What size lift could we detect from your ad network?
Being able to publish a defensible lift number from a retail media campaign depends on three things:
- How many stores run it
- How many campaigns you read together
- How much of the retailer's data you have visibility into
Use the calculator below to see how each one moves the minimum detectable lift.
This calculator computes the smallest lift a matched-store read could reliably find, given the size of your network, the length of the flight, and how noisy the category is. It is exact, not a rule of thumb: the formula is the sampling variance of the estimator we actually use, and it agrees with a simulation of the same design to within a percent. Nobody's data is involved, and nothing you enter is sent anywhere.
Set these to match the campaign you have in mind. Everything here is something you already know about your own network, so there is nothing to look up.
A single flight at 40 stores already finds 3% on its own. Reading your year together tightens it to 1.4%.
Assumes 26 weeks of clean pre-period history, a typical amount of week to week persistence in store sales, 80% power and a 5% two-sided test. Reading campaigns together assumes campaign effects vary by about 1.2 points from one to the next.
Caveats
- The default here assumes we can get item and category detail. Without it the floor moves 15%, not 47%.
- Brand share of category used to remove noise from overall store traffic. Limits our ability to see overall category growth.
- Dose response approximates sales lift per unit of ad exposure rather than looking at aggregate sales while the campaign was live.
- Promo activity could overlap with the campaign and confound the analysis, which is why we ask for the discount calendar.
- Reading campaigns together lowers the floor and limits how far we can isolate any one campaign. A campaign needs a strong result or a large store count to move away from the library read.
- Multipliers come from simulation on synthetic store panels, not a completed engagement. The first real read replaces them with your own numbers.
Three ways to lower the floor
- Run it in more stores.Precision improves with roughly the square root of the exposed store count, so four times the stores halves the detectable lift. Adding weeks does much less.
- Read more campaigns together.Each campaign is measured on its own stores against its own comparison stores, then the answers are combined and weighted by the evidence each contributes. A campaign at 90 stores counts for more than one at 12. A campaign too small to prove anything alone still adds evidence.
Illustrative. Individually, 6 of the 8 campaigns include "did nothing". Read together, the same 8 do not. - Get more of the retailer's data.Weekly category sales by store are enough to run a read. Item-level sales alongside the category are worth more than everything else on this page combined, and have to be asked for rather than bought.
The floor is also what makes a null result honest. A four-week flight across 25 stores in a typical category cannot see a 3% lift. If that campaign comes back flat, the true statement is that the effect was smaller than the floor, not that the screens did nothing. We publish the floor next to every estimate for that reason.
Sharing individual campaign results in context
A single campaign at a handful of stores is real evidence, just weak evidence, so it should move an estimate by a small increment. When we report a campaign that ran at 12 of your stores, we combine what those stores showed with what the rest of your library showed, and we give you the split between the two. No other attribution report in this category tells you that, and it is part of our commitment to transparency.
What about share of voice?
Weight changes how big your lift is. It does not change how small a lift you could detect, and detection is what this page computes.
More weight, more lift. This is the effect getting bigger.
The detection floor does not move with weight. It is set by store counts, flight length and category volatility.
Where weight does matter
- As the dose in the readStores almost never run identical share of voice, and that variation is the strongest evidence a read can produce. We test whether the measured lift rises with it.
- In the specificationA lift measured at four slots in a twelve-slot loop is a different finding from the same lift at four slots in sixty, so the weight is stated rather than left implied.
Two things we set for you rather than ask
- Week-to-week persistenceHow much of a store's sales persist from one week into the next.
- Category volatilityHow much the category above moves week to week with no campaign running.
Both are properties of a network rather than of a campaign, both are computed from a couple of years of weekly sales by store, and before an actual read we compute them from your own history rather than leaning on the defaults here.
This is the check to run before a matched-store read. The read itself is on Channel Reads.