In many distribution companies a campaign report is a narrative: what we aimed at, what we ran, what window sales were. Narrative is familiar in a board pack and insufficient for a budget decision. Didrah’s AI agent - agentic AI built for FMCG distribution - splits the answer into three layers that have to be read together: claim, evidence, and risk.
The claim must be a falsifiable sentence
“The campaign went well” is not a claim. “The spring tiered discount came in 12 percent under budget and sales effect was not measured” is a claim, because it can be wrong. If a claim has no number, you are still in story. If it has a number whose source is a slide rather than an invoice, you have not reached evidence yet.
Do not bury risk and data gaps
Risk is what could invalidate the claim, for example that the campaign had a goal other than short-term sales. A data gap is what was never measured at all. Reading campaign results without both manufactures false confidence. The agent should say both; the manager should decide with them in view. That is human control, not a signature at the bottom of the pack.