Amazon Marketing Cloud Analysis That Moves a Budget

Amazon Marketing Cloud suppresses any output row that represents fewer than 100 distinct users, which is why instances under roughly 40,000 attributed conversions in a 28-day window return path tables that are mostly empty cells. We check that in the first week and before any larger fee is quoted. If your event volume cannot carry path work, we say so, return the readiness fee, and tell you which of the three questions below is still answerable at your scale. Selling a quarter of blank grids to a team that has to present them is not work we take. The questions worth opening an instance for: What display adds on top of sponsored, measured on exposure groups rather than on the DSP report's own arithmetic. Where frequency stops paying, which in the last publishing instance we ran was the sixth impression in fourteen days, after which incremental orders per thousand impressions fell below the cost of serving them. And how long the path actually is, because a median of eleven days between first exposure and order in travel goods, against two in consumables, tells you the reporting window your finance team should be using and usually is not. Audiences only when a line item is waiting: Rule-based and lookalike audiences are simple to build and easy to accumulate. We build one when there is a DSP line item scheduled to spend against it inside the next month, and not otherwise. An instance full of unused segments is a record of analysis that never reached a media plan. You keep the SQL: Every query is left in your instance as commented .sql, written so your analyst can change a date range and rerun it without us. We hand over the query, the output CSV and a decision memo naming the budget line each result should move. We do not build a dashboard on top of it; a monthly refresh of a chart nobody argues with is not measurement. Limits stated before you buy: Lookback runs to roughly 12.5 months, so seasonality older than one cycle is not available. Only events from accounts linked to the instance appear, which matters if your DSP sits with a second agency. Path analysis of any depth needs DSP data present. Amazon's paid datasets and clean-room subscriptions are billed by Amazon at their rates, not resold through us. Not included: DSP media buying and campaign management. Warehousing into Looker, Tableau or your own BigQuery. First-party audience uploads, which we will specify and hand to your data protection officer for approval, because the clean room agreement is between you and Amazon and signing it is not our decision to make. Ongoing analyst hours beyond the queries agreed at kick-off. Not for you if: You want a recurring insight deck. This engagement produces a small number of answers with budget consequences attached, and if none of them changes what you spend next quarter, it was the wrong purchase. Also not for you if the honest answer might be inconvenient: overlap analysis regularly shows that a share of DSP-attributed orders would have arrived through sponsored anyway, and we report that figure as we find it. How it runs: Week one, readiness and volume check against your instance. Week two, query development and a first read with your analyst in the room, because the person who will maintain this should watch it being built. Week three, the analyses and the memo. Where a result argues for a budget change, we write the change as a number, not as a recommendation to consider a change.