Publishing 700 Unlisted Videos: Our First Customer

Yoga Marga is a yoga school operating out of New York. When we met them they had somewhere around seven hundred videos sitting on their YouTube channel, all unlisted, all effectively invisible.

The recordings existed. They were Zoom sessions, captured over years of teaching, uploaded to YouTube and then stopped there — because between an uploaded file and a published video sits a title, a description and a thumbnail, and doing that seven hundred times is not a task anyone starts on a Tuesday afternoon.

They were our first customer, and the product was built while working through their backlog.

Why the backlog existed

This is the part worth understanding, because it is not laziness and it is not unusual.

A school teaching over Zoom produces video at the rate it teaches, and the recording is a byproduct rather than a production. Nobody hits record on a class thinking about thumbnails. Zoom makes the capture free and effortless, which is exactly why the backlog forms: the cost of creating another file is nothing and the cost of publishing one is half an hour. So the files accumulate, someone uploads them to YouTube so they are safe somewhere, unlisted because they are not finished, and finishing them becomes a project with no obvious owner and no deadline.

By the time anyone counts, it is seven hundred. At even fifteen minutes each that is roughly four and a half working weeks of uninterrupted metadata work, which is why it does not happen. It is the same arithmetic that stops smaller back catalogues, at a scale where it stops being a chore and becomes structurally impossible.

What the work actually was

Each video needed the same four things, derived from the video rather than from a template.

A transcript, because everything else depends on knowing what is actually in the recording. For a class, that is not guessable from a filename.

A title that distinguishes it from six hundred and ninety-nine others covering related material. This is harder than titling a normal channel's uploads: on a channel where every video is broadly the same subject, a generic title makes the whole catalogue indistinguishable.

A description in a consistent format, so the channel reads as one library rather than seven hundred separate decisions.

A thumbnail. A Zoom recording is close to the worst possible source for this. Every frame is the same speaker view or the same gallery grid, in the same room, at the same webcam exposure. Scoring candidate frames against each other returns a winner that is not meaningfully better than the loser, because there is no moment in the footage to find. This is the same problem podcast channels have, and it is why best-frame extraction and image generation exist as separate templates rather than one with a fallback.

What it taught the product

Nearly everything in Growati that constrains the output came from this engagement.

The brand profile exists because seven hundred videos need one consistent voice, and the only way to get that is to store the rules once instead of re-deciding per video. The review step exists because publishing seven hundred pieces of metadata unattended, to a real channel with a real audience, is not something I would do or ask a customer to do. The templates exist because it became obvious that "generate everything" and "generate metadata only" are different jobs.

Building a product against a single customer is normally bad practice, and it is fair to say that is what happened here. What made it survivable was that the constraint we were solving for — volume, consistency, and no usable frames — turned out to generalise better than a niche yoga catalogue had any right to.

What this case study does not contain

Numbers. I want to be direct about that rather than let the omission pass.

We did not capture before-and-after performance for this engagement, because we were building the thing while running it and measurement was not in place. I cannot tell you what publishing those videos did to the channel's views, and I am not going to reconstruct a figure after the fact.

There is also no quote here. We have not asked Yoga Marga for a written testimonial and I am not going to paraphrase a client into sounding like one.

So read this as a description of work rather than as evidence of results. The beta findings post is the closest thing we have to the latter, and it is careful about its own limits too.

If you have a backlog

The specific thing worth copying from this is the order of operations: transcript first, then metadata derived from it, then thumbnail, then publish in reviewed batches rather than all at once.

And if your catalogue is more like sixty videos than seven hundred, start with the ones that already have impressions and poor click-through. Yoga Marga's videos had no impressions at all, because unlisted videos are shown to nobody. That made the decision simple in a way most back catalogues are not.