The 40% you already paid for

Ask a content team what a piece costs and you will get a division: monthly spend over monthly output. It is the number on the board, the number in the deck, and the number a vendor is measured against. It is also the wrong number, and the gap is not small.

On the account in our four-month production review, May’s spend was $2,995.71 and May’s output was 400 pieces. That is $7.49 a piece and it is what the board saw. Only 240 of those pieces cleared review. The real cost of a piece the client could publish was $12.48 — 67% higher than the number anybody was working from.

Accuracy is a cost multiplier wearing a quality badge

Right first time gets filed under quality. It sits in a different report from the budget, gets discussed in a different meeting, and is reported as a percentage next to other percentages. That filing is the mistake, because arithmetically it is not a quality metric at all. It is a divisor on everything you spend.

Read the first column alone and the improvement is 2.43×. Read the third and it is 3.99×. Same spend, same output, same four months. The 1.64× between them is not a rhetorical flourish; it is the part of the budget that was buying work nobody ever published.

Notice what happens between the second row and the third. From May to June, accuracy did not move at all — 60% both months. The team got faster and cheaper per piece produced, and the waste ratio sat exactly where it was. Speed alone does not touch this line. It is entirely possible to make a team 32% cheaper per piece and improve the cost of publishable work by precisely the same 32%, having fixed nothing.

Why the waste stayed at 40% while everything else moved

Rework in a hand-built content operation is not one problem. It is a dozen small ones that happen to land in the same bucket: a script that drifts off brief, a voiceover with the wrong emphasis on a term of art, an intro that runs long, a thumbnail that fails a platform check, a hashtag block that trips a filter, a claim that the subject-matter expert will not sign off.

Each is individually fixable and none of them stay fixed, because the fix lives in a person’s head and the next piece is made by a different person on a different day. The failure rate is a property of the process, not of the staff, which is why it did not move when the team got smaller and faster.

A pipeline changes the shape of that problem rather than the difficulty of it. When the same steps run the same way every time, a rejection is diagnosable: it happened at a known stage, with known inputs, and the correction goes into the stage rather than into somebody’s memory. Fix the intro-length rule once and every subsequent intro is the right length. That is the mechanism behind 60% → 98.5%, and it is unremarkable. It is just that determinism compounds and good intentions do not.

The honest counter-argument

There is a version of this where we are overstating it, and it deserves to be on the page rather than in a footnote.

The $12.48 figure assumes the failed 40% was discarded. If the team absorbed its rework inside the same month, on the same salaries, then the work was re-done rather than thrown away and the money gap is narrower than the third column suggests. May’s true cost per usable piece would sit somewhere between $7.49 and $12.48 depending on how much of that 40% was salvage rather than scrap.

We do not have a clean measurement of the split, so here is the comparison at both bounds:

The truth is somewhere between 2.39× and 3.99×, and we would rather publish the range than pick the flattering end of it. The time figures are unaffected either way, because rework hours were already inside the 1,512 person-hours the month consumed — you cannot rework something for free even when the salary is fixed.

Note also that the lower bound still describes a team more than twice as efficient at producing publishable work, with the accuracy improvement thrown in for nothing. The argument does not need the flattering number.

What to do with this on Monday

You do not need a system to start measuring this, and you should not buy one on the strength of somebody else’s percentage. Three things are worth doing regardless of what you build:

• Count rejections at the deliverable, not at the draft. A piece that went back twice and shipped is not a success with an asterisk; it is one usable piece that cost three times what the model says.
• Publish cost per usable piece next to cost per piece. If the two are close together, your process is healthy and automation will buy you speed rather than money. If they are far apart, you have found where the budget is going, and it is not where anybody thinks.
• Ask which stage the rejections come from. If they cluster, they are fixable in the process. If they scatter evenly, they are a briefing problem and no pipeline will help you.

That last one is the honest filter. On the account above the rejections clustered, which is why the number moved. We have seen briefs where they do not, and the right advice there is to fix the brief before anybody builds anything — which is why the fit conversation comes before any quote.

The figures in this report come from one account’s records and are set out in full, with their estimates named as estimates, in the production review.