The Rise of Datumo: Disrupting Scale AI’s Stronghold  ## Metadata - Author: [[Hard Fork AI]] - Full Title: The Rise of Datumo: Disrupting Scale AI’s Stronghold - Category: #podcasts - URL: https://share.snipd.com/episode/d1d7e6a4-b264-4324-b3d5-b6e11af2bf74 ## Document Notes ## Highlights - **Labeling To Model Evaluation Pivot** - Daidumo expanded from annotation into model evaluation and safety benchmarking as customers requested output scoring and comparisons across models. - This vertical move leverages their labeling roots to deliver enterprise-focused model monitoring and evaluation tools. Transcript: Jaeden Schafer And this is basically where the company Dedumo comes in. They started as a data labeling company, and now they want to also help companies basically do like a safety benchmark for AI models so that you're able to monitor and improve your models. You're able to test what kind of responses come out. And that's kind of a new thing that they've been doing. They've been a data labeling similar to Scale.ai. That's kind of been their core product up until this point. So on Monday, they just announced another $15.5 million. They've raised about $28 million total. And so this round, their investors were Salesforce, they had KB Investments, ACVC Partners, and SBI Investments. There's a bunch of other people. But their CEO, his name's David Kim, and he is a former AI researcher at Korea's Agency for Defense Development. Basically, he was really frustrated. The reason he started the company is it just takes a really long time and it's very time consuming to label data for any sort of AI training. So he came up with an idea, which is basically he built this reward based app. I think this is super, super funny. But basically, it's a word based app, anyone can get on there. And you basically if you have free time, you can sit there and label data in your spare time, and you get paid for it. So I thought this was kind of funny. But it's obviously a huge business. It needs a lot of humans to do this. And so they basically crowdsource this data labeling on this app. So this is kind of cool. They did like a startup competition that they won, yada, yada. But I think overall, great idea. Before the app was fully built, they actually had 10s of thousands of dollars in pre contract sales during their customer discovery phase. So obviously, they're going to talk to customers like, hey, would you use this? A ton of people are like, yes. And they are, you know, signing pre contracts. So basically, they're like, all right, this is a great idea. In their first year, they actually passed a million dollars in revenue. So a ton of people wanted this well labeled data. Today, they have a whole bunch of really big companies that use them mostly it's Korean companies. So they have Samsung, LG Electronics, LG CNC, Hyundai, Navr. There's, you know, the Seoul-based telecom giant, which is SK Telecom. So a lot of big companies in Korea are basically their main clientele. And I imagine they'll try to push to expand a little bit more to America in the coming years. Over the last couple of years, their clients also started asking them for other things other than just data labeling. I talked about this in the intro, but basically they're going to start doing, they're going to start helping companies to benchmark models. Right now, they're seven years old. They have over 300 clients and they generated about $6 million in revenue last year. So growing really well. Here's a quote from one of the co-founders, Michael Huang. He said, and basically this is talking about the new service that they're going to be offering, which is kind of this benchmark thing. He said, they wanted us to score their AI model outputs to compare them to other models. That's when we realized we were already doing model evaluation without even knowing it. ([Time 0:02:19](https://share.snipd.com/snip/40b9a6f3-aeac-4c22-a73e-a15ac9e72df5)) - **Crowdsourced Labeling Launched Growth** - Daidumo began as a data-labeling startup using a reward-based crowdsourcing app to pay people to label data in their spare time. - The founders secured pre-contract sales and passed $1M revenue in year one, proving product-market fit. Transcript: Jaeden Schafer But their CEO, his name's David Kim, and he is a former AI researcher at Korea's Agency for Defense Development. Basically, he was really frustrated. The reason he started the company is it just takes a really long time and it's very time consuming to label data for any sort of AI training. So he came up with an idea, which is basically he built this reward based app. I think this is super, super funny. But basically, it's a word based app, anyone can get on there. And you basically if you have free time, you can sit there and label data in your spare time, and you get paid for it. So I thought this was kind of funny. But it's obviously a huge business. It needs a lot of humans to do this. And so they basically crowdsource this data labeling on this app. So this is kind of cool. They did like a startup competition that they won, yada, yada. But I think overall, great idea. Before the app was fully built, they actually had 10s of thousands of dollars in pre contract sales during their customer discovery phase. So obviously, they're going to talk to customers like, hey, would you use this? A ton of people are like, yes. And they are, you know, signing pre contracts. So basically, they're like, all right, this is a great idea. In their first year, they actually passed a million dollars in revenue. ([Time 0:03:02](https://share.snipd.com/snip/98b1d36d-4335-41c7-8df0-82a4f41d4852)) - **Licensed Book Data Boosts Reasoning** - Daidumo licenses curated datasets, including cleaned book text, which they claim improves models' structured human reasoning. - Owning licensed datasets differentiates them from pure service labelers and adds recurring product revenue potential. Transcript: Jaeden Schafer I think basically the drama in the industry is that Meta just spent $14.3 billion. They did this kind of acquisition investment into Scale AI. I think they bought like, they got like 50% the company i mean honestly everyone like says like oh my gosh meta bought scale ai but it's kind of exact same thing that microsoft did with Open ai for 10 billion dollars to get 50 of the company so or 50 of one of the shell companies that owns shares in the blah blah blah i know it's all convoluted for open ai but in any case i don't See it as being super far super far off. But when this happened, a lot of the big customers of scale AI pulled out open AI being one of them opening actually stopped using scale AI after the whole meta deal, because they're like, Well, metas are competitor. And you know, how can we trust it, all this stuff, right? So there's a whole bunch of similarities between these companies between Dadoomo and scale AI, I think basically they have this like pre-trained data that they're