When machines start buying on a household's behalf, they will start with the categories nobody minds losing. Two markets, forty-eight brands, and what the machines already believe about them.
David Roth · August 2026 · BAV consumer data 2026, Aura machine readings, United Kingdom and United States
A category is safe to hand to a machine when the human loses nothing they value by handing it over. In the four-pillar model that condition has a signature: high Knowledge and almost no Differentiation. Everybody knows the brands, nobody finds them distinctive, so the choice expresses nothing and delegating it costs nothing.
That is testable, so here it is tested. Thirty British and twenty-four American household brands, set against sets of well-known identity brands in the same markets, and then read a second time through Aura to see what the machines make of them.
Percentile ranks within each market's own brandscape, 0 to 100. Not percentages, and not comparable across markets as a league table.
Exhibit one
The gap is not a matter of degree. Household brands and identity brands are not near each other on the measure that predicts growth, and both markets say so independently.
Average Differentiation percentile rank. Replenishment sets are household consumables; identity sets are well-known brands in cars, watches, fashion, technology and entertainment, chosen deliberately rather than sampled at random.
Exhibit two
Sorted by Differentiation, the categories line up in almost the same order in both markets. Paper goods are the most exposed thing in the house. Air fresheners, which sell on scent and mood rather than function, are the least.
Average Differentiation percentile rank by category. Several rows rest on two or three brands, so the ends of the ladder carry more weight than the middle.
Exhibit three
Each line is one British brand. The left end is where consumers place its Brand Stature; the right end is where the machines place it. Eighteen of twenty-four fall. Some fall a very long way.
Brand Stature percentile, consumer against machine, United Kingdom. Hover any row for the numbers. The full table is below.
Exhibit four
American brands, on Differentiation. Bars to the left are brands the machine finds less distinctive than the public does. Bars to the right are brands it finds more distinctive. Neither error is small, and the paper's argument is that the first kind is the one you cannot see happening.
Machine reading minus consumer reading, Differentiation percentile, United States.
Exhibit five
This is the finding the American data produced on its own, and it tests the central claim of the paper directly. On the same brands, in the same market, in the same wave, the machine's reading of Brand Stature is roughly twice as accurate as its reading of Differentiation.
Average absolute divergence between machine and consumer percentile readings, twenty-four US brands.
The categories most ready to be surrendered are the ones the machine reads worst. Delegation will arrive first in the aisle where machine judgement is least reliable, among brands that spent thirty years becoming famous rather than becoming distinct. Both conditions are already true, and neither was designed.
Britain is the more exposed market. The machine reads American household brands about twice as accurately, which is what you would expect if machines read a record rather than a market: far more has been written about Tide and Clorox than about Comfort and Harpic. A British brand is not being treated worse. It is less written down.
A category average will not save you. The errors scatter in both directions within the same aisle, so a brand has to read its own position. That is the practical case for measuring machine perception as a standing item rather than assuming the category's fate is your own.