Forward buying means the customer buys more in the promotion window because they will not buy the same goods later at full price. Window sales rise and sales uplift sits near zero. In FMCG distribution the pattern shows on both distributor and retail invoices: the warehouse fills, then a dip follows the campaign. Didrah’s AI agent scores effect against a baseline, not against how busy the festival week looked. If a matching dip follows, the claim should be forward buying, not a winning mechanic.

How forward buying differs from cannibalization in distribution

Cannibalization eats a neighbouring brand in the same week. Forward buying borrows the same brand’s sales from later weeks. Both can make the window look busy and leave effect empty. The question to the data is not the same. Cannibalization needs a second brand dropping in step. Forward buying needs the baseline after the window. If an AI agent mixes the two, budget follows the wrong finding.

How an AI agent sets the promotion baseline in distribution

The baseline must be the same brand, same channel, same seasonal window. Comparing Yalda with a quiet week and no campaign is a calendar contest. Several weeks after the campaign must be read as well, or forward buying stays hidden. Trade marketing often closes a mechanic on window sales. Agentic AI should put the window, the pre-window, and the post-window in one claim. If the post-window is missing from the data, that is a data gap.

When the agent flags forward buying, what you do with the mechanic

Rerunning the same tiered discount usually reruns the same loan. The rational call is a shorter window, a tighter purchase cap, or moving the mechanic to a channel whose warehouse cannot sit on stock. The agent proposes that. Resetting the budget is still human. If the campaign had a goal other than short-term sales, write that in the risk so the claim is not oversold.

What to ask the agent about window sales versus effect

  • Do not call window sales effect until the pre and post baseline is in view.
  • Read the post-campaign dip apart from a neighbouring brand’s drop.
  • If the post-window is missing from invoices, the agent should stop, not declare a win.
  • Do not rank a mechanic by campaign slogan; rank it on the same baseline.

After the festival, what the distribution AI agent should rerun

The week after the campaign matters more than the campaign week. Same brand, same channel, same region: did volume return to baseline, fall under it, or stay up? A fast return to baseline is classic forward buying. Staying up may be real effect, if cannibalization is not in the same window. The agent should write those two apart. If you leave the post-window for “next month in Excel,” you no longer have agentic AI. You have a delayed report. Hold the budget to rerun the mechanic until the post-window is in view.