Amazon DSP Off-Amazon Campaigns with Holdout Testing

Amazon DSP reports view-through conversions on a fourteen-day window, and a view is not a decision. Every campaign I plan starts with a holdout: 15 to 20 percent of the addressable audience, or a set of matched regions, withheld for four weeks minimum, with the result read on total orders and total installs rather than on attributed ones. The number that comes back is almost always smaller than the platform's, and it is the number I plan the following quarter on. If a measurement that makes your own channel look smaller is politically impossible where you work, this is the wrong engagement and I would rather say so in the first call. Supply, and where the money quietly goes: Amazon owned inventory, meaning detail pages, Fire TV, Twitch and the Prime Video ad tier, behaves nothing like the open exchanges, and the exchanges are where budget disappears without anyone lying about it. I run an allowlist rather than a blocklist, read the placement report weekly for the first six weeks, and cap frequency per line item instead of per campaign. Saturation is measured rather than assumed: on consideration campaigns the point where the next impression stops moving anything usually lands between six and ten in a fortnight, and that is where the cap goes. App installs and off-Amazon events: For mobile app clients, DSP runs against your mobile measurement partner, whichever you already hold, with the install event and one downstream event both wired before any spend: trial start, first funded deposit, first lesson completed. Optimising to installs alone buys installs, and installs are the cheapest thing in the funnel to manufacture. For web conversions the Amazon Ads Tag goes in with a consent check, and I confirm the events fire myself in staging rather than accept a screenshot from an implementation team I have not met. Regulated categories: Financial services creative goes through category review, and licensing and disclosure differ by market, so creative is built per market rather than translated. I plan a two-week review cycle into the timeline and I will not commit to a launch date that assumes approval first time. Seat, invoices and fees: I work on your seat, or on an agency seat where you hold the contract. I will not book your media through a seat of mine and mark it up, and I am not paid a percentage of spend. You see the platform invoice. This is why I am more expensive than an arbitrage shop and cheaper than one over a year. Not included: No video or display production; I write the brief and the specs. No measurement vendor implementation beyond the event wiring described above. No sponsorship or upfront negotiation for Twitch or Prime Video inventory. No first-party audience onboarding into Amazon Marketing Cloud without written sign-off from your legal team, which I will ask for and then wait for however long it takes. Who should not buy this: Below roughly EUR 25,000 a month in DSP media, a holdout will not reach a readable difference and you are buying a measurement you cannot act on. Put the money into search and come back when the ceiling there is real. Also not for teams who want DSP because a competitor announced they were running it; that is a reason to ask a question, not a reason to fund a channel.