Skip to content
allanninal.dev/build
Collaborate

Part 3 of 7 · Packaging data reporter series ~5 min read

Weighing what you cannot weigh

Nobody can weigh three thousand product lines, and nobody is expected to. What is expected is that every number in the submission can say where it came from, and that turns an impossible measurement problem into an ordinary data problem.

Key takeaways

  • Three sources: supplier specification, weighed in house, sampled and modelled.
  • The method and the date travel with the weight. A bare number is not an answer.
  • Weigh empty packaging. Subtracting product weight from a full case compounds two errors.
  • Split composites by material. A card box with a plastic window is two lines.
  • Small per-unit errors are large at volume, and they do not cancel out.

Three ways to get a weight

Three sources for a per-unit packaging weight converging on a tonnageThree boxes on the left feed one box on the right. Supplier specification, from a drawing or a written spec sheet, described as best if it exists. Weighed in house, meaning empty packaging on real scales, described as slow and exact. Sampled and modelled, meaning a representative few applied to the rest, described as unavoidable. All three converge on Per-unit weight in grams, by material and by component, which feeds a box labelled Tonnage: units shipped times the weight. A note says all three are acceptable, that what is not acceptable is a number that cannot say which of the three it is, and that this is what most first submissions contain.Supplier specificationthe drawing, or awritten spec sheetbest, if it existsWeighed in houseempty packaging,on real scalesslow, and exactSampled and modelleda representative few,applied to the restunavoidablePer-unit weightgrams, by material,by componentTonnageunits shipped timesthe weightAll three are acceptable. What is not acceptable is a number that cannot say which of the three it is, and that is what most first submissions contain.
Fig 1. Three routes, one weight. The route is stored beside the number because it is the first thing an auditor asks and the last thing anybody records.

Empty, not full minus product

The shortcut everybody reaches for is to weigh a full case and subtract the declared net weight of the contents. It saves opening anything and it produces a figure that is wrong by the sum of two errors: the tolerance on the fill weight and whatever else is in the case that is neither product nor the component being measured.

For a case of tinned goods that shortcut puts the tin weight into the cardboard line, which is not a small error and is in the wrong material entirely. Weighing the empty case takes thirty seconds more and produces a number that survives being questioned.

How the tonnage was established

Tonnes of packaging grouped by the method that established the weightThree vertical bars showing tonnes of packaging placed on the market grouped by how the weight was established. Supplier data accounts for one thousand one hundred and ninety tonnes. Weighed in house accounts for four hundred and five tonnes. Sampled and modelled accounts for two hundred and eighty-five tonnes. A note says the third bar is the one an auditor asks about first, and also the one that shrinks every year as suppliers answer.0500100015002000~1190Supplier data~405Weighed in house~285Sampled and modelledTonnes placed on the market, by how the weight was establishedThe third bar is the one an auditor asks about first. It is also the one that shrinks every year, as suppliers answer.
Fig 2. The same 1,880 tonnes, grouped by where its weights came from. This chart is the honest summary of a submission’s quality and it takes one query.

The chase list falls out of this

Once weights carry their method, the system can rank suppliers by how much tonnage currently rests on an estimate. That list is short, it is ordered by something real, and sending it to purchasing is the single highest-value output of the first year of running this.

It also reframes the conversation with the supplier. Please send packaging specifications gets ignored. We are currently estimating 61 tonnes of your packaging because we do not have your spec sheets does not, because it is specific and it implies somebody is counting.

Where the error goes

Per-unit weights are in grams and volumes are in millions, and that ratio is what makes small mistakes structural. An outer case recorded forty grams light, across one point two million cases in a year, is forty-eight tonnes of packaging that does not appear in the submission at all.

The temptation is to assume errors cancel, and they do not, because they are not random. A business that estimates conservatively estimates conservatively everywhere, and a business that uses a nominal board weight from a catalogue uses the same catalogue for every line. Systematic method produces systematic error, in one direction, at scale.

So sampling records what was sampled: how many units, when, the mean and the spread. A component whose sample spread is wide is flagged rather than averaged, because a wide spread usually means two different packs are being sold under one code, and that is a data problem rather than a measurement one.

What a weight record holds

  • Grams. Per unit of the component, empty.
  • Material. From the fixed list, with composites split by material.
  • Component. Primary, secondary, transit, or shipment.
  • Method. supplier_spec | weighed | sampled, never blank.
  • Sample detail. n, mean and spread, where the method is sampled.
  • Established on. A date, because packaging changes and specs go stale.
  • Evidence. The spec sheet, the weighing record, or the sample sheet.

Composites and the material list

The material categories are fixed and their boundaries are not always intuitive. A card sleeve with a plastic window is two materials and reports as two lines. A fibre-based composite carton is its own category rather than paper. A steel lid on a glass jar is steel and glass, in the proportions they actually are.

This is tedious and it is also where a surprising amount of tonnage moves between categories that are charged differently. The system therefore models a component as a list of material portions rather than as a single material, which costs one level of nesting and removes an entire class of argument.

The next post is the field that decides whether any of this tonnage costs money: whether it ends up in a household bin or a trade one, and why that is a question about the buyer rather than about the pack.

All posts