Laura Scotten

01 · Audience

Meta kept handing us people twice our target age

For as long as we’ve been able to track the data, our Meta-determined audience has skewed old. 45 to 54 was the sweet spot and 55 plus was right behind it, which has been great for building our business so far, but we have been trying to turn our younger YouTube audience into paying customers and have had a tough time finding the best way to do that. Meta kept delivering our ads to who it thought would buy, and that kept being Gen X and Boomers.

My read was that the polished, branded, obviously-an-ad variants were doing the sorting for us. A 58-year-old will happily sit through a nicely made commercial. A 28-year-old scrolls past one.

I pushed for comedy, skits, lo-fi, and UGC and EGC style cuts, things that look like they belong in the feed instead of interrupting it. The statics stayed mostly polished and on-brand, because that’s where the brand actually needs to hold. The video team changed tack and started building for the feed.

Trials per thousand impressions is more than doubled, and that one doesn’t care how big the budget was. Click-through went from 2.10% to 2.68% over the same comparison.

02 · Test design

One pain point, six angles

Bar chords are where beginner guitarists quit. It’s not a metaphor, it’s a specific thing your hand won’t do yet, and a lot of people quietly decide at that point that they’re not a guitar person after all.

This static went out in July and beat everything else in its set. On its own that means nothing, but it had me curious, so I pushed to point the next testing round at that one barrier and nothing else.

Six angles went out as 11 statics and 9 video cuts.

“Learning bar chords SUCKS” made the best static of the lot, and it nearly didn’t run at all. Our CMO thought the language was too harsh. I pushed to keep it because it’s how the instructor actually talks, and because a campaign from years back had already told us this register works and we’d just forgotten.

The rest of the set

Five more of the 11 statics. All mine, concept through build. The bottom two barely ran, so read those numbers as a look rather than a result.

Video

All 9 cuts, built by our video team to angles I set. Thumbstop is 3-second plays over impressions. Trials over spend is an ad’s share of the set’s trial starts divided by its share of the money, so anything above 1.00 paid for itself.

Naming the exact technical failure, the buzz, the F chord, the grip of death, made the strongest hooks by a distance.

Attention and trials came apart almost entirely, and you can read it straight down the grid. The cut that held the most people returned the least of anything we funded. The one that returned the most is sitting seventh on attention. I don’t have a tidy explanation for that yet and I’d rather say so than invent one.

And the angle my team normally bets on, the demographically precise one aimed at stuck intermediates in their forties, did nothing at all. It described a person instead of a moment they’d recognise.

Video and statics split the money nearly evenly and came back nearly in proportion, so I can’t tell you one format beat the other. What mattered was which angle you picked.

This launched in early September and is still under our significance floor, so treat all of it as an early read. Ratios instead of counts to protect Musora’s numbers.

03 · Reading the signal

When the variable turns out to be a person

A static was beating everything around it in both spend and trial-starts. Same template, same offer, same messaging angle as the rest of the set (every instrument, and as many demographics as I could test for each). The only thing actually different was the guy in the photo.

Not knowing if it was a fluke or something about the model, we built a video asset from the b-roll of the same shoot. It’s now the biggest creative by spend in that portfolio, and having a video instead of a static gives far more data to learn from.

He’s booked for an upcoming shoot so we can pull it apart properly and find out whether it’s his face, his energy, or just the room he happened to be sitting in. We will use our best performing hooks, angles and formats, and try on different instruments (the model is a multi-instrumentalist). If it does turn out to be him, that’s a much better thing to own than a concept, because we’ll be burning through concepts forever and he’ll still have that face.

The read and the argument were mine. Our video team cut the footage.

04 · Customer voice

774,000 student comments, finally readable

Every lesson on our platform has a comment thread under it and some of them go back to 2012. There’s an enormous amount in there about our customers and none of us could read all of it. We would occasionally parse through a few videos to find a nice quote for a sales page, but it was time-intensive and not particularly effective.

So I pulled the lot into a database and built a tool that queries it. 774,546 comments, five brands, fourteen years.

It isn’t only about the size, it’s also what I threw out. Our student experience team comment constantly and they’re warm and encouraging because that is literally their job. If you leave them in and use it as a source for “student voice” it ends up being your own brand talking back to you. The working set strips staff, moderators and deleted threads.

I can now find the real language our students use, real pain points they are struggling with, and real wins they are experiencing in our app. It has been a game changer for headline writing and scripting in our campaigns.

05 · Creative systems

The boring layer everything else sits on

We could always tell you which ad won. But as a data set, we couldn’t tell you why, because ad names were whatever the person building the thing felt like typing that day.

So I built the naming convention, and then the taxonomy underneath it, and a living document that outlines details from every naming segment. Every ad gets coded for hook tactic, messaging angle, visual format, audience, pain point, offer, and model. So we know whether “Olivia” is a coach, a student, or a celebrity, we know what hooks work best for what audience, and we can parse out whether the ad hit because the offer was great, or failed because the pain point isn’t relevant to the audience.

06 · Creative engine

Ten times the output, no AI-generated photos

Testing at volume turns into a production problem long before it’s a creative one. Dozens of our top of funnel campaigns sat untouched for years because building variants by hand cost more than the variants were worth.

I have been leading the development of an asset manager that holds the whole library, 200,000-odd photographs. Filenames stay exactly as they are but every image gets AI-tagged on upload for shot angle, emotion, demographic, instrument, location, face recognition and quality. I built the catalogue into an automated Figma template so one set of branded static ads of our best performing designs got turned into our entire course catalogue, complete with updated AI generated headlines and hooks.

If you can query what’s in the library you can query what isn’t, and what came back was an uncomfortably specific list of the people we can’t currently cast. That list is now the brief for a full day of shooting, so we’re filling holes the data found instead of holes we’d have guessed at, or just casting whoever was available.

The one thing I won’t move on is AI-generated imagery as a subject. As a human-first organization, we pride ourselves on casting real students and representing real human creativity and education. Using AI to get real photographs into templates quickly is a completely different thing from using it to fake the photograph.

Creator direction

Writing to the person, not the brief

Our roster of partners and coaches runs from classical concert pianists to a Spanish metal drummer with a NSFW personality. You obviously can’t hand those two the same script.

I research their persona and write to it. Vocabulary, pacing, format, hook, all of it shifts depending on what their audience already expects from them. Threading the needle of our brand voice and that of our guest stars develops trust for our brands.