Every media team faces a version of the same question: we're spending across a dozen channels, but how much of what we're seeing is actually because of that spending? And if we shifted the mix, what would happen?
For a streaming service, the challenge was compounded by the nature of the product. Shows aren't toothpaste — demand spikes at premiere, decays fast, and varies dramatically based on whether a show has an established audience or is trying to build one. A one-size-fits-all media model won't work.
The Approach
We built a marketing mix model covering 13 tentpole shows over 15 months, drawing on stream data alongside spend across more than a dozen platforms — Meta, The Trade Desk, connected TV, TikTok, Reddit, and a range of publisher-direct buys. The methodological challenge for streaming is that spend and streams move together by design — you spend more when you're launching, and streams peak at launch. We isolated the signal by studying variation across geographies, normalizing all data by population so large markets didn't mechanically dominate, and modeling each show separately given how differently they behave.
The question isn't whether streams and spend move together. They always will. The question is whether, in markets where you spent more, you got more — after accounting for everything else.
What the Model Found
streams from paid media
tentpole shows
incremental stream
Across the 13 shows, paid media contributed an estimated 3.8 million incremental streams — a 10% lift over baseline demand. That average concealed a wide range: a documentary aimed at newer audiences saw a 25% lift from paid support, while a long-running drama with a deep loyal following saw only 4%. The model doesn't penalize that show for a smaller lift — it's a sign of a strong show, not a weak media plan.
Channel Efficiency
| Channel | Cost / Incremental Stream | ||
|---|---|---|---|
| Display | $0.72 | Most efficient | |
| Paid Social | $0.94 | ||
| Direct Digital | $1.08 | ||
| CTV | $1.32 | ||
| Pre-Roll | $1.44 | ||
| OEM | $1.64 | ||
| Podcast | $2.76 | Least efficient |
Display was the most efficient channel by a meaningful margin — but display is a small portion of total spend, and efficiency at current scale doesn't guarantee efficiency at higher spend levels. CTV drove more total volume than any other channel but at a higher cost per incremental stream; its overall contribution was driven heavily by a handful of shows where existing audience strength and CTV reach combined effectively. Podcast showed the weakest efficiency across the board.
Reallocation Simulation
For one nonfiction title, we simulated shifting 10% and 20% of the budget toward the most efficient channels. The model estimated a 5.7% improvement in streams at 10% reallocation and 11.5% at 20% — meaningful, hundreds of thousands of additional streams without additional spend. Directional rather than contractual, but a useful planning input.
Geography as a Strategic Lever
Normalizing streams by population revealed striking variation. Some markets significantly over-indexed on per-capita streaming; others were well below average. That creates a strategic choice: concentrate spend in high-performing markets to maximize near-term ROAS, or shift toward under-served markets to build long-term audience breadth. Neither answer is obviously correct. What the model does is make the choice visible and quantifiable.
Want to understand what's driving performance in your media mix? Say hello.