Seoul AI Startup Datumo Secures $15.5M from Salesforce to Rival Scale AI

A recent McKinsey report reveals that most organizations feel underprepared to implement generative AI safely and responsibly. A key concern is explainability—understanding the rationale behind AI decisions. Although 40% of respondents see this as a major risk, only 17% are taking concrete steps to mitigate it.

Datumo, a Seoul-based startup that began as an AI data labeling firm, is now helping businesses develop safer AI. It provides tools and data for testing, monitoring, and refining AI models, designed to be accessible without deep technical knowledge. The company recently raised $15.5 million, bringing its total funding to around $28 million. Investors include Salesforce Ventures, KB Investment, ACVC Partners, and SBI Investment.

CEO David Kim, a former AI researcher at Korea's Agency for Defence Development, grew frustrated with the slow pace of data labeling. He conceived a reward-based app that lets people label data in their spare time for pay. The concept was validated at a KAIST startup competition, leading Kim and five KAIST alumni to co-found Datumo (formerly SelectStar) in 2018.

Even before the app's full launch, Datumo secured tens of thousands in pre-contract sales during the competition's customer discovery phase, primarily from KAIST alumni-led ventures.

In its first year, revenue exceeded $1 million with several key contracts. Current clients include major Korean corporations like Samsung, Samsung SDS, LG Electronics, LG CNS, Hyundai, Naver, and SK Telecom. As client demands evolved beyond basic labeling, the seven-year-old startup expanded its services. Today, it serves over 300 clients in South Korea and generated approximately $6 million in revenue in 2024.

"Clients began asking us to score their AI model outputs or compare them with alternatives," co-founder Michael Hwang told TechCrunch. "That's when we realized we were already performing AI model evaluation—we just didn't have a name for it." The company deepened its focus in this area, releasing Korea's first benchmark dataset for AI trust and safety, Hwang added.

"We started with data annotation, then moved into pretraining datasets and evaluation as the large language model ecosystem grew," Kim explained to TechCrunch.

Tech and VC heavyweights join the Disrupt 2025 agenda

Netflix, ElevenLabs, Wayve, Sequoia Capital, Elad Gil — just a few of the heavy hitters joining the Disrupt 2025 agenda. They’re here to deliver the insights that fuel startup growth and sharpen your edge. Don’t miss the 20th anniversary of TechCrunch Disrupt, and a chance to learn from the top voices in tech — grab your ticket now and save up to $600+ before prices rise.

Tech and VC heavyweights join the Disrupt 2025 agenda

Netflix, ElevenLabs, Wayve, Sequoia Capital — just a few of the heavy hitters joining the Disrupt 2025 agenda. They’re here to deliver the insights that fuel startup growth and sharpen your edge. Don’t miss the 20th anniversary of TechCrunch Disrupt, and a chance to learn from the top voices in tech — grab your ticket now and save up to $675 before prices rise.

Meta's recent $14.3 billion strategic investment in data-labeling firm Scale AI underscores the market's significance. Shortly after, AI model maker OpenAI, a Meta competitor, ended its use of Scale AI's services. This move highlights the growing competition for quality AI training data.

Datumo overlaps with companies like Scale AI in providing pretraining datasets, and with Galileo and Arize AI in AI evaluation and monitoring. Its key differentiator is its licensed datasets, which include content crawled from published books. CEO Kim notes this data offers rich, structured human reasoning but is challenging to clean.

Datumo also offers Datumo Eval, a full-stack evaluation platform that automatically generates test data and assessments to identify unsafe, biased, or incorrect AI responses without manual coding. This flagship no-code tool is built for non-developers, such as policy, trust and safety, and compliance teams.

Regarding attracting investors like Salesforce Ventures, Kim recounted hosting a fireside chat with DeepLearning.AI founder Andrew Ng in South Korea. After Kim shared the event on LinkedIn, Salesforce Ventures took notice. A series of meetings and Zoom calls led to a soft commitment, with the full funding process taking about eight months, according to Hwang.

The new capital will accelerate R&D, particularly for automated enterprise AI evaluation tools, and expand global go-to-market efforts in South Korea, Japan, and the U.S. The 150-person Seoul-based team established a Silicon Valley presence in March.