By: Matt Schiffman, VP Pricing Strategy & North America
Super Bowl Sunday morning, Costco in Seattle. The TVs were moving.
Not the way TVs usually move. People had them stacked in flatbeds, lined up at the exit, loading seventy-five inch screens into cars that were not designed to receive seventy-five inch screens.
For about 48 hours, those TVs were KVIs. Traffic builders. The item people came in for. By Monday morning they were profit generators again. Same SKU, different role. Nothing about the product changed. The shopping context around it did.
That’s the thing most segmentation frameworks get wrong. They treat KVI, sales driver, and profit generator as properties of products. They’re not. They’re properties of moments. A model that assigns static roles is modeling the wrong thing.
And modeling the wrong thing has a cost. Most retailers still run segmentation as a quarterly ritual. Teams sit in a room, assign roles manually, and those roles hold until someone calls another meeting. The TV is a KVI in a spreadsheet on Saturday and a profit generator in reality by Monday, but the pricing logic hasn’t caught up. The gap between those two states is where margin leaks.
Scoring the Moment, Not the Product
That gap is the problem we set out to solve when I designed our new article segmentation model at Quicklizard. The starting assumption was different from most approaches in the market: the relevant question isn’t “what is this SKU” but “what is this SKU right now.” That reframe drove the entire architecture. The model scores continuously across competitive price position, search volume, repurchase rates, traffic cost, footfall conversion, and cross-category substitution relationships. It reclassifies as those signals shift. When the Super Bowl ends, the model knows before the category manager does.
The Long Tail Is Where Segmentation Breaks
Getting this to work at catalog scale required solving a real data problem. SKU-level scoring breaks down when individual item history is thin. We built roll-up logic that aggregates from SKU to segment to subcategory to category, so every product in the catalog carries a score even when its own transaction history is sparse. Coverage doesn’t degrade at the long tail. That matters because the long tail is where most of the catalog lives and where static segmentation is most obviously wrong.
What Happens When the Buyer Changes
But the reason I think this architecture matters beyond operational efficiency, and the reason I pushed for this approach specifically, is what happens to segmentation when the buyer changes.
The signals that define a “moment” right now are behavioral. Where human attention is pointing, what events are driving traffic, which items have entered consumer price memory. All of that assumes the buyer is a person with cognitive biases that pricing teams have spent decades learning to exploit. Anchoring on the original price. Responding to charm pricing. Feeling urgency from a countdown timer. At most retailers, these behavioral tactics contribute 200 to 400 basis points of margin. That’s not optimization upside. For many retailers, that’s survival margin.
Autonomous purchasing agents, already operating at the margins of retail, don’t have these biases. They don’t anchor on the TV. They don’t respond to Super Bowl urgency. They parse substitution value across a full catalog simultaneously, weighted by whatever preference function the consumer configured. The behavioral signals that make a TV a KVI for a human shopper simply don’t register.
Why Contextual Scoring Survives the Shift
This is where the connection matters. The same signals our model already reads, competitive position, substitution relationships, category-level price sensitivity, are the signals that survive the transition to agent-mediated commerce. We didn’t build a dynamic segmentation tool that happens to be useful when agents arrive. We built it because the shift from static product roles to contextual scoring is the same shift the market will be forced into as behavioral pricing loses its grip. The abstraction is right: score the context, not the product. The specific inputs that define “context” will evolve as agent-originated demand grows from edge case to meaningful share. But the retailers running quarterly segmentation rituals today aren’t building on an architecture that can evolve. They’re building on one that has to be replaced.
Static segmentation fails twice. It fails today, when the TV is still classified as a profit generator on game day. And it fails again in an agentic environment, where the entire behavioral foundation it was built on doesn’t apply. The retailers who can’t reclassify their own catalog in real time right now, when the disruptions are Super Bowls and TikTok trends, have no plausible path to competing when agents are reclassifying it for them continuously by simply choosing not to buy.
Costco won’t have this problem. They run 3,800 SKUs, price consistently, and make their margin on memberships. The entire segmentation question is irrelevant to their model. Their pricing never depended on exploiting bounded rationality in the first place.
Everyone else should be asking whether theirs does, and whether the architecture underneath can handle the answer.