labeling. One thing I do think that's interesting that they're doing, it differentiates themselves, but I think this is a great business model for a lot of these types of companies today is that They actually have some licensed data sets. So beyond just labeling data sets for companies, they have their own licensed data sets that they'll also license to people. One of theirs that's really interesting is it is a whole bunch of data crawled from published books. And the company actually says that it's a really good, they say a rich structured human reasoning. So they said it's notoriously difficult to clean this kind of data, but it's actually it helps with the reasoning. ([Time 0:05:31](https://share.snipd.com/snip/8643a39d-1b19-4ece-96e7-d156dc52ca3e)) - **Meta-Scale Deal Opened Market Opportunity** - After Meta's big investment in Scale AI, customers worried about competitor entanglements and looked for alternatives, creating openings for challengers like Daidumo. - Daidumo targets enterprise safety, transparency, and global expansion to capture customers fleeing concentrated vendor risk. Transcript: Jaeden Schafer I think basically the drama in the industry is that Meta just spent $14.3 billion. They did this kind of acquisition investment into Scale AI. I think they bought like, they got like 50% the company i mean honestly everyone like says like oh my gosh meta bought scale ai but it's kind of exact same thing that microsoft did with Open ai for 10 billion dollars to get 50 of the company so or 50 of one of the shell companies that owns shares in the blah blah blah i know it's all convoluted for open ai but in any case i don't See it as being super far super far off. But when this happened, a lot of the big customers of scale AI pulled out open AI being one of them opening actually stopped using scale AI after the whole meta deal, because they're like, Well, metas are competitor. And you know, how can we trust it, all this stuff, right? So there's a whole bunch of similarities between these companies between Dadoomo and scale AI, I think basically they have this like pre-trained data that they're labeling. One thing I do think that's interesting that they're doing, it differentiates themselves, but I think this is a great business model for a lot of these types of companies today is that They actually have some licensed data sets. So beyond just labeling data sets for companies, they have their own licensed data sets that they'll also license to people. One of theirs that's really interesting is it is a whole bunch of data crawled from published books. And the company actually says that it's a really good, they say a rich structured human reasoning. So they said it's notoriously difficult to clean this kind of data, but it's actually it helps with the reasoning. So all these reasoning models, apparently reading books is a good way for them to like reason through, learn how to reason through problems, which I thought was absolutely fascinating. Unlike other companies, they have a full stack evaluation platform. So that's their data eval, which they basically have recently launched. One of their main products is kind of a no code evaluation tool it's for non-developers so people that are on policy trust safety compliance teams will use this you don't have to be a developer Really know how any of that works i thought that the uh i thought the story of how they actually got their the most recent investor which was salesforce ventures was interesting their Ceo that they previously hosted a fireside chat with Andrew McGee, who's the founder of deeplearning.ai. So obviously very famous person in the AI space. And this was hosted at an event in South Korea. But afterwards, they shared that event on LinkedIn. And from that, someone at Salesforce Ventures actually saw it. And then they said they had a whole bunch of Zoom calls and they got like a soft commitment. But the whole funding process took about eight months to actually roll out. And then they were able to close that. But I thought it was so interesting. It's like hosting some big famous person and then posting it on LinkedIn is like the best way for a startup to raise money. Thought that was very interesting. I may have to take a page out of their book for my startup AI box. This new round of funding that they're going to be using is basically used to accelerate their R&D. Specifically, they said in developing automated evaluation tools, they're doing this for enterprise AI and also to help them scale their global go-to strategies. We'll talk about a little earlier, they're very focused right now on Korea. I think that they say they want to expand to Japan and the US. They have about 150 employees in Korea. And they also have a presence in Silicon Valley. I think they might have opened like a little office or hired some people there in March. So starting to expand. Very interesting. I'm really excited about this company, to be honest. Very bullish on the data labeling side of things. Love kind of the journey where they're coming from and the industry they're going into, right? Scale AI taking on $14 billion. This is obviously a very big industry. So excited to see what they do with it and how they're able to grow. ([Time 0:05:31](https://share.snipd.com/snip/e19afc7a-849a-4254-b36b-c38a9886032d)) - **LinkedIn Post Turned Into VC Lead** - Daidumo attracted Salesforce Ventures after posting a fireside chat with Andrew Ng on LinkedIn, which led to Zoom calls and an eventual investment. - The CEO described the funding process as taking about eight months from initial contact to close. Transcript: Jaeden Schafer So obviously very famous person in the AI space. And this was hosted at an event in South Korea. But afterwards, they shared that event on LinkedIn. And from that, someone at Salesforce Ventures actually saw it. And then they said they had a whole bunch of Zoom calls and they got like a soft commitment. But the whole funding process took about eight months to actually roll out. And then they were able to close that. ([Time 0:07:50](https://share.snipd.com/snip/7113f66c-9ffb-4fa5-b02b-33748b826946)) - **Invest Funding Into Evaluation & Global GTM** - Use new funding to accelerate R&D in automated evaluation tools and to scale global go-to-market execution for enterprise AI products. - Prioritize expanding from a strong home market into Japan and the US while building productized, no-code evaluation features. Transcript: Jaeden Schafer This new round of funding that they're going to be using is basically used to accelerate their R&D. Specifically, they said in developing automated evaluation tools, they're doing this for enterprise AI and also to help them scale their global go-to strategies. We'll talk about a little earlier, they're very focused right now on Korea. I think that they say they want to expand to Japan and the US. ([Time 0:08:26](https://share.snipd.com/snip/bbc0d30b-c833-473a-953e-82c78a74a99a))