Datumo in Seoul Raises $15.5M to Compete with Scale AI, Backed by Salesforce

A recent study from McKinsey indicates that numerous organizations feel ill-equipped to utilize generative AI in a safe and responsible manner. A primary concern identified is explainability—the ability to understand how and why AI makes particular decisions. Although 40% of respondents in the survey view this as a major risk, only 17% are actively working to mitigate it.

Datumo, a startup located in Seoul, originally specialized in AI data labeling but has since transitioned to helping companies develop safer AI through tools and datasets for testing, monitoring, and improving their models—no technical skills required. Recently, on Monday, the company raised $15.5 million in funding, bringing its total funding to approximately $28 million, with contributions from Salesforce Ventures, KB Investment, ACVC Partners, and SBI Investment, among others.

David Kim, the CEO of Datumo and a former AI researcher at Korea’s Agency for Defence Development, found the data labeling process to be overly time-consuming, prompting him to devise a reward-based application that allows individuals to label data during their free time while earning money. This concept received validation during a startup competition at KAIST (Korea Advanced Institute of Science and Technology). In 2018, Kim co-founded the company—initially named SelectStar—with five KAIST alumni.

Even before the app reached full development, Datumo managed to generate significant revenue in pre-contract sales during the customer discovery stage of the competition, primarily from businesses and startups founded by KAIST graduates.

In its first year, Datumo surpassed $1 million in revenue and secured several key contracts. Presently, the firm counts major South Korean companies among its clients, including Samsung, Samsung SDS, LG Electronics, LG CNS, Hyundai, Naver, and telecom behemoth SK Telecom. Several years ago, these clients began seeking services that extended beyond basic data labeling. After seven years of operation, Datumo now serves over 300 clients in South Korea, with an expected revenue of around $6 million in 2024.

“Clients asked us to evaluate their AI model outputs or compare them with other outputs,” said Michael Hwang, co-founder of Datumo, in an interview with TechCrunch. “That’s when we realized we were already conducting AI model evaluations—without fully understanding it.” Following this realization, Datumo sharpened its focus on this domain and released Korea’s first benchmark dataset centered on AI trust and safety, Hwang noted.

“We started with data annotation and then transitioned into pretraining datasets and evaluations as the LLM ecosystem evolved,” Kim told TechCrunch.

Meta’s recent acquisition of data-labeling company Scale AI for $14.3 billion emphasizes the importance of this market. Shortly after the deal, OpenAI, a competitor to Meta, ended its relationship with Scale AI, highlighting the growing competition for AI training data.

Datumo aligns with companies like Scale AI by offering pretraining datasets, while also standing beside firms like Galileo and Arize AI in monitoring and evaluation. Nevertheless, it distinguishes itself through its unique licensed datasets, including those obtained from published books, providing valuable structured human reasoning that is typically difficult to clean, as noted by CEO Kim.

Unlike its competitors, Datumo features a comprehensive evaluation platform called Datumo Eval. This tool automatically generates test data and evaluations to detect unsafe, biased, or incorrect responses without the need for manual scripting, according to Kim. The centerpiece of the offering is a no-code evaluation tool designed for non-developers, including those working in policy, trust and safety issues, and compliance.

Reflecting on attracting investors like Salesforce Ventures, Kim mentioned hosting a fireside chat with Andrew Ng, founder of DeepLearning.AI, at a South Korean event. Following the chat, he shared highlights from the session on LinkedIn, which piqued the interest of Salesforce Ventures. This initiated several meetings and Zoom discussions, ultimately leading to a soft commitment from the investors. Hwang noted that the entire funding process took roughly eight months.

The recently secured funding will be directed towards enhancing R&D efforts, particularly in developing automated evaluation tools for enterprise AI, and expanding market operations globally in South Korea, Japan, and the U.S. At present, the startup has 150 employees in Seoul and set up operations in Silicon Valley this past March.

We continuously seek to improve, and your feedback on TechCrunch and our coverage and events is invaluable. Please complete this survey to share your thoughts and enter for a chance to win a prize!