Scrap control: measuring, attributing and acting on material lost in production

Scrap you cannot attribute is scrap you cannot reduce

The critical field is not how much but where it was made wrong. A part rejected at final inspection may have been ruined at the second operation, and recording it against final inspection sends every improvement effort to the place that merely found the problem. Attribution needs the person doing the disposition to name the operation that generated the defect, which sometimes takes a metallurgist or a quick sectioning rather than a guess. Where the answer genuinely cannot be determined, record that honestly as unattributed — a growing unattributed share is itself a finding, and usually means inspection sits too far downstream of the work.

The code list and its predictable failure modes

Scrap codes drift toward uselessness in two directions. Some plants keep a handful of broad categories so everything lands in dimensional or cosmetic and no mechanism is identifiable. Others keep hundreds, so operators pick whichever appears first and the distribution reflects the ordering of a menu. A workable list names failure mechanisms in the language people on the line use, keeps categories mutually exclusive, and is reviewed after a few months of real use to prune what nobody selects and split what absorbs too much. Adding a code without retiring one is how the list gets to hundreds in the first place.

Value it where it was lost, not by material weight

A casting scrapped before machining costs the metal. The same casting scrapped after machining, heat treatment and coating costs the metal plus everything spent since, plus the capacity consumed at every operation it passed through, plus whatever the plant now has to expedite to replace it. Valuing scrap at raw material cost makes late-route losses look trivial and misdirects the improvement budget. Cost it at the accumulated value at the point of loss, and where the operation was at the constraint, add the contribution the plant could not earn during that time. That figure is what wins the argument for a fix.

Who is allowed to scrap, and the bin that eats the evidence

If any operator can drop a part into a skip, the plant loses both its count and its evidence. Scrap should be segregated, tagged with the order and operation, and dispositioned by someone with the authority to decide, so that parts can be examined before they are destroyed. This costs a little floor space and a little time and it is what makes a proper investigation possible at all. It also stops the quiet practice of remaking a part off the books to keep a shift's figures clean, which leaves the plant with an unexplained material variance and no record that a problem occurred.

From the record to an actual change on the floor

Rank losses by value rather than by count, take the top few, and give each a named owner with a date. Verification is the part that gets skipped: after the countermeasure, the scrap rate for that specific code at that specific operation has to be checked against the level before, or the plant accumulates closed actions and unchanged losses. Waste-elimination methodology as a discipline belongs elsewhere; what is needed here is narrower and harder — that this month's largest coded loss has an owner who will be asked, in public, what changed.

Frequently asked questions

Should setup scrap be recorded separately from running scrap?

Always, because they have completely different causes and owners. Setup losses respond to changeover method, tooling condition and first-off approval practice; running losses respond to process control, material variation and tool wear. Merging them produces a figure that moves whenever the product mix changes and tells nobody anything. Splitting them also makes the true cost of frequent changeovers visible, which is often the piece missing from an argument about batch sizes.

How do we handle scrap that a supplier caused?

Record it against the operation where it was discovered, code it as incoming material defect, and hold the physical evidence with the lot identity attached. Recovering cost from the supplier is a commercial matter that depends entirely on having that evidence and the traceability to the delivery. Internally, the more useful question is why it reached production at all — repeated supplier-caused scrap found on the line usually means goods-in checks are looking at the wrong characteristic, or none.

Is a target scrap rate a good idea?

A target is fine as a planning assumption and dangerous as a performance goal, because a rate that is being met stops attracting attention. Standard allowances built into costing should reflect what the process currently does, and they should be reviewed downward as improvements land. What works better than a rate target is a value target with named projects attached, since it forces a conversation about which losses matter rather than an average that hides them.