Seoul's Datumo raises $15.5M for AI safety and evaluation tools
The rapid proliferation of generative artificial intelligence has brought with it a pressing challenge: ensuring these powerful models operate safely and responsibly. A recent McKinsey report highlighted this concern, revealing that while 40% of organizations view AI explainability—understanding how and why AI makes specific decisions—as a significant risk, a mere 17% are actively addressing it. Stepping into this critical gap is Seoul-based Datumo, a company that began its journey in AI data labeling and has now pivoted to offer tools and data for testing, monitoring, and improving AI models, all without requiring deep technical expertise. The startup recently announced a $15.5 million funding round, bringing its total capital raised to approximately $28 million from investors including Salesforce Ventures, KB Investment, and SBI Investment.
Datumo’s origin story traces back to its CEO, David Kim, a former AI researcher at Korea’s Agency for Defence Development. Frustrated by the laborious nature of data labeling, Kim conceived a novel solution: a reward-based mobile application that allowed individuals to label data in their spare time for compensation. This innovative concept was first validated at a startup competition held at the Korea Advanced Institute of Science and Technology (KAIST), where Kim, alongside five fellow KAIST alumni, co-founded Datumo (initially known as SelectStar) in 2018. Even before the application was fully developed, Datumo secured tens of thousands of dollars in pre-contract sales during the competition’s customer discovery phase, primarily from businesses and startups led by KAIST alumni.
The company quickly gained traction, surpassing $1 million in revenue within its first year and securing pivotal contracts with major South Korean conglomerates such as Samsung, Samsung SDS, LG Electronics, LG CNS, Hyundai, Naver, and telecom giant SK Telecom. With over 300 clients in South Korea, Datumo reported approximately $6 million in revenue in 2024. However, several years ago, client demand began to evolve beyond simple data labeling. “They wanted us to score their AI model outputs or compare them to other outputs,” explained Michael Hwang, Datumo’s co-founder. “That’s when we realized: we were already doing AI model evaluation — without even knowing it.” This realization prompted Datumo to double down on AI evaluation, leading to the release of Korea’s first benchmark dataset specifically focused on AI trust and safety. As Kim elaborated, “We started in data annotation, then expanded into pretraining datasets and evaluation as the large language model ecosystem matured.”
The burgeoning importance of AI training data and evaluation is underscored by recent industry moves, such as Meta’s significant $14.3 billion investment, akin to an acquisition, in data-labeling firm Scale AI. This deal not only highlighted the market’s value but also intensified competition, notably when Meta competitor OpenAI subsequently ceased using Scale AI’s services. Datumo navigates this competitive landscape with a distinctive approach. While sharing similarities with companies like Scale AI in pretraining dataset provisioning, and with AI evaluation and monitoring platforms such as Galileo and Arize AI, Datumo sets itself apart. It leverages unique licensed datasets, particularly data meticulously crawled from published books, which, according to CEO Kim, offers rich, structured human reasoning but is notoriously difficult to clean.
Furthermore, Datumo offers a comprehensive, full-stack evaluation platform called Datumo Eval. This platform automatically generates test data and evaluations to identify unsafe, biased, or incorrect AI responses without the need for manual scripting. Crucially, its signature product is a no-code evaluation tool, making it accessible for non-developers, including professionals in policy, trust and safety, and compliance teams. The journey to securing investment from Salesforce Ventures was notably strategic. Kim recounted how a LinkedIn post sharing a fireside chat with DeepLearning.AI founder Andrew Ng at a South Korean event caught the attention of Salesforce Ventures, eventually leading to multiple meetings and a soft commitment after an eight-month funding process.
The newly secured capital will be instrumental in accelerating Datumo’s research and development efforts, particularly in crafting advanced automated evaluation tools for enterprise AI solutions. It will also fuel the scaling of their global go-to-market operations across South Korea, Japan, and the United States. With a team of 150 employees in Seoul, Datumo has already established a presence in Silicon Valley since March, signaling its ambitions to become a global leader in AI trust and safety.