Roku, Samsung and LG Ads Bought Direct, Not Exchange

Buy the three of them separately, or do not buy them: Roku, Samsung and LG are three publishers with three sales models, and the placement each one genuinely owns — the home screen the set opens on — is sold on insertion order, not in any auction. Nothing in a DSP reaches it. In my Spanish and UK flights, OEM in-stream clears between 18 and 34 euros CPM bought direct; the same avail routed through an open exchange with the publisher name masked has come back 30 to 60 percent higher, and about a fifth of the time it was not the OEM at all. Where I buy: Roku through Ads Manager for self-serve tests and OneView where a household graph is attached. Samsung Ads direct, with the DSP used only for the audiences worth extending. LG Ad Solutions direct. Every line runs with publisher-level reporting at minimum, and where the deal size supports it I take log-level delivery and reconcile it myself. The reconciliation is where the money is found. The one thing these platforms have that nobody else does: Automatic content recognition from their own installed sets. It is the only honest way to see what a household already saw on linear and to price the overlap. For a Spanish client running Atresmedia and Mediaset schedules I build the incremental reach question first — which households the linear plan is not reaching at all, and what a thousand of them costs on Roku against a fourth exposure on television — and the buy follows from that answer. If linear is already reaching them at frequency nine, I say so and the budget goes somewhere else. Verification, because this is where the fraud went: app-ads.txt and sellers.json checked against the actual supply path before the flight, not in the wrap. Server-side ad insertion means the stream is stitched by the publisher and device-level signals can be forged, so I treat the OEM's own logs as the counting source and any third-party number that diverges by more than a few percent as an open question rather than a rounding difference. DoubleVerify or IAS on every line, pre-bid where the platform offers it. Measurement, given there is no click: Geo holdout at província level in Spain, DMA in the United States, drawn before the flight and held for its duration whatever the weekly numbers do. QR and on-screen response codes where the creative earns them, never as the primary read: scan rates on a living-room screen are a rounding error, and anyone quoting one as the KPI is selling you something. What I refuse: CTV bought through an open exchange when the same inventory is available on an IO. Flights where the platform will not name the app. Household frequency promises across all three OEMs, since there is no shared identifier and what you are being sold is a model, not a count. And Spanish igaming: gambling advertising here is confined to a narrow overnight window, these platforms do not daypart to it reliably, and I will tell a licensed operator to spend in Mexico or the United Kingdom rather than bill them for a Spanish plan. Not included: Creative production and versioning, linear or addressable-linear buying, and the streamer ad tiers. Netflix, Disney+ and Hulu are a different negotiation and a different engagement. Who should not book this: Anyone whose single-market flight is under roughly sixty thousand euros. Below that, Roku's self-serve tool is the right answer, and you can run it without me.