When planning your weekly grocery order on Monday morning, you think through meals, essentials, and staples. But do you add the crisps or chocolate you'll want to grab Friday evening? Probably not.

And that's exactly the problem AI shopping agents face with vice products.

The Planning Gap

Working with grocery data revealed a clear pattern: even next-day delivery significantly reduced vice product purchases. The longer between order and delivery, the fewer snacks and treats end up in baskets.

When we plan ahead, we shop for an idealized version of ourselves, not the person who craves cookies after a long day. This gap between intention and reality eliminates indulgent purchases, even though they're part of what customers actually want.

How Impulse Buying Works

Impulse purchases happen in the moment, triggered by visual cues, emotional states, or immediate contexts. You grab cookies because you see them and suddenly want them, or you're treating yourself after a long day. When your AI agent asks Monday what you need for the week, that Friday state of mind isn't there to trigger the purchase.

The conversational nature of AI agents may worsen this. Asking a human-like assistant to "add cookies to my order" feels different from quietly dropping them in your basket at a physical store. This conversational interface might create a subtle social judgment that doesn't exist when shopping alone, potentially systematically suppressing specific product categories.

With recurring orders, the problem compounds. Would you deliberately set up auto-delivery of chips, cookies, and chocolate to your doorstep weekly?

Why This Matters

Vice products aren't just a revenue concern. They're a genuine consumer problem. People still want these items, even if they don't plan for them rationally. Friday treats, movie night snacks, cookies to share with family and friends—these small comforts matter. Forgetting them means last-minute store trips, undermining the convenience AI shopping promised.

For retailers, the impact is substantial. Specifically for online grocery delivery where the economics only work with substantial basket sizes to offset operational costs. Missing products hurt profitability directly.

Worse, there's no guarantee customers will visit your(!) store for forgotten items, they might go to competitors or do without. Keep in mind that vice products often trigger complementary purchases. Browsing the snack aisle for crisps, customers frequently buy multiple vice-category items and often also add some soft drinks, beer, or wine. These additional purchases vanish in AI-assisted shopping where the initial item was never requested.

The Broader Challenge

Perfectly predicting grocery baskets is far more complex than current AI hype suggests. The average in-store basket contains 7-20 products, while eCommerce baskets can exceed 60 items*. Each product is mundane individually, but collectively they reflect highly personal preferences. Missing even one or two creates frustration and lost revenue.

AI agents excel at efficiency and prediction but struggle with the fundamentally irrational and spontaneous nature of vice purchases. Solutions need to go beyond simple prediction, recreating the spontaneous discovery and impulse purchase opportunities that happen naturally in physical stores.

_{* simplified for illustrative purposes. The size of the basket is not that straightforward to define. It could refer to unique SKUs, total number of items, distinct products, monetary value, etc.}