Seoul-based Datumo secures $15.5M from Salesforce Ventures and others to enhance AI model evaluation and global expansion.
According to a recent McKinsey report, most organizations assert that they are not adequately equipped to implement generative AI responsibly and securely.
One area of concern is explainability, which refers to the ability to comprehend the rationale behind the decisions made by artificial intelligence. According to the report, only 17% of respondents actively address the significant risk, even though 40% regard it as such.
Seoul-based Datumo was founded as an AI data labeling company. It is now committed to assisting businesses in developing safer AI by providing them with the necessary tools and data to facilitate the testing, monitoring, and improving their models, without technical expertise.
The startup raised $15.5 million on Monday, bringing its total to approximately $28 million. Investors include Salesforce Ventures, KB Investment, ACVC Partners, and SBI Investment.
David Kim, CEO of Datumo and a former AI researcher at Korea’s Agency for Defense Development, was dissatisfied with the time-consuming data labeling process.
Consequently, he devised a novel concept: a reward-based application that enables individuals to label data in their free time and earn money. The startup validated the idea during a competition at KAIST (Korea Advanced Institute of Science and Technology).
In 2018, Kim and five other KAIST alums co-founded Datumo, which was previously known as SelectStar.
Datumo secured tens of thousands of dollars in pre-contract sales during the customer discovery phase of the competition, primarily from KAIST alumni-led businesses and startups, even before the app was built entirely.
The venture achieved over $1 million in revenue and secured numerous significant contracts during its inaugural year. Samsung, Samsung SDS, LG Electronics, LG CNS, Hyundai, Naver, and Seoul-based SK Telecom are among the leading Korean companies that the startup currently serves as its clients.
However, clients began to request that the company expand its capabilities beyond basic data labeling several years ago. In 2024, the venture has been operational for seven years, generated approximately $6 million in revenue, and serves over 300 clients in South Korea.
Michael Hwang, co-founder of Datumo, told TechCrunch, “They requested that we evaluate their AI model outputs or compare them to other outputs.” “That was when we realized we were conducting AI model evaluations without our knowledge,” Hwang emphasized that Datumo redoubled its efforts in this field and released Korea’s inaugural benchmark dataset that concentrated on AI trust and safety.
Kim told TechCrunch, “We initiated our efforts with data annotation and subsequently expanded to pretraining datasets and evaluation as the LLM ecosystem matured.”
The significance of this market is underscored by Meta’s recent $14.3 billion acquisition-like investment in data-labeling company Scale AI. Following that agreement, OpenAI, an AI model creator and competitor of Meta, discontinued using Scale AI’s services.
The Meta agreement also indicates that the competition for AI training data is becoming more intense.
Datumo has similarities with companies such as Scale AI in the provisioning of pretraining datasets and with Galileo and Arize AI in evaluating and monitoring AI.
However, the company distinguishes itself by utilizing licensed datasets, particularly data from published books. According to CEO Kim, these datasets provide a wealth of structured human reasoning but are notoriously challenging to clean.
Kim also noted that Datumo provides a full-stack evaluation platform called Datumo Eval, which generates test data and evaluations automatically to detect hazardous, biased, or incorrect responses without the necessity of manual scripting, in contrast to its peers.
The signature product is a no-code evaluation tool designed explicitly for non-developers in the policy, trust, safety, and compliance teams.
Kim elucidated that the startup had previously hosted a fireside chat with Andrew Ng, the founder of DeepLearning.AI, at an event in South Korea, when queried about attracting investors such as Salesforce Ventures.
Kim shared the session on LinkedIn following the event, which piqued Salesforce Ventures’ interest. The investors provided a gentle commitment after numerous Zoom calls and meetings. According to Hwang, the funding procedure was completed in approximately eight months.
The new funding will be utilized to expedite research and development initiatives, with a particular emphasis on the development of automated evaluation tools for enterprise AI and expand global go-to-market operations in the United States, Japan, and South Korea. In March, the venture, which has 150 employees in Seoul, also established a presence in Silicon Valley.
