The Metric Nobody Tracks: What Share of the Parts You Scrapped Were Still Good?

Most industrial operations can tell you, to two decimal places, what they spent on replacement parts last year. Very few can tell you what proportion of the parts they replaced were actually broken.

That second number is not tracked anywhere, in most companies, by anyone. It has no owner and appears in no report. And it is almost always larger than people expect — which makes it one of the more reliable sources of recoverable margin sitting inside a heavy-industrial business, precisely because nobody is looking at it.

This is a piece about how to find it, using the analysis tools any operations team already has.

The metric that does not exist

Call it the condemnation error rate: of the components taken out of service and scrapped in a given period, what share would have passed inspection if anybody had inspected them?

The reason it goes unmeasured is structural rather than technical. Once a component is scrapped it stops generating data. It leaves the asset register, the physical item is cut up or melted, and the question of whether that was the right call becomes permanently unanswerable. The evidence destroys itself as part of the process. Compare that with a component that fails in service, which generates an incident, an investigation, a report and often a lasting change in procedure.

So one type of error is loud and the other is silent. Organisations respond rationally to that asymmetry by optimising hard against the loud one. Over time, the threshold for condemning a part drifts downward, nobody notices, and the drift never corrects because there is no feedback signal pointing the other way.

Why it costs more than the scrap value suggests

The instinctive way to value a scrapped component is by what it weighs. Steel has a recycling rate that most materials would envy and a well-established market, so the loss feels small — you get some of the money back.

That instinct is right for material and wrong for manufactured items, and the gap between the two gets wider the more work has gone into the part.

 Recycling recovers the grey band and nothing else. For a plain steel plate that is most of the value. For a precision-threaded joint it is under a third — the rest has to be bought a second time.

Melting a component recovers its atoms. It does not recover the machining, the heat treatment, the threading, the inspection or the freight from a mill that is rarely close by. Those have to be paid for again, in money and in emissions, to produce a replacement that is functionally identical to the item that was discarded.

This gives a clean rule for deciding where to look first. Sort your scrap stream by the ratio of manufacturing cost to material cost. Anything near the bottom — plate, bar, structural sections — can be scrapped without much thought, and the existing recycling infrastructure handles it well. Anything near the top deserves a second look before it goes in the skip, and the further up you go, the more expensive each unnecessary condemnation becomes.

The break-even is lower than people assume

The usual objection to inspecting more is that inspection costs money on every item, including the ones that turn out to be genuinely dead. That is true, and it is the right way to frame the question.

But it also makes the analysis simple. Inspection is worth doing when its per-item cost, paid across everything you check, is less than the replacement cost of the items it saves.

 Illustrative, with inspection priced at 18% of replacement. Below that break-even, checking is a waste. Above it, every point of condemnation error is money returned. Substitute your own inspection cost; the shape does not change.

The chart is indicative rather than authoritative — your inspection cost may be higher or lower, and the honest answer is that you should run it with your own figures. But the structure holds regardless of where the numbers land. The break-even sits at whatever fraction of replacement cost inspection consumes, and for expensive components that fraction is small.

Which means the practical question is not "is inspection worth it in general". It is "is our condemnation error rate above or below that threshold" — and almost nobody knows, because nobody measures it.

How to measure it without a project

You do not need a transformation programme. You need one month and a corner of the yard.

Stop scrapping a defined category of component — the one with the worst manufacturing-to-material ratio — and set it aside instead. At the end of the month, inspect the pile properly. Not a visual glance: dimensional checks against specification, and where the component has a functional interface, a controlled test of that interface.

Count what passes. That number is your condemnation error rate for that category, and it is the only figure in this entire analysis that cannot be argued with, because it came from your own stock.

Two warnings about running this. First, do not let the people who made the original condemnation decisions perform the re-inspection; the incentive to be right is too strong. Second, record the results whether they are flattering or not. An operation that discovers its error rate is two per cent has learned something genuinely valuable — it means the existing judgement is good and the money is somewhere else.

