Last week I explored how AI shopping agents struggle with vice products because we don't plan for Friday's cookie craving on Monday morning. But there's a deeper challenge: the less you see, the less you buy. And AI agents, by design, show you almost nothing.

The Efficiency Trap

Over nearly two decades analyzing grocery data, I've watched the same pattern emerge repeatedly. Online shoppers get more efficient over time. They use favorite lists, use targeted search, and develop routines that enable them to build their baskets faster. Great for convenience, terrible for basket size.

These experienced shoppers nail their weekly staples. But they also forget items that don't pop up in their routine. When products go out of stock or get reformulated with new SKU numbers, they simply disappear from the shopping list. We documented cases where baskets actually shrank over time, not because customers needed less but because they saw less.

This hurts both sides. Retailers lose basket size and margin. Customers forget items they genuinely need or want, only remembering them mid-week when it's too late. That efficiency gain evaporates the moment you make an extra store trip the next day.

Discovery Drives Purchases

Consider wine during the holiday season. Many retailers suggest a bottle that pairs with specific ingredients, both in-store and online. Smart merchandising. Add one bottle, done. Efficient shopping experience delivered.

But now that shopper never visits the wine section. They don't see the dessert wine that would've been perfect for after dinner. They don't notice a few bottles worth stocking up on. There's no opportunity for up-selling from the suggested bottle to something special at a higher price point. You made shopping efficient by collapsing an entire category into a single recommendation.

Physical stores understand this intuitively. End caps, cross-merchandising, strategic placement of promotions. All designed to put products in your path. You came for milk, but you left with milk, cheese you didn't know you needed, and that interesting new yogurt you spotted. There's a reason research consistently shows grocery stores with larger assortments sell more. Despite what the paradox of choice might suggest, more options create more opportunities for discovery.

There's another layer too: browsing triggers subconscious inventory checks. You see laundry detergent and suddenly remember you're running low. Online shopping reduces these serendipitous discoveries and limits these mental triggers. AI shopping agents may eliminate both entirely.

This matters more than you might think. Industry research suggests impulse purchases can represent over half of grocery revenue, with some categories reaching 80%.

The Agent's Impossible Balance

AI shopping agents face a fundamental dilemma. Show too many options and you're not really adding value; nobody wants to review multiple suggestions for every item on their list. But show too few and customers only get what they explicitly requested.

An AI agent could take some liberty and add items proactively. But this is tricky to get right. Would you accept your AI agent splurging occasionally on your behalf? Maybe. But when? And by how much?

When hosting that couple you usually meet at expensive restaurants, upgrading the wine makes sense. But what about the casual Friday dinner with friends who aren't wine people, should it surprise you with something special anyway? And if so, by how much?

These judgment calls are hard enough. But they miss the deeper issue: no one asks their AI agent for things they don't know they want, need, or even know exist. The average grocery supermarket launches thousands of new products annually. A generic AI shopping agent will show you precisely none of them unless you specifically ask. And you can't ask for what you don't know is there.

Why This Matters

This discovery problem compounds the vice product challenge I explored last week. Together, they create a systematic reduction in basket size that undermines both customer satisfaction and retailer economics. Online grocery delivery only works economically with substantial basket sizes to offset operational costs. The same is true for brick and mortar stores. If everybody only bought what they had in mind going in, the entire grocery retail model collapses.

For customers, the promise of AI-assisted shopping is convenience and completeness. Neither works if you're constantly remembering forgotten items or missing products you would've wanted. The efficiency gains become efficiency losses when you're back at the store a day or two later.

For retailers, smaller baskets mean lower revenue and potentially unprofitable delivery operations. Regular store visits also depend on customers finding more than they came for, making the trip feel worthwhile rather than just efficient. But there's a deeper strategic concern: if AI agents collapse product categories into single recommendations, entire market segments disappear from view. New products never get discovered. Premium options never get considered. The long tail of inventory becomes invisible.

The grocery industry has spent decades optimizing product placement, merchandising, and discovery. AI shopping agents may undo all of that in pursuit of efficiency. The question isn't whether AI can handle grocery shopping—it's whether efficiency-focused AI can preserve the discovery that makes grocery shopping actually work for the retailer and customers.