Most organizations admit that they are not fully prepared to use generative artificial intelligence in a responsible and safe way. According to a recent McKinsey ReportAI is a threat to human safety, according a McKinsey study. Explainability is a concern — understanding why AI makes certain choices. According to the report, 40% of respondents see it as a major risk, but only 17% actively address it.
Seoul-based Data started as a company that labels AI data. Now, it wants to help companies build safer AI by providing tools and data for testing, monitoring and improving models without the need for technical expertise. The startup raised $15.5m on Monday, bringing its total raised up to $28m. Investors included Salesforce Ventures and KB Investment.
David Kim was frustrated by the time consuming nature of data labels so he came with a new concept: a reward based app that allows anyone to label data and earn money in their spare time. The startup validated its idea at a competition held by KAIST (Korea Advanced Institute of Science and Technology). Kim co-founded SelectStar (formerly Datumo) in 2018 with five KAIST graduates.
Before the app was even fully developed, Datumo secured tens and thousands of dollars of pre-contract sales, mostly from KAIST alumni led businesses and startups, during the customer discovery stage of the competition.
The startup exceeded $1 million in revenue during its first year and secured several important contracts. The startup counts as clients major Korean companies such as Samsung, Samsung SDS and LG Electronics. It also counts Hyundai, Naver and the telecom giant SK Telecom. Clients began to ask the company several years ago to go beyond simple data labels. The startup, which has been around for seven years, now has over 300 clients in South Korea. It is expected to generate about $6 million of revenue by 2024.
Michael Hwang, cofounder of Datumo told TechCrunch that “they wanted us to score or compare their AI model outputs to other outputs.” Hwang said, “We realized that we were already doing AI model assessment without even realizing it. Datumo doubled-down on this area by releasing Korea’s first benchmark data focused on AI safety and trust.”
Kim told TechCrunch that “we started in data annotation and expanded into pretraining datasets, then evaluation as the LLM eco-system matured.”October 27-29, 2025
Meta’s recent $14.3 billion investment-like acquisition in data-labeling firm Scale AI highlights this market’s importance. OpenAI, a competitor of Meta and a maker of AI models, stopped using Scale AI services shortly after the deal. The Meta deal signals a growing competition for AI training data.
Datumo has some similarities to companies like Scale AI, which provides pretraining datasets. They also have some similarities to Arize AI and Galileo in terms of AI evaluation and monitoring. It differentiates itself, however, through its licensed datasets. Specifically, data crawled out of published books. According to CEO Kim, this data offers rich, structured human reasoning, but is notoriously hard to clean.
Datumo, unlike its peers, also offers a full stack evaluation platform called Kim said that Datumo Eval () automatically generates test results and evaluations in order to check for unsafe or biased responses, without the need for scripting. The signature product is an evaluation tool that does not require any code. It’s designed for non-developers, such as those in the policy, trust, safety, and compliance departments.
When Kim was asked about attracting investors such as Salesforce Ventures to the startup, he explained that they had previously hosted a fireside conversation with Andrew Ng founder of DeepLearning.AI at an event in South Korea. Kim shared the session with Salesforce Ventures on LinkedIn after the event. After several Zoom calls and meetings, the investors made a soft commitment. Hwang stated that the entire funding process took around eight months.
New funding will be used for R&D, especially in developing automated evaluation tool for enterprise AI and to scale global Go-to-Market operations across South Korea Japan and the U.S. The startup, with 150 employees in Seoul has also established a presence at Silicon Valley since March.
As we are always looking to improve, your feedback and insight into TechCrunch, our coverage, and events can help us. Fill out This surveywill let us know what we’re doing, and you can win a prize!
Kate Park, a reporter for TechCrunch with a focus in Asia on technology, startups, and venture capital, is a reporter. She was previously a financial reporter at Mergermarket, covering M&As, private equity and venture capitalism.
View Bio
