WHY THIS MATTERS IN BRIEF
If shopping moves to AI agents, the product data they read — not the storefront — decides what gets sold.
Matthew Griffin is the World’s #1 Futurist Keynote Speaker and Global Advisor for the G7 and Fortune 500, specialising in exponential disruption across 100 countries. Book a Keynote or Advisory Session — Join 1M+ followers on YouTube and explore his 15-book Codex of the Future series.
In a warehouse here bigger than two football fields, digital cameras rotate around vitamin bottles, strollers and washing-machine pods as manicurists, hand models and former theatre directors work to build what they hope is a digital catalogue for retail’s Artificial Intelligence (AI) future.
Eko, the Brooklyn firm that operates the facility, calls it a “capture factory.”
The goal: to improve the accuracy of online listings for millions of products at Walmart, Best Buy and other retailers and make them easily digestible by AI. It is a decidedly manual process. Hundreds of Eko employees work to shoot products from every angle on movie-studio-style stages, shifting lighting or buffing out fingerprints on metal surfaces as needed.
“The output of AI is only as good as the input, and there was no good input,” said Ben Kaufman, president of Eko, which counts Bentonville-based Walmart as an investor, as well as venture firms such as Sequoia Capital and Intel Capital.
At the heart of this effort is a fast-moving battle over how consumers will find and buy everything from toilet paper to furniture through AI Agentic Retail platforms such as OpenAI’s ChatGPT or Google’s Gemini. Both have cut deals with retailers such as Walmart and Target to offer shopping via their chatbots, while Amazon is building its own AI shopping tools.
The Future of Consumer Behaviours and Retail | KPMG, UK | Matthew Griffin | Futurist Keynote, by Futurist Keynote Speaker Matthew Griffin
With AI-driven shopping, “a lot of the specific features of that inventory become more important,” said Nathan Feather, an internet analyst at Morgan Stanley. A shoe’s width or a precise description of how it could be used becomes more necessary because AI chatbots tend to home in on those sorts of details to suggest products to shoppers, he said.
Retailers have long struggled to make traditional product listings accurate, in part because they largely outsource the task to their suppliers. And while AI-based shopping is still a niche practice, some retailers are giving their inventory descriptions another look to make sure their products are AI-ready if and when more shoppers arrive.
It is still unclear what AI shopping will look like, and there is inherent tension between retailers and AI companies as they figure that out. It hasn’t all gone smoothly. OpenAI recently backed away from an early version of its Instant Checkout feature in favour of directing users to the retailers to complete the purchase. That allows retailers to preserve their direct relationship with shoppers.
Several companies – including PayPal, Salsify, and retail AI upstart Verneek – are touting tools that aim to make a company’s product lists more accurate and AI-ready.
Eko – the company that runs the warehouse in Bentonville – started as an interactive video company but eventually zeroed in on interactive digital product pages and AI-ready product files, said Yoni Bloch, the company’s founder.
Walmart has invested more than $300 million in Eko since 2018, in large part through a joint venture created that year, according to people familiar with the situation. Eko’s revenue is tens of millions of dollars a year and growing, said a spokeswoman.
It is still unclear what AI shopping will look like, and there is inherent tension between retailers and AI companies as they figure that out. It hasn’t all gone smoothly. OpenAI recently backed away from an early version of its Instant Checkout feature in favour of directing users to the retailers to complete the purchase. That allows retailers to preserve their direct relationship with shoppers.
Several companies – including PayPal, Salsify, and retail AI upstart Verneek – are touting tools that aim to make a company’s product lists more accurate and AI-ready.
The factory process of creating an Eko file takes 10 minutes for something simple like a bottle of vitamins but half a day for a large refrigerator with lots of features, said Amy Long, who has worked at the Eko factory since August.
To get action shots, the refrigerator is hooked up to water and electricity, said Long, who has worked as a professional photographer for nearly two decades. It is a constant battle to remove fingerprints and glare from lights before each shot, she said. Live models display the features. “It can be hard to adjust a refrigerator shelf in a beautiful, natural way,” said Long.
In its Brooklyn headquarters, employees develop shot lists for each product, balancing how long it takes to capture video of every feature versus efficiency.
On a recent day, the team debated how to show a new Tide detergent tile’s ability to quickly dissipate in water. “We know that’s a pain point for similar competitor products – that it doesn’t dissolve well,” said Talia Halperin, head of merchandising for Eko. “We want to just show honestly how it happens.”
Why does AI shopping need a whole new product catalogue?
AI agents pick products from fine-grained details — a shoe’s width, exact materials, use cases — that messy retailer listings rarely capture, so firms like Eko are rebuilding catalogues so agents can read and recommend items reliably.
