When one item is on promotion, shoppers who would have bought a sibling — another size, flavor or variant in the same portfolio — switch to the promoted one. That is cannibalization: the promoted item’s lift is real, but part of it was paid for with sales the manufacturer was going to make anyway. Halo runs the other way: the promoted item draws shoppers who also buy complementary products, which sell more without being discounted.
Both effects sit outside the promoted item’s own numbers, so a review that looks only at that item misses them. A promotion with strong lift and heavy cannibalization can lose money across the portfolio; one with modest lift and a strong halo can be worth more than it appears. Retailers and manufacturers also count them differently — a retailer cares about the whole basket, a manufacturer about its own brands — which is why the same event can be judged a success by one and a failure by the other.
Both are estimates, built from the same kind of baseline as the promoted item, and they are only as defensible as the method used to build them.