Scot Wingo noticed customers’ procuring journeys whereas working ChannelAdvisor, {the marketplace} administration agency he began in 2001. He says customers method the method in three levels: researching the market, discovering appropriate merchandise, and shopping for the proper merchandise.
The acronym — ReFiBuy — is the title of his newest firm. It’s a generative AI optimization platform for retailers and types.
By any measure, Scot is an ecommerce pioneer. We first interviewed him in 2006, when he launched us to market promoting.
Final week, I requested him about ReFiBuy. Your entire audio of our dialog is embedded beneath. The transcript is edited for size and readability.
Kerry Murdock: Inform us about your ecommerce journey.
Scot Wingo: It started in 1999 after I launched Public sale Rover, an public sale search engine. We offered it to GoTo.com, which turned Overture, the corporate that invented paid search. The public sale search engine wasn’t nice after the dot-com bubble burst. However we had constructed the promoting instruments, which turned ChannelAdvisor, which I launched in 2001.
Murdock: Inform us about ReFiBuy, your new enterprise.
Wingo: The concept began with my expertise at ChannelAdvisor. The corporate went public in 2013, and I used to be nonetheless CEO and founder. By 2015, working a public firm had turn out to be a drag.
I left the CEO position however stayed on the board. ChannelAdvisor was finally taken non-public by a personal fairness agency, which merged it with Commerce Hub. It’s now known as Rithum.
So I left ChannelAdvisor in 2015 and launched an on-demand automobile care firm known as Spiffy. Then, in August of 2024, I made a decision to begin what’s now ReFiBuy. I wished to do one thing within the AI world. I’ve a technical diploma, and as a technologist, I assumed AI would create a lot disruption, which creates alternative.
So I used to be poking round, studying extra about it. After which, in December 2024, Anthropic, the makers of Claude, printed a paper on “agentic” AI that may carry out duties. Previous to that, giant language fashions have been read-only. The agentic element meant they may do issues.
And that jogged my memory of an issue we had at ChannelAdvisor. Our shoppers have been retailers and types with giant product catalogs. The problem for us was the absence of an business commonplace for electronically storing and sharing the product information, akin to specs, colours, dimensions, and weight.
Purchasers would ship us a file of their product catalog in a disorganized mess. But we had 100 marketplaces that wished to obtain stunning, clear catalog information. So our job turned catalog cleaners, to transform shoppers’ stock information right into a format acceptable to these exterior channels. Once more, there was no business commonplace.
We got here up with algorithms for cleansing the catalog that labored solely half the time. The opposite half required people. Finally, once we had 300 folks in Bulgaria engaged on it, serving our 3,000 clients and 15 billion annual transactions.
That reminiscence was my gentle bulb second for agentic AI. May we clear up the product catalog downside for LLMs? We began engaged on it late final 12 months.
Concurrently, Perplexity launched what we now name agentic commerce, or agentic procuring, the place you can’t solely analysis merchandise but in addition purchase them.
That’s the inspiration for our title. ReFiBuy is “analysis, discover, purchase.” It’s the consumer’s journey.
We launched our Commerce Intelligence Engine final week. It ensures that the LLMs — Perplexity, Claude, ChatGPT, and others — have correct, present, and complete product catalog information for our shoppers, that are retailers and types.
Murdock: How do you try this — set up the info after which make sure the LLMs digest it?
Wingo: We begin with the product catalog. We take a standard Google Procuring feed and even information from a service provider’s ecommerce web site. We analyze it via the lens of an LLM, which helps us establish lacking or incorrect elements. We then suggest adjustments, fixes, and additions. LLMs need every bit of content material that ties merchandise to the context of prompts. That features Schema.org markup, Reddit discussions, immediate historical past — way more than product information alone.
That’s our analysis part. Then we assist our shoppers whitelist the proper bots to crawl their websites. Most retailers and types block all bots aside from Google. Actually there are good causes to try this, as many bots are malicious or from opponents.
So we assist retailers know which LLM bots to permit.
Murdock: How have you learnt that an LLM receives and shops your optimized information?
Wingo: We monitor product playing cards, the visible representations by LLMs of really useful merchandise. We run 1000’s of prompts each day throughout all of the LLM engines to make sure our shoppers’ merchandise seem in these playing cards and that the info is correct.
Our AI brokers consider the playing cards and classify into buckets. If our shopper owns the product card, our job is finished. We’ve got achieved Nirvana for that SKU. If our shopper’s merchandise seems in a card of one other service provider, there are 20 to 30 issues which have seemingly gone fallacious. Our AI brokers detect it. Generally it’s so simple as a lacking slash or an additional area within the file.
The agent additionally detects lacking SKUs — when our shoppers’ items don’t seem within the playing cards in any respect. That’s normally attributable to an infrastructure downside with the crawler, or one thing is damaged on the service provider’s web site.
We maintain cranking the method till we’ve optimized our shoppers’ complete catalog.
Murdock: What’s the price of ReFiBuy?
Wingo: It relies on the variety of SKUs. We begin at roughly $2,000 per 30 days — $20,000 to $25,000 per 12 months.
Murdock: The place can retailers study extra?
Wingo: We’re at ReFiBuy.ai. My Substack publication is “Retailgentic.”
