Built This Week/AI by Industry

AI by Industry

AI in Retail and E-commerce

In retail, the AI on our show measures what shoppers do inside stores, runs orders across sales channels and makes product photos in minutes, while chat apps start to compete for the checkout.

By Jordan Metzner & Sam Nadler · From our episodes · Updated October 2026

Watch: Episode 21, Not your STANDARD Market from 10:18

Retail on Built This Week splits into two halves: the physical store and the online checkout. Angie Westbrook of Standard AI brings online-style metrics into brick-and-mortar stores through the security cameras already on the ceiling. Wiley Jones of Doss runs inventory and orders for brands selling across several channels. On the e-commerce side, we've used AI to make product photos for a tiny charity store, and Jordan, who spent years at Amazon, has weighed in as shopping starts to move into ChatGPT.

You know, most retailers that you talk to are still experiencing something like 85 to 90% of their sales are still happening inside the store.

— Angie Westbrook, Episode 21

How companies we talked to use AI in retail

Standard AI: "Google Analytics" for physical stores

That was Sam's phrase, and Angie took it. She said physical retail has been a black box: stores only see sales data, which is shaped by seasonality, traffic and weather and can arrive weeks late. Standard AI started in checkout-free stores, then moved to analytics, running computer vision on existing cameras so layout, merchandising and in-store media can be measured like pages on a website. (Episode 21, from 10:18)

It doesn't use facial recognition. Angie said the system labels about 26 points on each body, turning shoppers into digital stick figures it can tell apart but not identify. Because it follows where people move and look rather than only how long they stand still, it produces a visual engagement score, the in-store version of an online impression.

But I think it's, like, more than 85, 90% of the time a new product fails, and all you do is get the sales data that says it didn't sell.

— Angie Westbrook, Episode 21

Her rule for reading the score: lots of engagement without sales points to a product or pricing problem; no engagement points to placement, marketing or signage. Speed is the other gain. A typical store test splits about 40 stores into control and experiment groups for 12 to 16 weeks, and Angie said some customers reached statistical significance in three weeks with only a couple of stores.

Doss: one system for orders across sales channels

Wiley's demo was a fake coffee company, and Jordan spotted orders flowing in from Shopify. Wiley typed his name into a table of a million sales orders and got results in a couple hundred milliseconds. He then asked the built-in agent for the last 10 wholesale orders, and for how SKUs connect to the product catalog, sales channels and fulfillment. It answered from the system's own setup. For the brands he sells to, growth usually means opening new sales channels and launching new products. (Episode 44, from 05:23)

What we built

  • AI product images for a 3D-printed hoodie charm store: Sam and his 11-year-old daughter sell 3D-printed hoodie charms, with proceeds going to an orphanage. He photographs a charm and asks Nano Banana in Google AI Studio to add it to a yellow hoodie, then to put the hoodie on a girl of about 11 to 13 in specific pants and shoes. It takes minutes, and he rarely had to re-prompt. Oversized charms in the banner turned out to be a plus, because they make it obvious what's for sale.
  • AI image editing chatbot: Jordan pitched it at small businesses and e-commerce brands that constantly need product shots. Chained edits kept his face consistent through a new hat, shirt and ranch background, but the product placement test, holding a beer, didn't fully work.
  • AI retail DJ: built for Angie's episode, it picks the store soundtrack by how many shoppers are inside. Angie offered to test in one of Standard AI's stores whether it changes browsing. The music side is on our AI in music page.

What's still hard

  • Sales data lags. Transaction logs can take weeks, and Angie said you can't see a shopper's instant reaction in sales data alone.
  • In-store tests are noisy. Physical retail is about as dynamic as it gets, and Angie said people analyzing traditional tests often "have to squint" to tie a lift to a decision. A bad change can also hurt test-store customers for the whole 16 weeks.
  • Privacy. Angie said some vendors use facial recognition; Standard AI treats being private as a core principle.
  • Who owns the checkout. When OpenAI said it would take a cut of shopping sales, Jordan said it could hit Amazon ads, Google Shopping and Shopify, since a store might only need a landing page ChatGPT can check out from. When Walmart started selling inside ChatGPT, he called chat the new search interface. Will shoppers actually buy there? "Maybe. Maybe not."
  • AI images aren't finished. Jordan expects quality to slip over many chained edits, an image of an image of an image.

Episodes on AI in retail

FAQ

How is AI used in retail?

Standard AI, whose CEO Angie Westbrook joined us in Episode 21, runs computer vision on existing store security cameras to measure how shoppers move and engage, so stores can test changes like a website does. We also used Nano Banana to make product photos for a small online store.

How does AI measure shoppers without facial recognition?

Angie said Standard AI labels about 26 points on each body and turns every shopper into a digital stick figure. The system can tell people apart and follow where they move and look, but it can't identify who they are.

How fast can a retailer A/B test a store change with AI?

Angie said a typical in-store test splits about 40 stores into control and experiment groups for 12 to 16 weeks. Some Standard AI customers reached statistical significance within three weeks, in some cases using only a couple of stores.

Will people shop inside ChatGPT?

Jordan wasn't sure. Discussing Walmart selling inside ChatGPT in October 2025, he said chat is becoming the new search interface and Amazon may end up paying for traffic again, but whether shoppers will actually buy there is still open.