The Second-Largest Line on Your P&L Has the Worst Data Behind It

Written by Cynthia Aadal, Senior Director of Retail & CPG at SDG Group USA

The merchant has a daily sell-through. You have a report from three weeks ago. Guess how that meeting goes.

I spend most of my week with commercial and supply chain teams across CPG, and the pattern barely changes from one company to the next.

The promo runs. Volume moves 22%. Everyone claps.

Then the buyer asks what the baseline was, how much came out of the multipack, and how much of that lift was borrowed from week six. If that answer takes four days to assemble, nobody will run a promotion. Somebody funded the retailer's quarter and called it a win.

Then the second bill arrives. A chunk of those stores went out of stock by Thursday, so the lift was real and so was the demand that walked out without buying anything. The DC saw the order surge only after it happened, because the replenishment plan was built on a baseline that never knew the promotion was coming. Finance closed the month calling 22% a success.

Four systems watched the same event. Shipments in one, scan data in another, trade spend in a third, replenishment in a fourth. They told four different stories, and not one of them was lying. The hierarchies don't map across them, and nothing in that stack was built to answer a question on a Tuesday afternoon.

So the second-largest line on the P&L gets defended with slides built on data old enough to have a different price on it, forecast against a supply plan that was never told the promotion existed.

The standard fix doesn't fix it. More dashboards, more visibility, another portal login for a team that already has six. And the newest version of the standard fix is a chatbot sitting on top of that same plumbing, answering confidently from numbers that were already wrong. That isn't a forecast. It's a faster way to be wrong in front of a buyer.

Where AI actually earns its keep here is unglamorous. It reconciles. Mapping SKU hierarchies that don't line up. Tying shipments to scan data to trade events to replenishment signals, continuously, instead of as a three-week project every quarter. That's the part people can't do at CPG scale, and it has to work before any model sitting on top of it means anything.

Get that right and FP&A stops being a rear-view mirror. The forecast re-runs when the replenishment signal moves, not when an analyst has time. A promo scenario prices in the service-level risk before the event instead of the write-off after it. The margin number already knows what the supply chain is about to do.

That's the version of AI-driven FP&A worth paying for.

A plan that updates itself when the shelf changes.

You'll know you have it when three questions get answered in the room instead of four days later:

1. Did it sell, or did it just ship?
2. Could supply actually cover the lift we paid for?
3. Does that number survive the buyer's follow-up?

I'm not bringing a deck. I'm bringing the sell-through. Book time with me here.