Cannibalization in distribution usually arrives as a feeling in the meeting: “this bundle took that brand’s sales.” Feeling is not enough. You need to see whether the second brand’s drop lined up with the first brand’s campaign, in the same region and week, and whether that drop did not also appear in channels the campaign never touched. Without those two conditions you just have two unrelated charts. Spotting that overlap is exactly what an AI agent does across the FMCG supply chain: not by feeling, by invoice.
The question to ask the data
The right question is not “does cannibalization exist?”; in a multi-brand book it almost always does, a little. The question is whether its magnitude outruns the campaign’s measured uplift. If it does, the campaign spent the company’s own pocket. Agentic AI is useful here because it runs that question on real invoices, not on an industry average.
The decision after the finding
A finding on its own is a slide. The decision is not to rerun the mechanic on the injured brand, to move shelf share, or to take the campaign to a channel with less basket overlap. Human control is this exact point: the agent shows the pattern; the sales manager chooses which brand is allowed to lose, if any brand should lose at all.