Kroger Precision Marketing embed for incrementality

We embed two people with your retail media team for a quarter and design the reads before the media runs. A one-point lift claimed against Kroger's household base is a different statement from a one-point lift inside a forty-thousand-household segment, and most decks print them in the same typeface. Before anything launches we write down the effect size worth acting on, the cell sizes needed to detect it, and the decision each cell exists to inform. If the arithmetic says the test cannot answer the question at your distribution, we say so then, not in the readout. How the embed works: Two of ours in your standups, working in your platform seats under your login. We do not take the account, we do not sit between you and your Kroger Precision Marketing representative, and nothing we design depends on us staying. What your team keeps at the end is a way of working: a test calendar, written designs, and a reporting format that separates what was measured from what was inferred. Designing the read: Minimum detectable effect declared in writing before launch. Exposed and control households held apart for the full flight, usually eight to twelve weeks. No audience edits inside a live cell — a mid-flight change is the most common way a Kroger test quietly stops being a test, and it almost always happens because somebody wanted the weekly number to move. We hold that line, and it is occasionally unpopular in the room. Audience work: Household data is the reason to be on this network at all, so the segments have to be ones your brand team can act on: lapsed buyers past twenty-six weeks, category buyers who have never bought you, households that cross banners between Ralphs and Smith's or between Fred Meyer and QFC, and new-to-brand read by region rather than nationally. Each segment is built to be testable first and large second. Where the money goes: On-site sponsored product against items with real ACV in the banner set, off-site programmatic against the segments above, and in-store where that store's own footfall justifies it. We will tell you when the right answer is fewer segments carrying more weight each, which is usually the answer in the first quarter. What we refuse: We do not write a readout unless the specification was agreed before launch, and null results are published in the same format as positive ones. We do not work on two brands competing in the same Kroger category at the same time. And we will not run a programme with no unexposed households in it: if the holdout is negotiable, the engagement is not. Who this is not for: Brands with thin distribution across the banner set. Below roughly a fifth of ACV a household test cannot be powered, and what you would be buying is a coin toss with a decimal point on it. Teams that need somebody to own the day-to-day buttons; that is a different arrangement and we will point you at it rather than absorb the work quietly. Anyone whose leadership has already settled on the answer. And anyone who cannot leave a control group alone for two months, which sounds straightforward in a kick-off and turns out not to be.