Search the way you think
Today, we are excited to announce our series A investment in Marqo — a platform that allows everyone to search the way they think.
The manner in which humans search for information or things online has remained non intuitive for decades. For example, to allow consumers to search for a single t-shirt on an e-com website requires the platform to tag that SKU with attributes such as ‘color’, ‘fit’, ‘material’, ‘size’, ‘brand’, ‘price’, ‘reviews’, ‘year of manufacturing’, and more, in a dedicated data catalog. Creating, labeling, and maintaining this catalog can be highly manual and cumbersome, especially for companies that have 10s of thousands of SKUs, with frequent additions and replacements. Moreover, edge cases such as spelling errors in catalogs, or mis-attributing a blue shirt as a green shirt, occur often and lead to poor recommendations and loss of revenue.
Over the last year or so, a new, natural way of searching for information has become popular globally thanks to platforms such as ChatGPT. Within months, over 100 million users got a taste of what it feels to search for ‘I have a baby-pink shirt, build me a matching wardrobe for an evening, outdoor, winter office party.’ This is the future of search — natural, conversational, and highly contextual. We at Lightspeed believe that to support such searches, platforms need to move away from a keyword-mapping-based approach to a semantic-based one. An approach in which the underlying algorithm understands the inherent meaning & context of a user’s query, irrespective of the medium (text, voice, image). This is where Marqo comes in.
First and foremost, as always, our investment thesis in Marqo was anchored on the team. To offer a vector-based search experience requires not only a deep understanding of LLMs/AI but also of traditional information retrieval (IR) techniques — indexing, filtering, retrieval, caching, and more. When we first met Tom & Jesse, we were immediately impressed with their background & technical capability. Jesse, a PhD in Physics, was the principal scientist & Stitchfix & subsequently a team leader at Amazon’s Robotics AI division. Tom had spent a few years as a core team member at AWS, in their RDS team. We immediately felt that this team had a deep understanding of how to leverage both the AI & and IR aspects of this challenge to create a new age AI-powered search system.
Second, it was the exceptionally well-thought out product/platform Tom & Jesse had built. While we had been studying the sharp rise in the number of libraries & vector database products that allowed engineers to build semantic search experiences, we recognized there were gaps. Existing solutions are difficult to implement in production (often taking months!) and require you to create, store & manage vector embeddings for each asset. Also, current set of AI-first solutions are incomplete and need to be supplemented with traditional relevance and ranking algorithms. And finally, there are limitations on searching with different data modalities e.g. voice, text, image.
Marqo provides a vector database, an orchestration layer, as well as a model-tuning service right out of the box. Consequently, engineering teams no longer need to worry about vector creation, storage and retrieval — the platform handles it all for you, while also providing the flexibility to choose across various embedding models & traditional ranking algorithms. Moreover, the platform syncs directly with an enterprise’s data store, ensuring that the vector index contains only fresh and relevant embeddings. In the absence of Marqo, these would all be mandatory engineering workflows that require a considerable amount of technical capability and bandwidth. From the outside the market seemed crowded, but we quickly realized that Marqo is a highly differentiated offering, and is best positioned to emerge as a global category leader.
And finally, Marqo was excelling both with its open-source bottom-up GTM as well as getting top-down enterprise traction this early in the game. We believe that to ensure a smooth transition from non-AI to AI-first backends, any product has to nail the value proposition to the developers through ease-of-use, devX and community, and also influence the buyers to open up budget lines.
Through Marqo, developers can embed multimodal and multilingual search into their applications through a single API and just 3 lines of code.
With Marqo powering the search experience, customer queries no longer need to be parsed in a highly rigid & confined manner. Intuitive search queries like ‘outfit for a summer wedding’ or ‘suggest a wedding outfit similar to the attached image’ would output highly relevant & personalized results!
The team has successfully onboarded several large enterprises, including a few who receive 500M+ monthly hits on their website! Many of these customers are seeing a 10–15% uplift in their add to cart rates after deploying Marqo. Interestingly, enterprises are also using their product to power various RAG based applications across customer support, enterprise search and sales/marketing related use cases.
We are off to a very strong start and are confident that Marqo will enable organizations to implement search experiences in a seamless way, and allow customers to search the way they think! You can check out the product here.
Authors
