Media
Personalization isn’t a feature, it’s a platform—and most streaming services get it backwards
Open almost any streaming app and you’ll find a row called “Because you watched.” That row is what most companies mean when they say personalization. It’s also why most personalization quietly underdelivers.
A recommendation row is a feature. It sits on top of a content catalog, and you can ship it in a quarter. Real personalization is a platform with several layers that have to work together: content likeness (how titles relate to each other), who the user actually is, what user behavior across the whole catalog tells you, and what you’re allowed to do with that information. Content likeness is the visible layer—the row. Most teams build it first and never build the others.
I spent two years building the layers underneath at People Inc., across 40+ brands and 22M+ users. Here’s why the order most companies choose is backwards, and what it costs them.
The recommendation row is the last layer, not the first
“Because you watched The Bear” is the output of personalization, not the system. By the time a row renders, a series of harder questions have already been answered—or skipped. How is The Bear like other content we serve? Who is this viewer, really? What does their behavior across everything they’ve touched tell us? What about the behavior of other users? And what does our agreement with them let us use?
Skip those, and the row is just a content popularity chart addressed to the user. A lot of streaming personalization is exactly that: catalog-wide trending, lightly filtered by the one show you finished.
Identity is the layer everyone skips
You can’t personalize for a person you can’t recognize. At People Inc., personalization didn’t start with a model—it started with identity. The mandate I was handed looked like a newsletter migration. What it actually needed was a multi-year identity, registration, and onboarding layer so we could recognize the same human across 40+ brands instead of treating every brand visit as a stranger.
That layer is unglamorous, and it is the whole game. Once you can recognize a person, behavioral context compounds across every interaction. Until you can, every interaction starts from zero.
Streaming has a sharper version of this problem than publishing did: shared accounts, four profiles on one login, a kid’s show and a prestige drama an hour apart. Get identity wrong and your “personalization” is averaging across people who aren’t actually the same person.
Behavioral context lives across the catalog, not inside one title
The most effective recommendation work I did at People Inc. wasn’t a better algorithm. It was a decision about where to draw the training boundary. We relaunched MyRecipes on a model trained across reader behavior on every food brand in the network, not just the brand a reader happened to arrive on MyRecipes from—and it doubled traffic against the recipe-of-the-day baseline.
The model was ordinary. The cross-property training set was the unlock.
Streaming has the same boundary problem hiding in plain sight. Behavior on one title, one franchise, or one genre is a thin signal. Behavior across the full catalog—what someone abandons, what they rewatch, whether they browse or binge—is the rich one. Personalize within a title and you get more of the same. Personalize across the catalog and you start getting the person.
Governance is part of the product surface, not the legal review at the end
Here’s the layer growth teams treat as someone else’s job: what you’re allowed to do with the data. I sharpened my product sense through an MS in Law focused on data privacy, so I read consent and data rights as product inputs, not compliance overhead. What you can collect, what you can join, how long you can keep it, and what a user can revoke—those constraints shape the personalization surface as directly as any model does.
Teams that bolt governance on at the end build systems they later have to tear out. Teams that treat it as a design constraint from the start build personalization that survives a regulator, a breach disclosure, and a user who actually reads the settings page. In streaming, where the data is intimate—what you watch alone, late, repeatedly—that isn’t a nicety. It’s the license to operate.
You can tell who made the platform bet
Personalization as a feature is a row you can ship this quarter. Personalization as a platform is content likeness, identity, behavioral context, and governance built as one surface—the row you see riding on three layers you don’t.
And you can tell, as a user, which bet a company made. When a service recognizes you across devices, gets sharper the more you use it, and never makes you feel surveilled, someone built the platform. When it recommends the show you just finished and forgets who you are on your phone, someone shipped the row.
Most streaming services shipped the row. The opportunity is in the layers underneath.