Putting numbers on it

Abstract break-evens are easy to nod at and hard to act on, so it is worth walking through the arithmetic with a concrete case. The figures below are illustrative — substitute your own — but the structure is what matters.

Take an operation that condemns 600 high-value components a year, each costing 10,000 currency units to replace. That is a six-million-unit annual replacement bill, which will appear in the accounts as a procurement line and attract the usual pressure to negotiate a better unit price.

Suppose inspection costs 1,800 per component — a deliberately unflattering 18% of replacement — and suppose the condemnation error rate turns out to be 35%, which is within the range operations typically discover when they check for the first time.

Inspect all 600: that is 1,080,000 spent on inspection. Of those, 210 pass and are returned to service; 390 are genuinely dead and get replaced at a cost of 3,900,000. Total outlay: 4,980,000, against 6,000,000 for scrapping everything. The saving is a little over a million units a year, on a process change that requires no new technology and no supplier renegotiation.

Now note what happens to the sensitivity. Halve the error rate to 17.5% and the saving nearly vanishes — you are close to the break-even and the exercise is marginal. Double the inspection cost and the same thing happens. This is genuinely a case where the decision turns on two numbers, one of which you can measure in a month and the other of which you already know.

Which is the actual argument for running the trial rather than debating the policy. The disagreement in most companies is not about the arithmetic. It is that one side assumes the error rate is 5% and the other assumes it is 40%, and neither has checked.

Pricing inspection honestly

One trap worth avoiding: the inspection cost in that calculation should be the fully loaded figure, not the invoice from a service provider.

It includes handling and transport to wherever the work happens, the labour to clean components properly — thread compound and fine debris hide precisely the defects you are looking for — the time the equipment occupies, and the cost of holding items in a yard while they wait. It also includes the components that fail inspection, whose inspection cost was, strictly, wasted.

Counting only the direct testing cost makes the case look better than it is, and inflated business cases tend to be discovered at the worst moment. The version that survives contact with a finance function is the one that already priced in the losses.

There is a corresponding trap on the benefit side. A requalified component is not always worth the full replacement price. It may have less remaining life, or carry restrictions on where it can be used. If your operation tracks remaining life at all, the honest saving is the replacement cost multiplied by the fraction of life the component still has — a smaller number than the headline, and a much more defensible one.

The reporting angle, briefly

There is a second-order benefit that most operations stumble into rather than plan for, and it is worth being clear-eyed about it.

Avoided replacement is avoided manufacturing, and avoided manufacturing is avoided emissions — the machining, the heat treatment and the freight that the earlier chart shows sitting above the material band. For companies with Scope 3 reporting obligations, that is a real reduction in embodied carbon achieved without buying anything or changing any product.

The catch is that it is difficult to claim credibly. Scope 3 accounting captures what you purchased, not what you sensibly declined to purchase, so the benefit shows up as an absence rather than a line item. Demonstrating it requires exactly the records described above: what was inspected, what passed, and what would otherwise have been ordered.

The pragmatic view is that this should be treated as a bonus rather than a justification. Programmes justified primarily on emissions tend not to survive a bad quarter. Programmes justified on avoided capital expenditure, which happen to reduce emissions, survive fine — and that is the honest order of the argument here.

The interface is usually where the argument is settled

For most high-value components, the thing that decides serviceability is not the body of the part but its connection to something else: the threads, the sealing face, the bore. That is where wear concentrates and where damage is hardest to see.

Take threaded oilfield tubulars, which are close to a worst case. A joint is machined to tolerances in the thousandths of an inch across a sealing surface several inches wide, then repeatedly assembled and disassembled in conditions that are nobody's idea of a clean room. A galled sealing face can look almost normal under a torch and still leak under pressure.

Assessing one properly means making the connection up again under controlled conditions and watching what the torque does as it tightens. The horizontal machines built for this work clamp one component, rotate the other, and plot torque against rotation while doing it. The curve is the diagnosis: a clean engagement rises smoothly to a sharp shoulder where the sealing faces meet, a contaminated thread produces a step, and a galled one climbs steadily with no inflection at all — reaching the specified torque purely through friction, having never sealed.

