Last year, we reported that Angi’s former CEO and chief product officer Oisin Hanrahan had raised an $18 million seed round to found a new company, Keychain, dedicated to building an AI-powered web platform that could better connect consumer packaged goods (CPGs) retailers in the U.S. — namely, large grocery chains selling in-house brands — with the manufacturers who actually make and package the food items those retailers sell.
Now, a little more than two months later, the New York City-based Keychain has debuted its web platform to the world for the first time. And it’s a massively ambitious undertaking even in its 1.0 iteration, encompassing more than 24,000 CPG manufacturers and more than 760,000 different products, with an eye on transforming the sector.
Is Keychain the Airbnb of food manufacturing?
With Keychain, retailers can log into a clean, orderly web interface, view different manufacturers, read a description about them and identify their location, sort them based on the capabilities they provide (nut-free, organic, chips, crackers, etc.), see the costs associated with them and connect with them directly on the platform, even reserving manufacturing times, introducing a new level of clarity and convenience.
Retailers can also elect to have products manufactured through Keychain’s exclusive partners (who pay for the convenience of being featured, as part of Keychain’s revenue generation model) or get wholesale excess manufacturing capacity from companies not advertising their services on the open market (who are revealed to the retailer customer alone).
The AI Impact Tour – NYC
We’ll be in New York on February 29 in partnership with Microsoft to discuss how to balance risks and rewards of AI applications. Request an invite to the exclusive event below.
Keychain uses the latest in generative AI on the backend to organize and enable precise searches across this massive volume of data, designed to replace the heretofore largely analog system by which CPG retailers found manufacturers — calling around, attending trade shows, leafing through individual catalogs and PDFs, and tracking everything themselves in their own spreadsheets and systems.
Instead of all that, Keychain allows CPG retailers to simply log into the website once, and, if they are a large brand, an account will already be there waiting for them to set up with many of the products they already offer, and suggestions for who can manufacture them.
According to Hanrahan in a video call interview with VentureBeat, this system allows the users of Keychain — both retailers searching for manufacturers to make their goods, and manufacturers searching for customers — to more easily find one another and keep tabs on the overall market.
“If you’re a brand or retailer, you know who can make what, and you’re a manufacturer, you know which products you should be making,” Hanrahan told VentureBeat.
In a way, it’s similar to Airbnb, providing the software that links interested customers with those providing the capacity they need — except, instead of travelers seeking lodging, it caters to retail CPG brands (think Trader Joe’s or Target, as hypothetical examples) and the manufacturers who can service them.
It’s also similar to Hanrahan’s previous joint, Angi, formerly Angi’s List, which pairs homeowners and renters with workers who can help them complete household projects, repairs and maintenance.
More than $120 million in manufacturing capacity, digitized and organized by AI
Keychain’s rollout already contains $120 million in manufacturing capacity across more than 40 product categories.
This massive trove of data — collected by the Keychain team’s scraping and discussions with CPG manufacturers about the many different and unique steps of their processes — is all analyzed and organized by AI on the backend: three different machine learning models, to be specific.
“We have a computer vision model which analyzes [product] images and turns them into description and finds things in the images in terms of packaging,” Hanrahan said. “We have a language model which is assessing the ingredients and descriptions and names, and then we have a another language model which is assessing manufacturer capabilities and basically doing a match or fuzzy match between what manufacturers can do and the processes that [go into the goods.] So we have three primarily three different private models doing the work.”
Asked whether the models were based on existing off-the-shelf models from the likes of OpenAI, Anthropic, or other leading provider companies, or open source code such as Meta’s Llama 2 or Mistral’s Mixtral, Hanrahan declined to specify but did say the company was using a mix of open source and closed source models that had been fine-tuned and retrained — echoing an emerging trend at the forefront of enterprise that VentureBeat founder Matt Marshall previously wrote about here.
Key industry endorsements
Keychain has already won endorsements from Paul Voge, co-founder and CEO of Aura Bora, and Kelly McGoldrick, chief customer officer at Wyandot Snacks, who commended the platform for enabling the sharing of production process and capacity information — novel features in the industry.
Keychain plans to make its core platform available to select retailers and brands in 2024, currently limiting access to invited partners.
With AI capabilities and an extensive network of manufacturers, Keychain aims to simplify the process of establishing and managing manufacturing partnerships, enabling brands to bring products to market faster and more efficiently.
VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.