Are Behavioral Pricing Tactics Dying? A Study of Price Endings & Partitioned Pricing Effects on Machine Buyers

Source: Utpal Dholakia; The Pricing Conundrum (Substack) · 2026-10-17

When an AI agent shops on a customer’s behalf, the efficacy of behavioral pricing tactics must be reassessed. I tested four leading AI agents as shoppers and came away concerned about the future.

Without a doubt, machine buyers are rising rapidly. Over the past year, most major technology platforms have built and implemented infrastructure for AI agents to shop and make purchases (including payments) on a consumer’s behalf. For instance, in September 2025, OpenAI launched Instant Checkout in ChatGPT, initially for US-based Etsy sellers, allowing users to make a purchase without leaving the chat. In January 2026, Google announced its Universal Commerce Protocol, developed in collaboration with Shopify, Walmart, Target, Etsy, and Wayfair, which lets shoppers check out inside Google’s AI Mode and Gemini. And more recently, Meta’s Muse agent went viral for its shopping proficiency, so much so that Amazon blocked it from its retail site in September 2026. A few weeks later, Meta and Sierra proposed a standard for how personal AI agents should deal with businesses. McKinsey estimates that agents could orchestrate $900 billion to $1 trillion in US retail sales by 2030 (only four years from now!)

For obvious reasons, this is a watershed moment for the practice of behavioral pricing. Everything we know, seek to understand, and apply in stores and on websites and apps is based on the understanding that our buyers are human, undergoing a human decision making process that culminates in a purchase by humans. As a simple example, charm prices (i.e., prices ending with 9) are posited to work because people read numbers from left to right and give greater weight to the leftmost digit. This makes a price like $49.99 feel much lower than $50.00 than it really is. In the same way, partitioned prices (the division of a total price into two or more components, which may be mandatory or not) lead people to anchor on the base price and underweight secondary (typically smaller) components, such as surcharges or shipping fees, resulting in an underestimation of the total price.

Thus, based on extensive bodies of research on charm prices and price partitioning, our advice to pricing practitioners is to liberally use prices ending in 9 or 99, and to avoid presenting prices in all-inclusive formats, and present itemized bills instead. Like these two effects, virtually every behavioral pricing tactic is based on an understanding of consumer decision making where the consumer is a human. What happens to behavioral pricing tactics when the buyer is no longer human, but a machine? This is the pricing conundrum I want to explore in this post.

There is interesting emerging research that gives us some clues. The first is the idea that machines will be rational buyers. Back in 2017, well before the current mainstreaming of AI agents was even on the horizon, Gal and Elkin-Koren argued that so-called ‘algorithmic consumers’ could strengthen the buyers’ side of the tussle with brands by searching widely and resisting persuasive manipulation. But the authors also perceptively cautioned that much would depend on who controlled the algorithms running the algorithmic consumers. It seems the current consultant and industry discussion on this issue has adopted the authors’ optimistic take without acknowledging their caution. For example, McKinsey has written about dynamic pricing and how agents and platforms will be paid, while other expert analyses have focused on the infrastructure sellers will need to serve machine buyers, such as payment cards for agents and machine-readable product data and pricing.

In a recent HBR piece, Sabbah and Acar reported running thousands of simulated shopping rounds with four AI models covering four different product categories. They found that marketing tactics such as scarcity messages, countdown timers, strikethrough prices, vouchers, and bundles had inconsistent effects on machine purchases, varying by model. Only star ratings based on reviews consistently increased choice by AI agents in their study, while higher prices consistently reduced it. Their advice to marketers was to “treat AI models as distinct segments.”

While I agree with the authors’ overall conclusion, one issue needs to be probed further: the fact that marketing tactics designed for humans fail with machine buyers does not mean that the machine buyers are rational. It raises two important follow-on questions that are crucial for pricing managers and consultants (and indeed all marketers) to know: (1) What do machine buyers, a.k.a. AI agents, respond to instead, and (2) Who decides the machine’s decision calculus? In the rest of this post, I summarize research on how AI agents shop, describe two experiments I ran with four AI models, including the method and results, and discuss what the findings mean for pricing managers and consultants.

A small but fast-growing body of research has had AI agents perform shopping tasks to study how they differ from humans. In many ways, AI agents do shop rationally. For example, agents favor higher-rated products and prefer lower prices. But the research also shows that agents often respond to cues that are not informative and to small differences between options as if they were large. One comprehensive study tested 17 AI models in a simulated web store and compared them with human shoppers. A product’s position on the page shifted one model’s choices by 89%, even though the position was entirely incidental, compared to only about 4% for humans. The authors also found that modest differences in price or ratings moved agents “in a nearly binary fashion,” with larger differences adding little incremental shift in choice (because the ceiling had already been reached). And they responded far more intensely to ratings than humans did. A higher star rating increased the share of a product’s selection by 30% to 80% in 14 of the 17 models, compared to only about 5% for people.