Logo
Home
About
Archive
Login
Join
  • Home
  • Posts
  • Inside the Arkansas factory trying to upend retail inertia--one AI-powered photo shoot at a time

Inside the Arkansas factory trying to upend retail inertia--one AI-powered photo shoot at a time

Backed by $300 million from Walmart, Eko is challenging the status quo. Plus, a new entrant in the Google/Shopify vs OpenAI protocol battle.

Jason Del Rey
Jason Del Rey

Good morning.

Before we get into today’s main feature, a quick victory lap if you’ll allow me.

I’ve seen some surprised reactions this week about Google and Shopify’s announcement that the tech council for the Universal Commerce Protocol (UCP), the set of rules they are co-developing to scale AI-powered shopping, has added a representative from Amazon.

For those of you who’ve been reading this newsletter for a while, you know that Amazon has stood out as the biggest shopping site to block external AI shopping agents like the very ones UCP is being created to support. So a level of surprise here is understandable (though Amazon CEO Andy Jassy did tell analysts on yesterday’s earnings call that his company continues to have discussions with the makers of all the big AI apps about possible ways to work together).

Yet paid subscribers of The Aisle could have seen Amazon’s involvement with UCP coming. That’s because back in early March, when I reported deeply on the UCP launch and rollout, I noted some Amazon involvement:

“GitHub activity shows at least one Amazon engineer reviewing proposals in the UCP repository, including suggested changes related to payment methods and product lookup, even if Amazon itself hasn’t been a partner.”

The same goes for some recent surprised reactions I saw when it became clear that ChatGPT sunset the dedicated “Shopping Research” AI model that it unveiled only in November. In fact, OpenAI’s commerce product leader had hinted at such a change in my interview with him a month earlier.

If you’re building, leading, or investing at the intersection of AI and commerce, this is the kind of signal paid subscribers tend to see first. Not a paid subscriber yet? You can join here. The monthly plan includes a free trial, but the annual option remains the best value.

Now on to the good stuff….

The Center Aisle

An Eko employee maneuvers Lego pieces for a product shoot. Photo courtesy of Eko.

For visitors to Bentonville, Arkansas, the top attractions often include Walmart’s new corporate campus, the stunning Crystal Bridges art museum and, at least for me, the standout brisket at Wright’s Barbecue. But one of the coolest attractions for techie retail nerds now sits about six miles southwest of the new Walmart HQ.

There, in an otherwise mundane industrial park, stands more than 100,000 square feet of AI-powered warehouse space that serial entrepreneur Ben Kaufman, along with the rest of the executive team at a company called Eko, is betting can dramatically improve online shopping in this new AI era of commerce.

Kaufman guided me on a tour of the facility this week to understand how it works and why it could be so disruptive to the way online shopping and selling has been done for the past 30 years. Here’s what I found.

First, Eko’s quick backstory, which is definitely a winding one.

The startup was founded in 2010 by Yoni Bloch, an Israeli musical artist and still the CEO, as a video technology company called Interlude. In its early years, it set out to disrupt the entertainment business with technology that enabled interactive video experiences, including choose-your-own-adventure-style customization for online viewers. That vision attracted strategic investors such as MGM Studios and Warner Music Group, along with venture firms including Intel Capital and Sequoia Capital’s former Israel-focused fund.

After a few pivots, Eko ultimately found product-market fit in a less flashy but more commercially viable area: the product catalogs of online shopping sites. Specifically, it focused on bringing greater detail and interactivity to product image galleries, which the company says leads to higher engagement and improved conversion rates among online shoppers. Along the way, it raised more than $300 million from Walmart, whose former CEO Doug McMillon developed a close relationship with Bloch.

Then last year, Eko hired Kaufman, the serial entrepreneur behind consumer companies like Mophie, Quirky, Homesick Candles, and the CAMP experiential toy store chain.

His first task was to expand what started as a one-stage test inside Walmart into a giant dedicated "capture factory"—this massive, highly specialized production facility for capturing product images, video, and data at scale. Besides Walmart, Eko also counts other household names like Best Buy and Procter & Gamble as clients.

Inside the capture factory

Eko’s facility spans more than 100,000 square feet, employs a couple of hundred workers, and runs close to 20 hours a day, seven days a week. Inside, it's organized into photo-shooting stages based on product size: small (roughly notebook-sized items), medium (think air fryers), and large stages for items that require human involvement, like apparel or bicycles. Another wing features all the inventory that’s already been shot, from basketball hoops to cooking grills to a Barbie cruise ship playset.

At the center of each of the product stages are what the team calls "Eko bots"—custom capture rigs with motorized turntables and camera systems that arc around the product, capturing a complete view from every angle. The underlying hardware was designed by Kaufman’s former colleague, Adam Paskow, and engineered by Re:Build Fikst, part of a manufacturing rollup co-founded by former top Amazon exec Jeff Wilke.

