I have spent most of my career selling things I never once watched anybody buy.

That is the job description. In consumer goods you make the product, somebody else sells it, and the moment money changes hands happens in an aisle or a search box you do not own and cannot see into. Everything we built after that is a workaround. Panel data. Share reports. Clean rooms. I have described a data clean room as a middle school dance: your first-party data lined up along one wall of the gym, the retailer's along the other, a heavily chaperoned temporary interaction, and then everybody goes back to where they started. It gets a laugh in a lecture theatre. It is also an admission. We invented the dance because we were never invited to the transaction.

Fine, until you start making spending decisions as though you had been.

Nik Sharma wrote up the cleanest example of this I have seen. A brand expanded nationally into Target. Its direct-to-consumer revenue fell sharply, Amazon grew, and total company revenue was reportedly up around 30 percent year over year. Through all of it, the DTC dashboard's return on ad spend and its cost of acquisition both got worse (Nik Sharma, Limited Supply, August 9 2026). In the same piece, a second brand pulled back its web-focused media and watched Amazon revenue dip almost immediately.

Those are one story told from both ends. In the first, the company grew and every screen in the building reported a problem. In the second, a screen reported a saving and the company got smaller.

The shopper is not doing anything strange. She sees it on Meta on a Tuesday, checks the reviews on Amazon on Thursday, and buys it at Target on Saturday because she was already there for washing powder. Sharma's line for this is the right one: customers do not organise their purchases around your channel P&L.

Here is what changed, and it is why a measurement problem belongs in an AI newsletter.

A wrong number used to move slowly. A director read it quarterly, weighed it against what she knew about her own business, and made a call that carried her context with it. Most of those calls have now been handed to a system that optimises whatever signal you gave it. Feed an automated bidding system a conversion event that only fires on your own website, and it will do precisely what you asked. It will find more people who buy on your website. It will also defund, steadily and without malice, the work that grows the sales it was never shown.

Let the machine be the machine. It is better than any of us at aiming spend at the outcome you specified. It has no way to tell you that you specified the wrong one. And it runs that decision hourly, where a human ran it four times a year.

Where this lands is not the reporting pack.

The media that gets cut first is whatever sits furthest from a checkout you can see, which in most large-cap consumer goods brands is the work that builds demand for somebody else's shelf. The spend that survives is whatever sits closest to a measurable transaction, which is usually your smallest channel and often your least incremental, because the people it finds were already coming. Retail media keeps growing partly on merit and partly because it arrives with a receipt attached. And in the review itself, the person defending brand investment walks in with a story while the person cutting it walks in with a number. I have watched that argument end the same way for fifteen years.

One rule, and it costs nothing to start before the next planning cycle. No channel's budget moves without writing down first, on one line, which other numbers you expect to move, in which direction, and by when. Amazon. Nielsen or Circana share. Retailer sell-through. Total revenue. Then go back and read them.

Change the order of the review to match. Total company revenue and total media spend at the top, for the whole business. Channel splits underneath. The order you read them in decides which one you end up defending.

If you want a single number to argue with, use new-to-brand. Absolute new buyers coming into the brand over time. It is the one I keep returning to, because a shopper can only do it once and it does not care which door she walked through to do it.

You are running a business whose customers buy in buildings you do not own. The dashboard will keep reporting on the part of it that it can see. The work is remembering how small that part is.

-- Imteaz

If this one landed, two earlier pieces on the same thread:

This one is for whoever owns a channel P&L, and for whoever has to defend a budget they cannot prove.

Or just hit reply and tell me one thing: which channel in your business is currently getting credit for sales it did not create? I read every one.