Datumo Raises $15.5 Million to Develop Tools for Building Safer AI and Improve Model Evaluation
Generative AI safety concerns pervade most organizations, according to a recent McKinsey report, with explainability – understanding AI decision-making processes – being a significant concern. While 40% of respondents view this as a major risk, only 17% are actively addressing it, the report states.
To help businesses build safer AI, Seoul-based Datumo offers tools and data for testing, monitoring, and improving models, requiring minimal technical expertise. The startup recently secured $15.5 million in funding from investors such as Salesforce Ventures, KB Investment, ACVC Partners, SBI Investment, and others, bringing their total raised to approximately $28 million.
Founded by David Kim, a former AI researcher at Korea’s Agency for Defence Development, Datumo began as an AI data labeling company. Kim’s frustration with the time-consuming nature of data labeling led him to develop a reward-based app that allows anyone to label data in their free time and earn money. The startup won validation for this idea at a startup competition hosted by KAIST (Korea Advanced Institute of Science and Technology). Kim co-founded Datumo, originally known as SelectStar, alongside five KAIST alumni in 2018.
Before the app’s completion, Datumo secured tens of thousands of dollars in pre-contract sales during the customer discovery phase of the competition, primarily from businesses and startups led by KAIST alumni. In its first year, the startup exceeded $1 million in revenue and secured key contracts with major Korean companies like Samsung, LG Electronics, Naver, and SK Telecom.
Clients soon began requesting more than simple data labeling from Datumo, which now offers services such as AI model output scoring and comparison. The company released Korea’s first benchmark dataset focused on AI trust and safety in response to these needs.
As the Large Language Model (LLM) ecosystem matured, Datumo expanded into pretraining datasets and evaluation. Meta’s recent significant investment in data-labeling company Scale AI underscores the importance of this market, with AI model maker OpenAI subsequently ceasing to use Scale AI’s services after the deal. The Meta deal also signifies an increasing competition for AI training data.
Datumo shares similarities with companies like Scale AI in pretraining dataset provisioning and with Galileo and Arize AI in AI evaluation and monitoring. However, it distinguishes itself through its licensed datasets, particularly data crawled from published books, offering rich structured human reasoning but known for being difficult to clean, according to CEO Kim.
Datumo also offers a full-stack evaluation platform called Datumo Eval, which automatically generates test data and evaluations to check for unsafe, biased or incorrect responses without the need for manual scripting. The signature product is a no-code evaluation tool designed for non-developers like those on policy, trust and safety, and compliance teams.
Investors such as Salesforce Ventures were attracted to Datumo following a fireside chat with Andrew Ng, founder of DeepLearning.AI, at an event in South Korea. After sharing the session on LinkedIn, Salesforce Ventures expressed interest, leading to several meetings and Zoom calls culminating in a funding commitment. The funding process took about eight months.
The new funding will be used for accelerated Research & Development efforts, particularly in developing automated evaluation tools for enterprise AI, and to scale global go-to-market operations across South Korea, Japan, and the US. The startup also established a presence in Silicon Valley in March.
