Usually it is not enough. Causal lift needs daily granularity. A month-long campaign at weekly granularity gives the model four data points, which is not enough to detect anything with confidence.

Four weeks of campaign at weekly granularity is four observations in the intervention window. No model separates signal from noise on four points, regardless of how much pre-period you supply.

Daily data over the same flight gives roughly thirty, against a pre-period of several hundred. That is a model that can say something.

Six months of daily history before campaign start is the minimum. Twelve months is preferred, because a full year lets the model see a complete seasonal cycle and separate your normal annual shape from the campaign.

Between six and twelve months you will get a study. Confidence tracks the length and quality of the pre-period, and the significance figure will reflect it. Below six months there is not enough to model against, and we will tell you so.

It happens, and it is not always solvable. Some advertisers cannot permit pixeling and can only supply a weekly store-level sales file. That combination rules out both web attribution and causal lift.

What is left in that situation is brand lift and search lift, neither of which needs your transactional data, plus footfall attribution if the stores are advertiser-operated. If the product is sold through third-party retail rather than in your own stores, footfall drops out too, because a visit to a chain store says nothing about whether your product left with the shopper.

That is a smaller measurement story and it is better than nothing, but decide it before launch rather than discovering it in the wrap.

If daily data exists somewhere in the business and the obstacle is internal approval, it is usually worth asking. The export is normally a one-time request to a performance or analytics team.

Daily data that cannot leave the business in raw form can still be sent obscured. Pick one random number, multiply every event count in the file by that same number, and send the transformed file. When the results come back, divide by the number you chose to recover the true counts.

The model is unaffected, because the transformation is uniform: every row is scaled identically, so the shape of the series and the lift measured against it are unchanged. Keep a record of the number you used, since only you will have it.

This turns the daily data question from a disclosure decision into an arithmetic one, and it is worth raising before you conclude that causal lift is closed to you.

Myth: Weekly sales data is enough if there is plenty of history.

Reality: The constraint is observations inside the campaign window, not total history.