Amazon Marketing Cloud Analysis That Ends an Argument

Three properties of your AMC instance decide which questions it can answer. Any output describing fewer than 100 users is suppressed. The window is roughly 13 months, rolling, so a query written today cannot see the launch you want to compare against if it is older than that. And the join key is Amazon's identity graph, not your customer file. Most disappointing AMC programmes are questions the instance was never able to answer, discovered in month two. We check that in week one and tell you if the answer is no. What we establish before writing SQL: Whether there is DSP spend, because several of the interesting questions are about the interaction between DSP and Sponsored Products and collapse without it. Monthly conversion counts per segment you intend to cut by, tested against the 100-user floor. And whether your question is attribution, which can be answered looking backwards, or incrementality, which needs a holdout designed before the media runs. Those are different projects and we will not sell the first as the second. Questions that come with a decision attached: What conversion rate does at each additional exposure when a shopper has seen both Sponsored Products and DSP, which tells you whether the DSP line is buying orders search would have taken anyway. Where the DSP frequency curve flattens, which sets your cap. New-to-brand rate by placement, which decides whether Sponsored Brands video earns its rate against a static unit. Path length and time to the first 定期おトク便 order for a consumable, which tells you whether the subscription discount is being surfaced too late. Time to second order, which is the number that sets your allowable acquisition cost and is usually being guessed. Financial services and off-Amazon conversions: For a card or insurance client the conversion happens on your own application flow, and AMC cannot see it until those events are uploaded through Amazon Ads Events Manager. That is two to three weeks of engineering on your side before any of it works. We will scope it and specify the payload, but we say plainly at the start that the analysis waits on your team, not ours. What you receive: Six to eight queries in a git repository you own, each one documented with the assumption that breaks it and the sample size below which the result should be ignored. A decision memo naming what to stop, in order, with the money attached. One working session with your analysts so they can extend the queries rather than commission the next batch from us. We write the SQL to be read by someone else. What we refuse: Dashboards. AMC output refreshed weekly and looked at by nobody is a cost dressed as capability, and the questions worth asking have answers that stay true for a quarter. Audience builds we cannot measure the lift of. Holdouts sized so small that the read is noise, even when the client wants the chart. And continuing past the feasibility step when the honest finding is that your instance cannot answer the question: we bill the feasibility work only and stop. Who this is not for: Brands running Sponsored Products alone at modest volume, where the 100-user floor suppresses most of what you would want to cut by and a well-built search-term analysis gets you further. Teams that need AMC on the plan because a holding company asked for it. Anyone who wants the queries but not the decision, since the decision is the part of this that costs money and the SQL is the part that does not.