A batch is one recipe run across a shelf of products: same format, same look, one lot per SKU, one number at the end. Up to twelve products at a time.
Five stages, and the line between the fourth and the fifth is the whole point of the shape.
Everything up to and including creating the lots is free.
You can build a twelve-SKU campaign, read every shot list, see the exact total, and decide not to render any of it. Nothing has been spent until somebody renders.
1 · The library
A batch takes product ids, not twelve briefs. So it starts with making sure the products exist, which is usually a matter of checking, because most of the time they already do.
For anything missing, read the product page into a saved product with its photography. That is the fastest route from a list of URLs to a library. Anything you were given as text, save by hand.
Pay attention to images here. A SKU with no usable still will be refused by any vertical that requires one, and it is much better to find that gap now than in a refusal list at stage three.
More than twelve SKUs is more than one batch. Split them deliberately and say so, quietly dropping four is the failure mode that shape invites.
2 · The mix
Everything you choose here applies to every SKU, which is what makes it a campaign rather than a pile of lots.
| Decide | Note |
|---|---|
| Marketing or UGC | Photodump and Explainer take no product, so there is nothing to iterate, those are single lots, not batches |
| The format, or the structure | UGC also takes a hook and a platform, and the platform decides the runtime and the aspect |
| The vertical | Read what it requires before proposing it |
| The capture style | If the campaign has one look across it. It composes with the vertical rather than replacing it |
| The brand | Usually name none, each product brings whatever kit it is filed under, which is what a multi-brand library wants |
A UGC campaign needs a person for every SKU, and it should be the same person. One creator, cast once, is what holds one face across the whole campaign, a description renders a different person in every lot.
Do not agonise over provider or resolution unless somebody raised cost or a platform spec. The defaults are the answer.
3 · Dry run, always
This plans every SKU and creates nothing. Read it for three things, in this order:
- The refusals. Which SKUs the vertical turned down and why.
- The shot lists. Whether the beats are the campaign you described.
- The total. What it costs if every shot renders once.
Refusals before the total, deliberately. A quote for eleven SKUs when somebody asked for twelve is a wrong number quietly, and it is the easiest way to under-quote a campaign by a twelfth.
4 · Build
The same run with the dry run off. A refused SKU comes back as a named refusal beside the lots that were made, so a partial batch is a result rather than an error.
There is no rollback.
If a batch stops partway, the lots it already created are real. Nothing undoes them, and deleting them is a decision rather than cleanup, the shot graph is the record of what was billed, so archive rather than delete unless you mean it.
Then read one back. The brief on the first lot tells you whether the shape you quoted is the shape you built, which is a thirty-second check that catches a wrong format across twelve lots.
A null total is not zero. It means a lot that exists could not be priced, usually an uncatalogued model, and the useful report is which lot and why, not a total with a hole in it.
5 · The number, and a decision that is not yours
Report it in this order, every time:
- What was built, lots, shots, which SKUs.
- What was refused, by name, with the reason and the fix.
- What it costs: the total, that it is one take of every shot, that a second take of a shot costs again, and that nothing has been spent yet.
- What happens next, rendering, per lot, when somebody says so.
Then stop. A campaign is where a habit of rendering “one shot just to check” turns into real money on somebody’s own key, and a request to build a campaign is not a decision to pay for one.
Two habits worth keeping
A refusal costs nothing, so prefer being refused at the dry run to discovering a gap at build. Every gap found before stage four is free.
Fix the lots; do not rebuild them. A wrong beat in every lot is twelve small edits, not a second batch, and re-running a batch over one bad shot creates twelve more lots beside the twelve you already have.
Set a lot cap on client work before you render. It is the one control that turns “this must not exceed what we quoted” into something enforced rather than remembered.