New offering bundles keyword and vector search into a single API
Agolia has officially launched NeuralSearch, an AI-based offering that combines both keyword and vector search capabilities in one API.
In a news release, Algolia said NeuralSearch can deliver highly accurate, relevant results in milliseconds on a scale involving “huge” production loads. It features advanced Large Learning Models (LLM), which Algolia describes as “the same technology underpinning ChatGPT and generative AI.” NeuralSearch automatically adapts search results when new products are added, new content is uploaded, or even when words and terms take on new meanings.
Algolia CEO Bernadette Nixon touted the API’s ease of deployment in the news release, stating that since the NeuralSearch is backward compatible, “zero engineering is required for customers to become AI-enabled” through its implementation.
Vector-based search analyzes the relationships, concepts, and context between words in a more natural, abstract way than keyword search. According to IDC research cited by Algolia, vector search gives retailers the ability to capture revenue they’ve traditionally lost due to unsuccessful ‘long-tail’ searches.
“Long-tail searches [are] less commonly used search terms that may not find exact keyword matches and return null results when queried. [They] represent lost revenue when they return null results to users, forcing those potential customers to abandon searches and take their business elsewhere,” Hayley Sutherland, Research Manager for Conversational AI and Intelligent Knowledge Discovery at IDC, wrote in her analysis.
Sutherland added that through vector search, retailers can “provide customers with similar or related products when an exact match is not found, allowing customers to find relevant results using free-form natural language and helping to ensure their revenue does not go to competitors.”
Algolia estimates that long-tail queries account for approximately 55 percent of all online searches.
Although vector-based search itself isn’t new, NeuralSearch features an innovation called Neural Hashing, which compresses search vectors from numbers with thousands of decimal points down to miniscule, static digits.
In an interview at the Shoptalk conference in March, Matthew Eisnor, Algolia’s Director of Global Alliances, told Composable.com this compression dramatically reduces the use and cost of computing resources.
“The problem with vector search is the amount of processing horsepower it requires. It’s very resource and infrastructure intensive. Neural Hashing compresses all of those vectors into a binary of ones and zeroes that is very efficient to process. That is the differentiator,” Eisnor said.
Neural Hashing and Neural Search were both developed by Search.io, an Australian startup acquired by Algolia in September 2022. The story behind the acquisition started with a series of serendipitous tweets that ended up having surprising, monumental consequences for both companies.
When Hamish Ogilvy posted a twitter thread on May 12, 2022, he had no idea where it would lead.
In the thread, the Co-Founder and CEO of Sydney-based Search.io described how his company’s vector-based Neuralsearch (that was the original spelling) helped an unnamed retailer boost conversions by five per cent and revenue by $1.5 million in one month of A/B testing.
“A huge number of search and AI experts around the world chimed in, asked questions, and contacted me, and one of them was the founder of Algolia,” Ogilvy recounted to The Australian Financial Review.
A deeper conversation between the two executives (and their companies) eventually ensued. Four short months after Ogilvy posted his Twitter thread, a deal for Algolia to acquire Search.io was all sewn up.
As Eisnor told Composable.com at Shoptalk, the transaction meshes Algolia’s established expertise in keyword search with Search.io’s new innovations in vector search.
“With both of those search mechanisms together, you have sort of a unified and complete search experience, because sometimes you want narrow and accurate (keyword queries) and sometimes you want broad and sort of exploratory (vector search),” Eisnor said. “What does that mean? It means customers are going to get a result, they’re not going to get a null result. They’re not going to get frustrated and go to your competitor.”
In his March interview, Eisnor said Algolia was testing NeuralSearch with 18 of its customers. Among the results from that pilot phase, fashion retailer Frasers Group saw conversions rise by 17 percent and ‘zero searches’—that is, failed queries generating no results—drop by more than 65 percent.
Algolia also released a no-code automation tool in March called Merchandising Studio. The company has doubled its total employee headcount over the past year, including a hiring spree in Q2 of 2022 that created 145 new positions.
Christine Wong
Senior Technology Staff Writer, Orium
I've been telling enterprise technology stories for almost three decades in print, online, and on television. I started out in journalism, covering the telecom boom, the birth of social media and the emergence of digital commerce. I'm always looking for the human angle in every technology story I write.