SAN FRANCISCO — Kumo, a graph machine learning-centered AI platform that allows anyone in an organization to harness the power of data to make faster, simpler, and smarter predictions, has raised $18 million in Series B funding led by Sequoia Capital. This latest round also includes participation and/or advisorship from existing and new investors including A Capital, SV Angel, Ron Conway, Michael Ovitz, Frank Slootman, Kevin Hartz, Clement Delangue, and Michael Stoppelman, among others.
Kumo plans to use the new funding to continue its hiring efforts, bring its leading AI technology to more companies, and invest in R&D efforts to expand its platform and services. On October 16, anyone can sign up for Kumo’s first release of the product via Kumo’s website.
“Kumo brings a new paradigm for predictive AI over cloud-based data powered by graph ML which we are thrilled to be in a position to introduce more broadly to the world,” said Vanja Josifovski, Co-Founder and CEO. “We are building a platform that is end-to-end, automating every step from ingesting data from source systems all the way to making predictions that can directly help businesses grow faster and operate more efficiently. Democratization of AI to all users regardless of machine learning (ML) experience has been the promise for a long time, but our unique approach, leveraging the inherent connectedness of your data, is the first to truly deliver on that promise. The Kumo product is also a huge win for CTOs and Chief Data Officers, allowing their colleagues in other parts of the business to harness the predictive power of AI by their teams directly rather than needing to constantly rely on the data team.”
The Kumo platform enables all non-technical and technical users, regardless of ML experience, to traverse all the major steps of a best practice ML lifecycle, with just three steps: (1) One-click ingestion of raw data tables from a wide variety of underlying source systems, (2) Creating a ‘Kumo Graph’ defining how different ingested tables connect to each other, and (3) Querying the future as easily as you query the past today in SQL, through its Predictive Querying language.
With this dramatically simplified workflow, users can immediately tackle a wide variety of predictive problems all in a single sitting, in application areas such as new customer acquisition, customer loyalty and retention, personalization and next best action, entity resolution and knowledge graph curation, abuse detection, financial crime detection, generation of ML features for data science teams, and more.
Under the hood, Kumo automates data preparation, feature engineering, neural architecture search, model evaluation, prediction-specific explainability, and deployment for predictions. By doing so, Kumo makes predictive tasks as easy as analytics tasks, thus revolutionizing enterprise AI just as data warehouses revolutionized analytics.
With a combined 50 plus years of experience in the AI and data field, Kumo’s founding team has seen firsthand the incredible power of graph learning for AI and business ROI — and also the massive effort to implement a single, production-quality predictive model due to cost and time. To tackle this opportunity, Kumo recently rolled out an early version of its product to a first wave of pilot enterprise customers, many of whom have already seen promising results across use cases for customer churn and LTV prediction, affinity modeling, personalization, and more.
“At Whatnot, AI plays a critical role in personalizing the shopper experience, driving cross-sell across categories and predicting future aggregate shopper behavior so we can shape our broader marketplace,” said Ludo Antonov, VP of Engineering at Whatnot. “To this end, we are working with Kumo to deliver a service that is truly ground-breaking, allowing us to not only quickly launch these needed predictions with their very simple predictive querying language and accompanying APIs, but also drive dramatic model quality gains, including a doubling of both precision and recall over existing baselines in initial experiments. We’ve been thrilled by the progress so far, and the ability of the Kumo product to allow even non-technical teams to harness the power of AI from our data in the future.”
The core graph ML technology that underpins Kumo’s product has been in development for the past five years through Stanford and Dortmund University AI labs and Pytorch Geometric open-source software, the world’s most widely used graph neural network open source framework. Kumo’s three founders, Josifovski (former Airbnb CTO Homes, Pinterest CTO, Google), Jure Leskovec (Stanford professor, former Pinterest Chief Scientist), and Hema Raghavan, (former executive at LinkedIn, IBM, Yahoo) saw a tremendous opportunity to take graph learning expertise from an academic setting and operationalize that research for a broad set of use cases in a more user-friendly SaaS product.