Idea · 2026
4 min read
PlayableAn ad exchange inside the movie
Product placement is sold once, years early, to one buyer, for every viewer. Everything programmatic advertising learned about clearing inventory says it should not be.
In 2019 I built a Movie Online-to-Offline attribution model, proving digital ad spend drove real footfall into cinemas by measuring who actually turned up. I spent a long time thinking about the boundary where an ad ends and a film begins.
This idea is that boundary from the other side.
Product placement today is the least sophisticated inventory in advertising. A can of a particular cola is negotiated eighteen months before release, sold to exactly one buyer, and then shown to every single viewer in every market forever. No targeting, no clearing price, no way to buy it if you were not in the room in 2023.
Meanwhile the same industry clears billions of impressions a second by auction.
Run the exchange
The frame below is drawn live rather than played from a video file, which is the point, because compositing into a rendered scene is exactly what the architecture requires. Pick a viewer segment and clear an auction for the slot.
Bids · CPM
floor $11.00
- Volt$30.45
- Northwind Cola$24.29
- Kettle Roast$12.48
- Still & Co$11.59
Second-price: the winner pays a cent above the next valid bid, not their own.
The architecture
The auction is the easy half; programmatic solved that. The hard half is upstream, turning a finished film into inventory at all.
Ingest
The finished film is processed once. Nothing about this is real time. The expensive work happens long before any impression, which is what makes the serving path cheap enough to auction.
The three objections, taken seriously
"This ruins films." It can, and a badly implemented version certainly would. Which is why the constraints belong to the title, enforced as data: category allow-lists, hero versus background slots, a hard cap on how many slots a scene may carry. A director who wants zero slots sets zero. The architecture has to make restraint the easy default, or it deserves the objection.
"The rendering cost is impossible." Only if you render per viewer, which nothing here does. Variants are per title-and-product, rendered once, cached, and served like any other asset. The number of variants is bounded by how many products clear the floor for that title, which is small.
"Attribution is impossible." This is the objection I have the most sympathy for and the most experience with. It is the same problem I worked in 2019, and the honest answer is that it is partially solvable and never clean.
- Sold today
- One buyer, eighteen months early, every viewer, forever.
- Sold here
- Per segment, per market, clearing at auction, changeable after release.
- Unlocks
- Regional and mid-market advertisers who could never be in the room for a placement negotiation.
- Risks
- Creative integrity, viewer trust, and attribution honesty. In that order.
Why this is not a pitch
I do not think this gets built the way I have described it, and I am not trying to raise money for it. What I find useful about it is that it is a clean example of a pattern I keep running into: an asset that is already digital, still being sold as if it were physical.
Cinema footfall was one. Out-of-home was another. I spent two years turning DOOH from a media buy you negotiated into inventory you could plan against, and it grew from $2M to $5M largely because being planable made it buyable. In-film placement is the same shape, one step further along.
Finding those is most of the job.