Earlier this week, my colleague and Term Sheet lead author Allie Garfinkle wrote about a young AI-powered shopping startup called Phia that helps consumers find the best prices on a fashion product in an instant, whether new or used. Earlier this summer, I wrote about the launch of the AI-powered fashion styling app Daydream too.
As a self-admitted commerce nerd who has reported closely on e-commerce since 2013 (in tons of articles, yes, but also in book, podcast, and onstage interview form), the rise of companies big and small that are using generative and agentic AI systems to innovate in and around consumer purchasing decisions has been a breath of fresh air.
Today, even as a naturally skeptical person allergic to BS, I feel like the sector is on the verge of legitimate widespread disruption. Many new companies will fail, but a wave of new ideas should be welcomed. So when I saw two investors from Andreessen Horowitz—general partner Alex Rampell and investing partner Justine Moore—recently published their thesis on how an AI-powered commerce revolution might play out, I was intrigued. (If you’re interested in this space, it’s worth reading their thesis in full when you have the time.)
In short, their work outlines how they foresee AI assistants and agents playing a role in five different buckets of transaction types, from routine ones like grocery shopping where AI agents may execute a purchase when you’re close to running out of an item and the price drops to your preferred level (a modern version of price tracker camelcamelcamel perhaps), to functional purchases like an oven, or a bike for commuting, where specialized AI consultants might guide you toward the best purchase in exchange for a small fee.
“If you’re buying a bike for $3,500 that you’re going to use every day,” Moore told me, “and you have this AI coach who walks through with you exactly what you might need, based on all of these specifications, maybe you’re willing to throw an extra $50 that way, versus if you’re buying a $50 sweater, you’re probably not going to be willing to pay the AI.”
Another consideration is attribution. Moore relayed the use case of high school and college-age girls who subscribe to ChatGPT Pro predominantly to help them identify the outfits (or dupes) worn by celebrities or influencers that they’ve seen on TikTok and that they want to purchase. OpenAI likely isn’t getting financial credit for its role in that transaction, because the old ways of tracking purchase credit through cookies and last-click attribution largely don’t work in this new world.
But that could be a big startup opportunity too, Rampell said.
“If you actually conduct your shopping in ChatGPT [in the future], then no [webpage] is being rendered potentially,” he said. “So there’s an opportunity to completely rewrite this—for a startup to do that, partially because you can’t be a combatant and Switzerland. And what I mean by that is, sure it would make sense for ChatGPT or Gemini to [tell brands or retailers to] use our attribution system, but they’re combatants too.”
Data sets focused on product specifications and product reviews could also serve as a new battleground for innovation and competition.
“We’ve actually started to see startups say, ‘I’m going to have an agent look for every video…hashtagged with this thing,” Moore said, “and transcribe that video, pull out the product-related information…and add it to their knowledge base so that’s their differentiated layer to pull from when recommending products.”
At a high level, Rampell said that the VC firm is taking a patient approach on the consumer side, “kind of waiting for something to really show traction, as opposed to us trying to outsmart the market.”
He allowed that an incumbent like Amazon could end up growing even faster due the power of its closed ecosystem and crowding out many would-be startup disruptors. (Some entrepreneurs believe the opposite.)
But on the enterprise side, Rampell believes, there’s tons of green room to pursue, from new payments infrastructure that supports AI agents making transactions across disparate websites and apps, to consolidating and enhancing the right data to make existing product catalogues discoverable in this new world.
“No matter how this takes off,” Rampell added, “there’s a lot of infrastructure that needs to change.”
We’ll see you next week,
Jason Del Rey
X: @delrey
Email: Jason.delrey@fortune.com
Submit a deal for the Term Sheet newsletter here.
Joey Abrams curated the deals section of today’s newsletter. Subscribe here.
