The candy at the checkout exists because you're tired. You've made forty decisions walking through the store. Your willpower is depleted. A Snickers bar costs two dollars and requires no thought. The placement isn't random. It's the result of decades of research into when humans are most susceptible to purchases they didn't plan to make.
Impulse purchases account for somewhere between 40 and 80 percent of all retail transactions, depending on how you measure. The variance in that estimate tells you something: the behavior is so embedded in commerce that researchers can't agree on where planned purchasing ends and impulse begins. What they agree on is that the revenue is enormous and the margins are high.
An agent executing a grocery order doesn't get tired. It doesn't walk past endcaps. It doesn't experience the checkout aisle. It receives a list, optimizes against constraints, price, nutrition, availability, brand preferences if specified, and executes. The Snickers bar never enters the consideration set because it was never on the list and the agent has no willpower to deplete.
Where the Money Actually Is
Grocery is the obvious case, but impulse revenue runs through retail like a nervous system.
E-commerce upsells. "Customers who bought this also bought" exists because humans are suggestible and uncertain. We second-guess our choices. We worry we're missing something. Recommendation engines exploit that uncertainty. An agent doesn't second-guess. It selected the item that met the requirements. The requirements didn't include "things other humans bought."
Travel add-ons. Seat upgrades, insurance, priority boarding, lounge access. Airlines and booking sites present these after you've committed to the flight, when the psychological cost of abandoning the transaction is high. An agent booking travel has no sunk cost psychology. It evaluates add-ons against policy. If the policy doesn't authorize lounge access, the modal window is irrelevant.
Subscription expansions. "Upgrade to premium" prompts catch humans in moments of engagement, when the product has just delivered value and resistance is low. Agents don't experience engagement. They evaluate whether the premium tier delivers functionality the principal requires. Enthusiasm is not a factor.
Extended warranties and protection plans. These are profitable precisely because humans overestimate risk when making purchases and underestimate their own tendency to never file claims. An agent calculates expected value. The math rarely favors the warranty.
The Margin Structure Problem
Retailers don't think of impulse revenue as a separate line item. It's woven into how stores are designed, how sites are built, how checkout flows are engineered. The assumption that a percentage of purchases will be unplanned is baked into margin expectations.
When that percentage declines, the economics shift in ways that aren't obvious from the top line. Revenue from planned purchases may hold steady while the profitable unplanned additions disappear. A 10 percent decline in impulse revenue hits harder than a 10 percent decline in overall sales because the impulse purchases carried a disproportionate margin.
The products most dependent on impulse, candy, magazines, batteries, accessories, small electronics, face category-level disruption. Their placement value collapses. Their pricing power erodes. Manufacturers who paid slotting fees for checkout proximity discover they purchased access to a channel that's going quiet.
The Timeline Question
How fast this happens depends on how quickly agent-mediated purchasing scales. The technology exists. The adoption curve is the uncertainty.
Grocery delivery already removes the checkout aisle. Subscription services already automate replenishment. Voice assistants already handle simple reorders. Each of these chipped away at impulse revenue before agents entered the picture. Agents accelerate a trend that was already moving.
The retailers most exposed are those whose margin models assume the highest impulse percentages: convenience stores, airport retail, checkout-heavy formats. The categories most exposed are those whose purchase rationale is weakest under scrutiny: the products humans buy because they're there, not because they were needed.
What Survives
Not all unplanned purchases are irrational. Sometimes you genuinely didn't know a product existed until you encountered it. Sometimes discovery has value. Agents that surface options the principal would want but didn't think to request could preserve some of this function.
The difference is who benefits from the discovery. Human-era retail optimized discovery for margin. Agent-era commerce could optimize discovery for principal utility. The agent that recommends a product the human would have wanted isn't engaging in impulse exploitation. It's providing a service. Whether retailers can capture value from that service is an open question.
What clearly doesn't survive is the exploitation of cognitive depletion. The tired shopper, the committed traveler, the engaged subscriber, these are states that create extractable value only because humans experience them. Agents don't get tired, don't feel committed, don't experience engagement. The entire apparatus built to extract value from those states becomes overhead.
The Conversation That Isn't Happening
Retail strategy discussions focus on omnichannel, personalization, supply chain optimization. The assumption that humans will continue making purchasing decisions, and making them imperfectly, is rarely questioned.
CFOs modeling future revenue are not, in most cases, modeling a world where impulse declines by 30 percent over five years. The scenario isn't in the forecast because it requires questioning an assumption so foundational that it doesn't feel like an assumption. Humans buy things they didn't plan to buy. That's just how retail works.
Until it isn't.
The retailers who notice first will have time to restructure, margin expectations, category mix, store formats, checkout design. The retailers who notice when revenue declines will discover that the infrastructure of impulse exploitation is expensive to maintain and hard to repurpose.
This is part of a series on the robotic web from TinyFish, which builds infrastructure for machine operation of the web.