Primary sales in distribution are goods leaving the manufacturer into the distributor warehouse. Secondary sales are goods leaving that warehouse to retail and wholesale. Both numbers are real, and both can mislead. If you only celebrate primary, you filled a warehouse, not a shelf. Didrah’s AI agent may join the two only when SKU, period, and region match. Otherwise you have two charts, not one claim.
Why the FMCG supply chain keeps these two layers apart
The manufacturer’s system usually sees the distributor order. The distributor’s system sees the retail invoice. Dates differ, units differ, and a brand’s SKU code can differ on each side. Data normalization is not a side task here. Without it the agent must not say “sales grew.” Data quality is this exact point: if a primary row cannot be joined to a secondary row, write the data gap. Do not fill it with a guess.
What an AI agent checks before it joins the two layers
Grounding means the claim is built from rows that actually align. The agent should say which week primary rose while secondary stayed flat; that pattern is warehouse build. Or secondary fell while primary stayed high; the network is selling into a warehouse that retail is not emptying. Agentic AI is useful here because it runs that comparison on a fixed rule, every cycle, not once in an end-of-month spreadsheet.
What decision is rational after a primary and secondary gap
If primary is running ahead of secondary, the usual call is to stop pushing the distributor, not to launch a new shelf campaign. If secondary is ahead, stock is draining and the next question is promotion or coverage. The autonomy ladder means the agent proposes this; a human cuts or releases the production order. Automatic action on primary orders, before the evidence is stable, skips governance.
What to ask the agent about primary and secondary sales
- “In this region, this brand, these four weeks: how far is primary ahead of secondary?”
- If the SKU keys do not match, the agent should say unmeasured, not approximate.
- Do not treat a primary spike as sales uplift at the shelf unless secondary evidence is there.
- Do not bury the data gap; it limits the claim in the first sentence.
One month of reading primary and secondary with the agent
Keep the same three keys every week: item, region, period. If primary ran ahead one week, the next week shows whether secondary followed or the warehouse held. Two weeks of build with no retail outflow means ease the push into the distributor, not grow the shelf campaign. An AI agent repeats that comparison on a fixed rule so the meeting does not become a war of definitions. If the item key breaks one week, drop that week from the claim. Averaging over a broken week hides the data-quality problem.