation schedules built for sweated assets misprice a rotating fleet. Incentives lag worst of all: an operations director bonused on utilisation will fight fleet redundancy with every spreadsheet at their disposal, and by their KPI they will be right. If your board still applauds a utilisation number, the useful question is what unit cost was assumed when the target was set, and whether the curves above have already broken it. That question belongs in capex strategy rather than procurement, and it is the assumption audit technical strategy engagements exist to force into the open.
Utilisation was never the goal. It was a proxy for making the most of a scarce machine, and it worked while the machine stayed scarce. In a growing number of sectors the machine has stopped being scarce, and a KPI that outlives its own premise isn't discipline; it's nostalgia with a dashboard.
Questions people ask
Is high asset utilisation always a good sign?
No. It signals efficiency only while the asset is the binding constraint. Run near 100% with no slack and you have fragility dressed up as discipline: no headroom for failures, inspection or demand spikes, and queueing theory shows wait times exploding well before you reach full occupancy. If unit manufacturing costs have fallen sharply since the utilisation target was set, a very high number may actually indicate under-investment in fleet capacity.
Can a reliable service be built from unreliable machines?
Yes, and it is often the cheaper route. Redundancy converts unit failures into a scheduling detail: spare capacity carries the load while failed units queue for unhurried repair. Google's early data centres proved the pattern at scale, delivering continuous service from commodity servers that failed constantly, without ever engineering an individual machine that didn't.
How do falling manufacturing costs change maintenance strategy?
They turn time back into the maintenance team's ally. With a rotating fleet, units come out of service for thorough, unhurried inspection while spares carry the load, so condition-based maintenance stops racing a turnaround clock. Budgets shift from paying a premium for intervention speed towards funding fleet depth and a rotation discipline that keeps spares trustworthy.
Related
- Retention Bonuses Don't Reduce Key Person Risk. Doctrine Does.
- Bot Traffic Is Splitting Your Ad Budget, and a Human Signed It Off
- Your Cloud Provider Is Financing Its Own AI Revenue. Do the Financial Due Diligence
- Data & Strategy
Written by an AI editorial persona of Abyshire's proprietary editorial system and reviewed by our team.