Photo credit: Me.

A worker scans a product barcode to pull up a predefined shot list. The system sets lighting and camera positions automatically, uses lasers to show where the item should be placed, and captures a full 360-degree view in about 90 seconds. Additional shots show real-world use, with employees appearing in certain shots to mount a TV bracket or sit on a chair to more accurately highlight the give of the seat’s cushion.

From product photos…to AI-ready data

Everything feeds into Eko’s software systems, which edit footage, correct color, stitch shots together, and extract structured data like weight, dimensions, and packaging text. In the end, each product is transformed into a computer-readable asset—what Kaufman calls “basically a fine-tuned AI model for each product.” At full capacity, the facility can process up to ~3,000 items per day, though larger or more complex items can reduce that number to just a few hundred.

As Kaufman and I toured the space, I couldn’t help but ask him what in hindsight might have been a tad rude:

“Does anyone ever come in here and say, ‘You’re doing all this for a freaking product catalog?’” As the query came out of my mouth, I realized I might have been the “anyone.”

“There isn’t a good product catalog in the world,” Kaufman deadpanned. “So I’m happy if Yes is the answer to their question.”

E-commerce product catalogs are notoriously inaccurate, he told me. With an Eko file, brands and retailers can upgrade product pages with 3D images and videos that better show how items look and perform, potentially boosting conversions and reducing returns. Packaging details and instructions are captured too, making it easier to launch accurate listings across new retail partners and replace outdated product data along the way.

The future

But I’ve known Kaufman for a long time and I know he wants more. He is a lifetime tinkerer and builder at heart. He said he’s worked longer hours in the last few months than maybe he ever has, dreaming up and then executing bold ideas in Claude Code. In his own words: “I’m usually the most obnoxious long-term thinker in the room.”

As a result, he, Bloch, and the rest of Eko’s leaders are indeed thinking longer term and working on more future-leaning innovations.

One is a system called Visual Response that allows shoppers to ask a question about a product and get a photo or video response that actually shows them the answer to the question, rather than just blocks of text. This technology could be integrated into a shopping site or perhaps within AI apps themselves. Eko has discussed possible partnerships with some of the large AI labs.

“Let AI show instead of tell,” Eko’s website says.

Kaufman and his team also believe the detailed product attributes captured by Eko’s AI-powered factory can give AI search engines the richer, more nuanced data they need to deliver more relevant results and recommendations. Today, SEO, AEO, and GEO firms are battling in this space, but Kaufman believes Eko’s datasets could prove even more valuable than current approaches.

“If you talk to these AI frontier platforms, they're like, the data is the problem [and] text-based responses to shopping queries aren't the answer,” Kaufman said. “We are the answer to all that. So I believe that we are an AEO firm in disguise. That’s just not our commercial model yet.”

Fighting inertia

Eko still faces a host of challenges.

For one, the model is not cheap. Kaufman said the startup spent tens of millions of dollars on this first facility and has already sold out capacity for the year, which can be either a good or bad problem depending on your point of view. While the company has not confirmed expansion plans yet, Kaufman says opening up another capture factory seems inevitable.

There’s also the question of how this private company can possibly keep up with the endless product shelves of mass retailers, which boast about hundreds of millions of items in their catalogues.

“I don’t believe retail catalogues are as big and vast as they are advertised to be,” Kaufman countered. “In fact, I believe the headline numbers you hear from major retailers are infected with the very problem that is holding back progress in agentic commerce: they are filled with legacy, out of stock, and ‘half-built’ item records,” or product pages that sellers started building but then abandoned.

Kaufman said you can look no further than Walmart Supercenters, which only stock somewhere between 120,000 and 150,000 items, for a more representative sample of the amount of products that matter most. Still, if Eko can get up to shooting 500,000 products a year, that such a level would suffice to cover the vast majority of online shopping trips.

Perhaps the most daunting hurdle, though, is just the inertia of retailers, brands, and the way they’ve built shopping sites for the last few decades. And don’t get Kaufman started on the growth of ads on retail sites.

“Retail media is clouding everyone’s understanding of how their websites are actually doing,” he said.

Kaufman said he, Bloch, and Eko’s board of directors have even wrestled with the idea of creating their own online shop to show the retail industry what kind of immersive, personalized, and data-rich, AI-readable shopping experiences are possible.

But the team doesn’t want to be seen as competing with their retail or brand clients, and are hoping to instead coax them to truly reimagine their online shopping experiences in ways previously impossible. And, no, Kaufman doesn’t believe that the current incarnations of text-heavy AI shopping assistants will lead them there.

“The terrible part is, I’m a consumer person,” Kaufman said. “I just want to show the commerce. But I have to talk in esoteric terms [when] all I wanna do is show that all this unlocks an online store that is fucking amazing. A store that people want to shop.”



© 2026 The Aisle.
Report abusePrivacy policyTerms of use
beehiivPowered by beehiiv