Client delivery · Marketing content pipeline

The bottleneck was never ideas.
It was the handovers.One complete path from product photography to published

Built for a company making physical products, manufacturing and selling both domestically and for export. Work that used to pass between three roles now runs end to end with one person — and the two places this kind of project usually fails are the two things this was built to solve.

Status

Live

Deployment

Client's servers · source code theirs

Roles it used to take

Three

Where the publisher receives it

On their phone

Why content doesn't scale

Three real reasons, and none of them is creativity

Every company that has tried to increase content output has met at least two of these. Notice that none of them is solved by producing assets faster.

One piece of content passes through three people

The images wait for a designer's queue, the video waits for an editor's queue, the copy waits for a writer's queue. None of those three is idle — but the time a piece of content spends actually being worked on is far less than the time it spends waiting.

So output isn't limited by who has ideas. It's limited by how cleanly the handovers go.

A roughly-right image isn't good enough

Companies making physical products can't use general image tools, for one reason: what comes out is something that looks like that category of product, not your specific model. Texture, structure or colour drifting even slightly means the piece can't go out.

An image you can't publish doesn't count, however fast it was produced.

Made isn't delivered — someone still has to hand it over

The images are on a drive, the copy is in a document, the video is on another machine, and the person who actually publishes is holding a phone. Just assembling those pieces, transferring them and matching them up is enough to exhaust whatever enthusiasm the piece started with.

A content operation rarely dies from not producing. It dies from producing and nobody publishing.

The first hard part

A good-looking image isn't enough — it has to be your model

This is where general tools stop being usable for anyone selling a physical product. What they generate resembles the category; what you need is the specific thing you sell.

The principle the pipeline is built on is short: the product's real photography is the basis, not a mood reference. Texture, structure and colour have to survive into the output, because a customer will compare the image against what arrives in the box. How exact that has to be varies by category, and the tier is chosen accordingly.

The tiers themselves, and which one a category needs, are set out on the visual pipeline page →

What the system manages

Six areas, from raw photography to a publishing log

Described by what each area handles, not by how it's implemented.

A product asset library

Product photography comes in as a batch and is filed once, and every piece of content afterwards draws from it. Which images a product has used and what content it has appeared in are connected, rather than scattered across individual machines.

Scene imagery

The same product shown in different usage settings, several versions at once to choose from. Hero, cover and standard landscape crops come out together at sufficient resolution — once chosen, it's publishable without going back for a resize.

Three kinds of short video

Several scene images of one product cut into a piece with a sense of space; a before-and-after comparison; and a detail close-up with the product's specifications laid over the frame. Music, voiceover, subtitles, cover and end card are all handled, with configurable length.

Matching copy

Headline, body, hashtags, a spoken script, and the standard replies for the questions that always appear in comments. A company's own past high-performing content can be loaded in so it writes in the company's voice rather than a generic template's.

Pre-publication checks

Prohibited terms and absolute claims are caught before anything goes out. The value isn't convenience — it's that this is the only step that can be done beforehand. Deleting a post afterwards costs something else entirely.

A publishing log

What went out, from which account, when, and where the link is — recorded line by line, schedulable by week, with reminders when due. Not missing a post and not double-posting has never been a matter of memory.

Of the six, the pre-publication check is the one worth pausing on: it's the only step in the whole chain that can be done beforehand. Everything else can be corrected afterwards. That one can't.

Where the output lands

The usual death is producing it and nobody publishing

This is the second hard part, and the one most projects skip. Everything upstream is worth nothing if the finished piece stops on a shared drive while the person who publishes is holding a phone somewhere else.

A piece of content is a package, not a pile of files

Cover image, image set, video, headline, body and hashtags are bundled into one package. Nobody collects three things from three places, and you don't end up with this piece's copy attached to the last piece's images.

Pushed into the interface they already use

The package goes to a named person or group through WeCom or Feishu, and on a phone it opens straight into copyable text and saveable images and video. No back office, no folders, no new password — and this single detail decides whether anyone is still using the system in three months.

Multiple accounts don't collide

The same theme produces different versions per account group, so the brand account, staff accounts and regional accounts aren't saying identical things. Near-duplicate content is where multi-account setups fail, and that isn't a capacity problem — it's a distribution structure problem.

And the data comes back

Which pieces worked, broken down three ways

Platform performance is collected back and attributed by account, by usage setting and by product. Which means the next batch is chosen on evidence rather than on an impression of what did well.

By account

Whether a brand account, a staff account and a regional account should be saying different things is answerable rather than assumed.

By usage setting

Which settings a product performs in — the single most reusable thing this data tells you, because settings carry across products.

By product

Which products deserve more content and which have saturated. Content budget follows the answer instead of following the catalogue order.

The module clients mention most

The one in the back office that exists purely to show what it costs

It produces no content and adds no capability. It's also the part clients bring up most often — because a running cost nobody can decompose is the thing that eventually kills a system, whatever it produces.

Where the money goes, by module, by person, by month

The system tracks third-party usage and its cost against the specific module, person and month it belongs to. Not one total invoice at month end that nobody can decompose.

