Have you tried going into ChatGPT and asking for a product recommendation? Like “What are the best shirts I should get for X”?

Spoiler: it’s not that great.

In a world of AI shoppers, product info needs to be easily accessible to agents so that AI can make the best product decisions. ReFiBuy is building for that reality.

In other news…

Queue the drumroll

Announcing our Tech Week lineup! We’re back with not one, not two… but six SF Tech Week events this year.

Our recommendation: join us for a full day of programming on Thursday, October 8th, where we kick off the day with three content-packed firesides, followed by an evening festival with a celebrity DJ performance. 👀

Full list below. We’ll be putting out a more official announcement later this week. Enjoy this week’s deep dive into ReFiBuy!

Ethan and Arek 🦄

To an AI shopping assistant, a product it doesn’t understand might as well not exist. ReFiBuy makes sure the right product gets found. The company transforms the traditional product catalog used by online retailers into “conversational product information,” data structured and written so AI shopping agents can understand it and answer customer questions accurately. The company offers two core products to brands, retailers, agencies, and commerce partners. The first, a Commerce Intelligence Engine, continuously manages brand awareness and data integrity across engines like ChatGPT, Gemini, Claude, and Perplexity. The second is an optimization module built for AI shopping assistants like Amazon’s Alexa for Shopping and Walmart’s Sparky. The company's goal is to rewrite the way people Research, Find, and Buy products.

Check it out: refibuy.ai

ReFiBuy sells subscription access to its platform. Pricing is based on the total size of a customer’s catalog or SKU count, which is the number of distinct products that ReFiBuy optimizes for AI shopping agents and intelligence engines. The optimization for AI shopping assistants can be purchased as a separate module for qualifying clients.

In 2001, when Amazon was still mostly known as a bookstore, Scot Wingo founded ChannelAdvisor to help match a seller’s product listing to the correct entry in a marketplace’s catalog. And this wasn’t Wingo’s first attempt at solving this problem. It was his third. 

His previous company, AuctionRover, crawled hundreds of online auctions at once just to help shoppers find a needle in the haystack. After GoTo.com bought it in 2000, Wingo ended up building the opposite half of the same puzzle, automating how sellers listed products across those same fragmented marketplaces. When GoTo introduced that system to the market in 2001, it became ChannelAdvisor. The company was taken public in 2013, and then sold to CommerceHub in 2022 for $663 million. Wingo walked away from ChannelAdvisor with a payout, but not with an answer to the problem the company was built to solve.

For sellers, the problem was that mismatched listings would lead to their offers getting buried under the wrong catalog entry or lost altogether. For shoppers, what they would see on the page and eventually end up buying would not match what ended up at their door. This would generate a loss of trust in the platform on all sides. 

When Amazon introduced its own way of identifying products, it grouped every seller offering the same item into a single listing. Figuring out which listing a given SKU actually belonged to became a multivariable puzzle nobody ever fully cracked. Per Wingo, the best coding teams could only reduce errors by 50 percent, with no full solution in sight. 

The idea for ReFiBuy took shape in the fall of 2024. It was clear to Wingo and his co-founders that LLMs could finally take on the kind of judgement-heavy work required to help online shoppers find what they’re actually looking for, making it the solution to Wingo's long-standing problem. He co-founded the company with fellow ChannelAdvisor alumni Cameron Bowe, James Frawley, and Derek Conlin, betting that 25 years of marketplace experience would matter more than being an AI-native newcomer. 

For 30 years, e-commerce trained retailers to talk like machines. Every product was distilled down to four or five searchable attributes and handed to Google. That used to work when shoppers typed descriptors into a search bar and clicked whichever link looked right. It doesn’t work when shoppers are feeding 20-word prompts into ChatGPT or Claude, describing what they want in their own words instead of attempting a reverse lookup using keywords. 

Discovery through these engines is nonlinear in a way keyword search never was. A shopper might open a chat to plan a trip to Iceland and, four turns later, land inside a highly specific hiking-boot purchase funnel without ever typing the word "boots." But to get from customer conversation to the right product, the product data itself has to catch up. Answer engines expect something closer to a full product brief, listing material, country of origin, care instructions, organic certification. They also expect what Google calls conversational attributes, the harder-to-systematize details like which occasion a product suits or whether it performs in a humid climate. It's the kind of detail a good salesperson would offer on the floor, information no product feed has historically bothered to record.

ReFiBuy’s job is to make sure a brand’s product actually shows up and is described correctly at all different moments an AI agent is helping a user decide what to buy.

Wingo rejects how the industry currently measures visibility. The prevailing assumption is that basic answer engine optimization, simply getting mentioned favorably by ChatGPT, is sufficient. He believes that doesn't hold up. Brand mentions increase visibility, but that doesn’t necessarily increase revenue. "Getting cited doesn't ring the cash register," is roughly how he puts it. Visibility without a resulting sale is a vanity metric. 

ReFiBuy operates a level below generic Generative Engine Optimization, replacing it with Agentic Commerce Optimization. It continuously identifies missing product information and publishes it, then runs thousands of prompts a day, anywhere between a shopper’s first vague question and the purchase itself, against the answer engines to verify whether the right product is actually surfacing.

The market data backs up the urgency. Weekly active users on Walmart's Sparky, its generative AI shopping assistant, have more than doubled quarter over quarter, according to Walmart's own earnings disclosures, with average order values roughly 35 percent higher among users and purchase volume more than quadrupling. Amazon folded its Rufus chatbot into a unified Alexa for Shopping assistant in May 2026. Industry estimates put agentic commerce at 20 to 30 percent of total e-commerce volume by 2030. 

Wingo and his team already built the company that won the last shift in e-commerce. Now he wants to build the one that wins the next, and his case rests on 25 years inside the marketplace itself. ChannelAdvisor now runs under private equity as Rithum. While still profitable, it has no incentive to rebuild around agents. But keyword jail is emptying for online shopping. ReFiBuy is staking its future on the idea that only the person willing to make his own life's work obsolete can build what replaces it.