Attribution after iOS tracking changes
Platform numbers drifted apart in 2024. Our data-driven model closes the gap without guessing.
At Kelvra Analytics we see this question every month, so we wrote down what we actually do. The short version: measure first, change one thing, measure again.
Three details that matter in practice. First, write down the assumption you are testing before you look at the data. Second, keep the change small enough that you can revert it in an afternoon. Third, tell the people affected before you ship, not after.
We will update this post when our approach changes. If you have a case that does not fit, write to us; the counter-examples are the interesting part.