Your Scrap Rate Is an Allocation, Not a Measurement

The number everybody already agreed on

A mid-market frozen and prepared foods plant reported a four percent scrap rate. Nobody disputed it. The finance team had it, the operations team had it, and it appeared in the materials the leadership group reviewed. It was the most settled number in the building.

It was also useless, and for a reason worth naming precisely. That four percent came from the only arithmetic available: material issued, less product shipped, divided out across volume. The plant was not recording production to tolerance. There were no upper and lower limits captured at the fill points, so there was no distribution anywhere in the system. The team believed the lines were dialed in because nothing in their data was in a position to tell them otherwise.

Then the assessment interviews surfaced the floor version. Material was accumulating in places material should not accumulate. Four percent is a lot for food manufacturing, and the plant's own continuous improvement lead, asked for their priorities without prompting, put loss at the fill points in the top three alongside labor and downtime. They named three areas: product fillers, the hoppers feeding them, and the label and box makers downstream. All three were heavily manual.

That is the gap this article is about. Not a wrong number. A number with no shape.

Why an average cannot be attacked

A plant-level loss rate answers one question: how much did we lose. Every decision that follows needs a different question answered: where. Those are not the same measurement, and the first does not contain the second.

Run the two readings that a single four percent permits. In the first, three items with a difficult fill profile are losing twelve to fifteen percent each, and the rest of the plant is clean. In the second, every item on the book is giving away two or three points because the fill targets sit comfortably above the labeled minimum and nothing checks the upper bound. Identical reported rate. Opposite problems. Opposite fixes. Opposite capital.

Aim at the first problem as though it were the second and you buy plant-wide instrumentation to solve three items. Aim at the second as though it were the first and you chase three items while the whole book keeps leaking. The reported average is compatible with both, which means it cannot be used to choose, which means any project justified on it is justified on an assumption nobody has written down.

There is a harder corollary. Loss that becomes scrap is at least visible as mass on the floor. Loss that ships is not. Product given away inside the package never appears in a scrap number at all; it leaves the building as cost of goods and arrives at the customer as a slightly better deal than the one that was priced. A plant with a four percent scrap rate and systematic overfill is running two losses, and its single reported rate can only see the one that stays behind.

What the floor was already saying

The useful detail in that plant was not the four percent. It was that the three loss hotspots its own continuous improvement lead named were the three most manual stations in the process.

That is a pattern worth generalizing, because manual fill is where variance enters a process and where measurement is hardest to retrofit. An operator working a hopper and a filler is holding a target in their head and correcting against product appearance. They will hit an average. They will not hold a band, because nothing tells them where the band is, and nothing records whether they stayed inside it. Variability at those stations is not an operator problem. It is the predictable output of running a manual process to a target with no limits.

My rule on this is simple, and I apply it before any yield work: a loss number I cannot break down by station is a budget line, not a finding, and I will not let it size a project. Finding the loss and locating it are two different pieces of work, and skipping the second is how a plant ends up with a funded initiative aimed at an average.

Instrument the stations, then fund the project

The sequence matters more than the tooling.

Pick the stations the floor already suspects, not the ones with the best existing data. The plant in question had three named for free by the person closest to the work. Set a written target at each with an upper and lower limit, defined against the labeled minimum and the specification, so there is a band and not just a number. Then record to it for one full production week, long enough to cover the normal item mix rather than one favorable run.

What comes back is a distribution per station and per item, and that is the first artifact in this whole chain that can actually carry a decision. Now you can see whether the loss is concentrated or spread. Now the two readings separate. Now the project can be sized against the stations that hold the loss, and the cost per unit of the fix can be compared against the giveaway it recovers rather than against a plant average that was never the target.

Only then does the labor question become answerable, and it is usually the real one. If a station needs an operator holding a band they have no instrument for, the choice is to give them the instrument or to take the band out of their hands. Those are different investments with different payback, and the distribution is what tells you which station deserves which. A capital case for automating a manual fill point that rests on a plant-wide scrap rate is a case built on a number that cannot see the station being automated.

What a well-measured line reads like

Every fill, portion, and forming point has a written target with an upper and a lower limit posted at the station. Loss is reported weekly by station and by item, and the plant can produce that breakdown on request without a special study. Each station's recorded distribution covers at least one full production week of normal item mix. Line speed and yield loss are read together, because a station that holds its band only by slowing down has moved the cost rather than removed it. No yield or automation project is approved without the distribution for its own station attached to the request.

The first measurement to take

Ask your operations leader for last month's loss by station and by item, with the limits each station ran against. The answer is the diagnostic. If it arrives as a breakdown, read it and aim. If it arrives as one plant number, you have not been told how much you are losing; you have been told how much you cannot see, and the next move is one production week of recording limits at the stations your floor already named.