Datumo Raises $15.5M to Scale No-Code AI Safety Tools

Datumo raises $15.5M to commercialize no-code AI evaluation

A growing worry among organizations is clear: they don’t feel fully prepared to deploy generative AI safely. McKinsey finds 40% of respondents see explainability as a major risk, yet only 17% are tackling it actively. Seoul-based Datumo is betting there’s a market for simpler, enterprise-ready tools that let non-technical teams test, monitor and improve models.

On Monday Datumo announced a $15.5 million round, bringing total funding to roughly $28 million from investors such as Salesforce Ventures, KB Investment and SBI Investment. The startup started in 2018 as a crowdsourced labeling app built by KAIST alumni, where contributors earned rewards for annotating data. It quickly proved product-market fit, hitting $1M in revenue its first year and later signing major Korean enterprises.

Today Datumo counts more than 300 clients — including Samsung, LG, Hyundai, Naver and SK Telecom — and reported about $6 million in 2024 revenue. Customer requests pushed the company beyond raw labeling: clients wanted model scoring, output comparisons and safety checks. That customer feedback led Datumo to build evaluation capabilities and Korea’s first benchmark dataset focused on trust and safety.

Datumo’s differentiators are twofold: licensed, hard-to-clean datasets (including content crawled from published books that capture structured human reasoning) and a no-code evaluation platform, Datumo Eval, which automatically generates tests to flag unsafe, biased or incorrect outputs without manual scripting. That positions the startup somewhere between data providers like Scale AI and monitoring players such as Arize or Galileo.

The timing matters. Large players are doubling down on training-data supply — Meta’s multi-billion investment in Scale AI showed how strategic data access has become — and competition for high-quality, labeled, and curated datasets is intensifying. Datumo’s licensed content and automated eval stack aim to shorten the gap between model development and trust-and-safety sign-off.

Investors noticed not only the product but also Datumo’s community roots and credibility. A fireside chat with Andrew Ng and a well-timed LinkedIn share reportedly opened conversations with Salesforce Ventures, ultimately leading to funding. The new capital will accelerate R&D for automated enterprise evaluation and expand go-to-market across South Korea, Japan and the U.S.

What this means for enterprises is straightforward: expectations are shifting from ad hoc labelling projects to integrated evaluation pipelines that non-engineering teams can run. Policy, legal and trust-and-safety groups need accessible tools to validate model behavior, and Datumo’s no-code approach is an example of how the market is responding.

For organizations planning AI deployments, practical next steps include:

• Map which teams need no-code evaluation versus which need developer-level integrations.
• Select benchmark datasets that reveal real-world reasoning failures, not just synthetic edge cases.
• Automate recurring checks for bias, toxicity and hallucination and assign remediation workflows to non-technical reviewers.

Datumo’s trajectory — from crowdsourced labeling to licensed datasets and automated evaluation — shows a broader market trend: making AI safety accessible and measurable. As companies move from proof-of-concept to production, the ability to test models at scale and involve non-developer stakeholders will be a competitive advantage.

QuarkyByte’s approach is to blend practical governance with technical instrumentation, helping teams pick datasets, define benchmarks, and build evaluation pipelines that non-technical staff can run and act on. For enterprises wrestling with explainability and safety, solutions that bridge data, evaluation and operations are no longer optional — they’re operational necessities.