Those fees are paid by the client directly

We don't collect them and we don't mark them up. The proposal gives indicative volumes so the client can work out roughly what a month costs and where it goes. That's partly for clean accounting and partly so any component can be swapped out without asking us.

When something cheaper or better appears, it can be swapped

The capability layer is abstracted, so replacing what sits underneath doesn't change any business function, and the switch is inside the maintenance scope. This field turns over every six months; welding yourself to one vendor is a cost the client eventually pays.

The same week

Once it was one pipeline, and while it was three queues

Not how it runs internally — just what people spend the week doing.

One pipeline

  • Pick the products and settings, select a batch, and by the next morning they're queued and ready
  • One person goes from product photo to finished piece without waiting on anyone's schedule
  • Images, video, copy and tags arrive on a phone in one package, ready to post
  • Prohibited terms are caught before publication, not after
  • What went out, from which account and when is recorded by the system
  • Performance by setting and by product is on a dashboard you can drill into

Three queues

  • Book the designer, then book the editor once the images exist, then book the writer
  • Three roles confirming back and forth, with most of a piece's life spent in a queue
  • Assets on a drive, copy in a document, video on another machine — assembled by hand
  • Prohibited terms caught only if somebody happens to look twice before posting
  • What was published is reconstructed from chat history and individual spreadsheets
  • Which pieces worked is an impression, with no way to say whether it was the setting or the product

What the client received

Five things, all of them theirs

Delivered and running on their own servers. We finished and stepped back out.

A fully deployed system

Back office, every workflow and the rendering services, installed and working — not a proposal document the client has to build from.

Templates tuned to their own products

A set of usage-setting templates, three video templates and a copy template, each specifically tuned against the client's actual products rather than generic presets.

Their own content knowledge base

Product categories, specification wording, standard terminology and the questions customers keep asking, collected in one place and extensible afterwards.

Manual and training

An illustrated operations manual plus a recorded live training session — so when people move on, the recording is enough for the next person.

Deployment docs and source code

Source code into the client's own repository. Templates and parameters are configurable in the back office, so business staff adjust and extend them without going through a developer each time.

When this doesn't apply

Three cases where we'd tell you not to

Written down because a pipeline that suits everyone suits no one.

Content doesn't need to be continuous

A few pieces a quarter won't do it: this pipeline's value comes from repetition, and without the repetition it's overhead. For occasional content, commissioning it once is far cheaper than keeping a line running.

Your content lives on a person on camera

Presenter formats, visits and interviews are valuable precisely because of the person in them. This produces product content, not personality content, and the two shouldn't share one pipeline.

The product is different every time

One-off bespoke work with no reusable product asset library means every piece starts from zero. A pipeline can't hold that shape, and we'll say so.

Where this sits

Related, and often confused

Two of our systems get called “pipelines” and they do different jobs.

This one and the visual pipeline are not the same thing

One handles where the imagery comes from — production capacity, ours, used for launches and advertising. This one handles a piece of content travelling from product photography to published and back into data.

The knowledge behind the copy

Specification wording, terminology and the questions customers keep asking come from one place, so the content, the support answers and the internal answers agree.

You can start with one piece

Batch imagery, or the copy production, works as a single module with the interfaces left open for the rest of the line later.

Questions

What people ask about this one

The first is the most common, so it's first.
01How is this different from an AI image or video tool?+

A tool produces an asset. This produces a piece of content and gets it to the person who publishes it. The two places these projects fail are both outside what a tool covers: the image has to be your specific product rather than something in the category, and the finished piece has to arrive where the publisher already is. A tool solves neither, which is why buying one rarely changes output.

02Will the images actually look like our product?+

That's the requirement the whole pipeline is built around. The product's real photography is the basis the output is bound to, not a mood reference — what comes out has to be that model, not something that resembles it. Where a category needs more accuracy than that, the tier can be raised. Selection stays with a person, so nothing publishes that hasn't been looked at.

03Who checks it before it goes out?+

Two checkpoints, both deliberate. Selection is human — someone chooses which version ships. And prohibited terms and absolute claims are screened before publication, which matters because that's the only point at which the problem can still be prevented; deleting a published post is a different order of cost.

04What does it cost to run each month?+

Third-party usage is the running cost, and it's tracked in the back office by module, by person and by month. Those fees are paid by you directly to the providers — we don't collect them and don't mark them up — and the proposal gives indicative volumes so you can estimate before committing. If something cheaper or better appears, the component can be swapped without changing any business function.

05Can our own people change the templates?+

Yes, and that's a delivery goal. Templates and parameters are configurable in the back office so business staff adjust and add without a developer, and the training session is recorded so a new starter can pick it up later.

06Does adding a new content type or platform later cost extra?+

Yes, and it's written into the contract at signing rather than raised mid-project. Day-to-day maintenance — templates, wording, parameters, swapping an underlying component, routine checks and incident response — your own people can carry, or you can buy an annual maintenance subscription. A new content type, a new publishing platform or a new system integration is new development, scoped and priced as a new project.

Next

Count the handovers, not the ideas

The useful question is how many people one piece of your content passes through, and how much of its life is spent waiting rather than being made. The diagnostic fee is credited in full against the project.