Why a hot AI shopping startup has trouble selling its most ambitious vision

Good afternoon.

If you're living in the U.S. right now, it's impossible to avoid news of Tuesday's midterm elections, which saw Democrats regain majorities in both chambers of Congress.

But one result seems most relevant to this newsletter: the overwhelming rejection of AI data centers. According to Ballotpedia, voters in 10 states weighed in on 41 local ballot measures related to data centers on Tuesday. In 40 cases, majorities favored greater limits on development, either approving restrictions or rejecting measures that would have supported it.

That's a staggering rebuke. And a reminder that people want a say in what's built in their backyards, especially when they're being told that the “product” those facilities are powering is going to overhaul their work and lives.

I also think it's an indictment of how poorly Big Tech and the AI labs have made the case that AI's benefits will be significant enough to justify the disruption to…maybe everything. Plus, it didn't help that companies kept boasting to Wall Street about the efficiencies they're achieving with this technology, seemingly without appreciating, or caring about, how threatening that sounds to everyone but top leaders.

Still, I wouldn't trust anyone who claims to know precisely how this backlash will play out. I could see the biggest technology companies, as well as OpenAI and Anthropic, having the resources to navigate this new realtity even if it slows their plans. Smaller AI companies and data-center developers with less financial flexibility could have a much tougher time.

Moving on.

Friend of The Aisle Scot Wingo, founder and CEO of ReFiBuy, argued Tuesday that the third quarter marked a tipping point for AI-powered product discovery. He points to new McKinsey research showing that 53% of consumers across 11 markets had used AI for at least one shopping-related task in the preceding three months, along with surging usage of AI-powered search experiences at Google and Amazon.

His message to brands heading into the holidays: You'd better know if and how your products are showing up in AI shopping recommendations, even if consumers are still largely completing purchases themselves. Wingo has a horse in this race, of course. ReFiBuy says it helps companies, among other things, improve their visibility in AI shopping experiences through continuously updated product catalogues. Still, I think he's right about the growing importance of AI-powered discovery.

I'm a bit more skeptical of his argument that the shift could already be putting pressure on Amazon's lucrative advertising business. Google did acknowledge that its rapidly growing AI Mode currently generates less advertising revenue than traditional search. Amazon's advertising growth also slowed in the third quarter, although the timing of Prime Day this year versus last complicated the comparison.

But Wingo wondered whether consumers shifting from traditional search to AI-powered shopping experiences could help explain Amazon’s slowdown. His podcast co-host Jason Goldberg suggested Amazon's own Alexa for Shopping might be responsible. Wingo agrees that's plausible, but also wondered aloud whether outside assistants like ChatGPT and Muse are diverting shopping activity—and thus advertising opportunities—away from Amazon.

I find the internal explanation more plausible, in part because of Amazon's claim that more than 400 million customers have interacted with Alexa for Shopping in some way. But there’s little evidence that outside agents are making any sort of meaningful dent. Amazon CEO Andy Jassy explicitly denied that external agents were having such an impact.

Elsewhere, Shopify reported Tuesday that traffic from AI channels increased 2.6 times year over year, while orders grew 3.3 times, albeit from a small base.

And one last number caught my attention. On Uber's earnings call Wednesday, CFO Balaji Krishnamurthy said AI agents had generated approximately 500 Uber Eats orders and a grand total of four Uber rides during the Halloween week, out of 320 million total deliveries and rides. Some perspective.

Now on to the good stuff…

Maju Kuruvilla has some pretty strong feelings about the vendor expo section of the annual NRF retail conference in New York City.

"I want to, like, nuke the whole floor," the former Amazon executive told me during a recent interview.

He was, to be clear, exaggerating. But Kuruvilla has spent enough time in the retail technology industry to have strong opinions on its faults.

Every January, scores of tech vendors descend on the Javits Center in New York City to promote products and services promising to improve some part of a retailer's business. Many boast impressive results. Yet retailers often struggle to translate those improvements into meaningful gains across their operations.

"How is it possible that an entire Javits Center floor of vendors [is] declaring success, and the entire retail industry is going in the wrong direction?" he asked.

It's a fair question to explore, even if rational people can quibble with the suggestion that the entire retail industry is headed in the wrong direction.

Kuruvilla's answer is that retailers have accumulated too many disconnected technology products. And now, with AI shopping experiences emerging across surfaces like ChatGPT, Google, Muse and elsewhere, he believes the limitations of that approach are becoming more obvious.

His Seattle-based startup, Spangle, is trying to build an alternative.