Philo Turns to Reelgood to Clean Up Streaming's Metadata Problem

Every recommendation engine in streaming runs on metadata. The problem is that this data, across hundreds of streaming services, is a mess.

Cast, genre, runtime, release year, availability windows — the structured signals that tell a machine learning model what a piece of content is and where a viewer can watch it. There is no industry-standard ID for a title. Netflix, Disney+ and Prime Video can all carry the same movie and assign it three completely different internal identifiers. Release years don't match. Genre taxonomies don't map. Availability windows go stale within days.

Reelgood spent eight years and tens of millions of dollars building machine learning infrastructure to solve exactly this problem, matching titles across services without a universal ID and using cast, crew, runtime and synopsis as the connective tissue.

SOS Insight: The metadata layer isn't a feature. It's the floor. A recommendation engine is only as good as the data feeding it, and most streaming services are building on a foundation they don't fully own or control.

Philo is a $25/month live TV service with 70-plus entertainment channels and 75,000 on-demand titles. What the Reelgood deal actually buys Philo is three things: daily-refreshed metadata integrated directly into its machine learning stack, cross-platform signals showing what's resonating across 300-plus services in real time, and historical availability records dating back to 2019.

That third piece gets underestimated. Licensing decisions in streaming are often hundreds of millions of dollars riding on questions like: Where has this title been available, for how long, and on what tier? AVOD services don't want to license a title that's been free for two years on Tubi. SVOD services want to know if a piece of content has appeared on a major competitor before they pay a premium for exclusivity.

The intelligence gap is bigger than most executives admit. Reelgood CEO David Sanderson has been on calls with major studios that didn't know their own catalog was licensed to a competitor's service, content that should have been pulled when a rights window closed, still streaming somewhere it wasn't supposed to be. Not fraud. Logistics failure at scale.

"Any recommendation engine is only as good as the metadata feeding it," Sanderson said. "They came to us for complete, deduplicated coverage refreshed daily. And its content team came for the historical availability for every show and movie, so it can make licensing decisions based on evidence, not guesswork."

What It Means

Philo gains a metadata foundation that can support machine learning work at scale, plus competitive catalog intelligence it didn't have before. Knowing what's resonating across 300-plus services lets a live TV service make more aggressive content acquisition bets instead of chasing its own internal signals in a closed loop.

Reelgood moves up the stack from data vendor to strategic infrastructure layer. Integrating directly into Philo's recommendation architecture is a stickier business than licensing a data product.

Ecosystem builders compound. Content buyers transact. The metadata layer is what separates one from the other.