The CAC crunch goes mainstream

I don’t usually link to the New York Times from here, but make a note: Oct. 7, 2026, Section A, Page 20. Check the print copy at the library if you don’t have other access.

In Meta Is Crushing Small Online Retailers, Corissa Steiner writes,

New business openings, e-commerce included, have exploded since the pandemic, but long-term viability is questionable. Big Tech platforms like Amazon and Meta are turning the screws on American mom-and-pop online businesses and empowering deep-pocketed mega-retailers like Quince, which a Times newsletter called a “well-branded dupe emporium.” That’s reducing choice for consumers and limiting paths to upward mobility for millennials like me and generations that follow.

And it only gets worse from there. Meta and Google are already an advertising duopoly, but the market expects both companies to keep growing at startup-like rates. Which means taking a bigger and bigger piece out of every sale.

Sometimes the customer acquisition costs (CAC) jump up quickly, as the duopoly can pass the costs of any disruption on to the advertisers.

In 2021, Apple rolled out a new policy that limited iPhone apps’ ability to track user behavior. It was a win for privacy, but it also made ads on Meta platforms Facebook and Instagram less effective….Researchers found a 39 percent reduction in revenues for Meta-dependent advertisers as a direct result of the switch, an effect they noted was felt primarily by small businesses.

But, of course, after Meta successfully adapted to ATT, ad rates didn’t come back down.

In the second quarter this year, Meta’s average price per ad increased 12 percent over the previous year. These costs aren’t just shouldered by the business; they get passed on to you, the consumer — an overlooked answer to the question of why everything is so expensive.

Or maybe if a retail business model depends on the duopoly, it’s not so much of a business model? The first companies to start using Facebook for reaching people on the internet did well for a while. But now that the Meta ads are costing basically the profit from the sale, at some point the privacy nerd customers (who aren’t really reachable on the big social sites) are going to be a hot segment to go after.

Or are shopping agents going to save us?

The optimistic view of “agentic AI” for shopping is that the shopping bots are going to see through all the surveillance ads and the related “retail media network” scam (which is slotting fees, not ads) and just help you pick out the best stuff at the best price.

Shoppers will be able to send the shopping bot, I mean “agent,” to fight the surveillance advertising and surveillance pricing bots.

But that means retailers would have to under-invest in their own “AI” in order to focus on do content marketing as a core competency. Basically push every available piece of accurate product info and customer feedback into some well marked-up web page or other place where crawlers/bots/agents can get at it. Fix any capacity and compatibility issues that get in the way of “the more you tell, the more you sell” but other than that stay light on “AI” for marketing, and win on some combination of price, quality, and service.

That would be a culture shock because a zillion startups are going to try to scrape existing brands’ product info and support pages and make a different “customer journey” and marketers would have to sit back and let them—you can’t afford to both make the product that other people’s agents would select and win an AI-vs-AI info battle. The brand has to be able to rely on either an existing winning combination of price/quality/service or be confident in ability to improve based on what sells or doesn’t. But that’s probably a safer bet, considering that at any given time, somebody’s crawlers/bots/agents are going to be able to see through any info asymmetry design.

Also a culture shock because management decision-makers are often resume-driven, and will be uncomfortable with less buzzwordy approaches such as “we went back to real ingredients and waited for shopping AIs to train on food bloggers who say our product is tasty” instead of resume-friendly AI-heavy projects.

I guess it depends how well you think the balance of AI personalization vs. de-personalization will go. (Personally, I think markets work better than central planning.)

Bonus links

AI chatbots are ignoring this prompt from wealthy users, in a phenomenon called ‘adversarial delegation’ by Jude Cramer. Major models including Claude Opus 4.8, Gemini 2.5 Flash, and GPT-5 recommended products more than $100 more expensive to high-income users versus low-income users. (Robin Hood, Robin Hood, hm hm hmmm hm hm…I’m sure they’ll fix this to correctly implement “poor tax” pricing at some point.)

Advertising Measurement Is Optimizing for Technical Convenience, Not Business Impact by Rick Bruner. In our work running media ROI experiments, we often find that a trackable online channel shows little or no incremental advertising effect, while the advertising has a substantial impact on sales through an offline channel that the advertiser isn’t measuring. This is endemic to the ecosystem. It is a structural limitation of user-level advertising measurement.

Money can’t buy enterprise trust by Matt Asay. (This is what people are missing about AI code quality. A vibe-coded replacement for a SaaS application doesn’t have to be more trustworthy than the established SaaS codebase—it just has to be more trustworthy than the SaaS company founders, who can drop support, sell out to private equity, or whatever. Another clipping for the rugpull economy file, I guess.)

Dungeons & Dragons, Duel Masters, and Magic: The Gathering Teams Unionize At Hasbro by Nicole Carpenter (Roll perception for the union label.)

Dutch eyewear chain Hans Anders stops selling Meta’s Ray-Ban glasses by Ana Maria Constantin. The halt followed a report in Dutch newspaper AD. It described strangers filming people, especially women, in public without their knowledge. Last month the Dutch data protection authority, the AP, warned smart glasses users that they generally may not share videos of identifiable people without permission.