 A real log from a controlled make-up: a long shallow climb while the threads engage, then a near-vertical rise at the shoulder. The tables beneath are what lets the same joint be trusted years later.

Three joints can reach identical final torque with completely different histories. A tally sheet recording only the final figure cannot distinguish them, which is why the final figure has quietly stopped being sufficient evidence in operations that have looked at the data.

Reading those traces is a learned skill rather than an automated one, and it is not evenly distributed across a workforce. It is worth keeping a handful of annotated good and bad examples wherever the machine lives; there is also a fair amount of published material on how the curves are interpreted if you would rather not develop the reference set from scratch.

The same equipment run in reverse matters just as much, because a careless disassembly can destroy a thread that survived years of service intact. Anyone specifying this kind of kit should be looking for a single frame that handles both directions of the operation under the same control rather than treating removal as the crude half of the job.

What the data is worth after the fact

The output of all this is not really a pass or fail. It is a record, and records compound in ways single decisions do not.

A few thousand logged assemblies will tell you which suppliers produce threads that make up cleanly and which do not — a supplier scorecard built from your own operational data rather than from their sales material. They will tell you whether a particular yard, shift or thread compound correlates with problems. They will tell you whether a batch that behaved in January started drifting in June.

None of that requires a model or a platform. It requires that the data was retained in a structured, queryable form with the component identified, which is the step most operations skip. A curve saved as an image in a monthly folder is a picture of information rather than information.

There is also a use that shows up in commercial disputes, and it is often what finally justifies the spend. When a component fails, the question of who pays is settled by whoever has better documentation. Being able to produce the make-up trace for a specific joint, on a specific date, showing a clean shoulder at the specified torque, is a very different position from producing a tally sheet with a number written on it.

Why this does not happen on its own

If the analysis is this straightforward, the obvious question is why the practice is not universal. Three reasons, none of them technical.

The costs and benefits land in different budgets. Inspection is an operating cost incurred now by one team; the saving appears later as an absence of capital expenditure somewhere else. When a cost centre pays and a different one benefits, the cost gets cut first. That is not irrationality — it is an accurate reading of how the accounts are drawn.

The incentives around condemning things are asymmetric, as described above, and asymmetric incentives do not self-correct. They need someone senior enough to say explicitly that a measured error rate is more valuable than a clean record of never having approved a marginal part.

And the history usually is not good enough to support the decision. Requalifying a component honestly means knowing what it was made from, how it was treated, what loads it has seen and how often. In most operations that history is scattered across paper tickets, a spreadsheet on somebody's laptop, and the memory of a supervisor who left last year. Where the history is missing, the safe answer is to scrap — and the safe answer wins by default.

Who should own the number

A metric with no owner does not get measured twice. If the trial produces a useful figure, someone has to be accountable for it thereafter, and the obvious candidates are all slightly wrong.

Maintenance owns the condemnation decision but is judged on failures, so asking them to report their own error rate is asking for a number that trends conveniently downward. Procurement benefits most but has no visibility of the technical call. Finance can see the saving but cannot evaluate a thread.

The arrangement that tends to work is that the technical function measures it and the finance function publishes it, quarterly, alongside the replacement spend it sits against. Not because anyone needs a dashboard, but because a number that appears in a routine report keeps getting produced, and a number that appears in a one-off study does not.

Where to start on Monday

Pick the single component category with the highest ratio of manufacturing cost to material cost. Stop scrapping it for one month. Inspect the pile with someone who did not make the original calls. Count the passes.

If the number is small, you have learned that your judgement is sound and you can stop worrying about this. If it is large — and in operations that have run this exercise it usually is — you have found a recurring cost with no owner, no line item and no defenders, which is the easiest kind of cost there is to remove.

Either way it is one month of deferred scrapping and a day of inspection, against a figure that is currently a matter of opinion. For an analysis this cheap, the surprising thing is not that some operations do it. It is how few of them have ever bothered to check.