Web3 CMO Stories
Web3 CMO Stories is the podcast for marketing leaders and founders building in the next web, where blockchain, AI and immersive technology meet.
Every week, host Joeri Billast sits down with founders, CMOs and thought leaders to unpack how brands grow visibility and trust when technology rewrites the rules. No hype and no jargon: clear, strategic conversations in a language business leaders outside the tech bubble understand.
With more than 300 episodes since 2022, the show ranks in the top 5% of podcasts globally according to Listen Notes. Guests include Philip Kotler, Gary Vaynerchuk, Chris Do, Mark Schaefer, Joe Pulizzi, Andy Crestodina and Ben Goertzel of SingularityNET.
Joeri is a Fractional CMO, AI Visibility Strategist, TEDx speaker and author of The Future CMO, endorsed by Philip Kotler.
Sponsored by CoinDesk (2024), RYO (2025-2026) and Metricool (2026). Featured in Cryptopolitan.
"One of the sharpest marketing shows running right now." Matt Wilkinson, Founder of Strivenn
Articles and transcripts: The Future CMO Library, webdrie.net
About Joeri: joeribillast.com
Web3 CMO Stories
The Smarter AI Gets, the More Trusted Data Matters with Daniela Braga | S6 E49
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Everyone is obsessed with smarter AI models, but the uncomfortable truth is that the next wave of wins will come from something less flashy: trusted data. I’m joined by Daniela Braga, founder and CEO of Defined.ai, to unpack why ethically sourced training data, provenance, and governance are quickly becoming the real competitive advantage for any company building with generative AI and agentic AI.
We look back at what changed in the last few years, including the surprising explosion of frontier model competition and the rapid arrival of AI agents. From there, Daniela explains why the open web is no longer the differentiator and why closed source data and private datasets matter more than ever. We also tackle synthetic data: when it’s the only way to cover rare scenarios, when it introduces hidden bias, and how model collapse becomes a real risk as AI generated content floods public channels.
If you’re a CMO, founder, or business leader, this conversation makes the case that data is not just an engineering input. It shapes brand reputation, compliance posture, and product outcomes, especially in regulated spaces like healthcare, finance, legal workflows, robotics, and physical AI. Daniela shares what “trusted data” means in practice, why metadata quality is a silent failure point, and how serious teams evaluate models through testing, red teaming, and human in the loop review at scale.
We close on accountability and trust: who is responsible when autonomous systems cause harm, what responsible AI should include beyond slogans, and why credible sources and paid rights still matter for the future of knowledge. Subscribe for more Web3 CMO Stories, share this episode with a colleague, and leave a review if you want more conversations like this.
This episode was recorded through a Descript call on November 11, 2026.
Read the blog article and show notes here: https://webdrie.net/the-smarter-ai-gets-the-more-trusted-data-matters-with-daniela-braga/
..........................................................................
📘 Read The Future CMO: Amazon
👥 Join the Future CMO Community
🌿 Explore Sintra Synergies Retreat
💼 Connect on LinkedIn
..........................................................................
Welcome And Why Data Matters
Daniela BragaHumans are not going anywhere, even though everybody's worried about the future of work, and there's more and more, human curation on the data is required as we are moving towards, specialized, mechanics, electricians, type of data, surgeries, dental, and all sorts of beauty.
Joeri BillastHello everyone, and welcome to the Web3 CMO Stories podcast. My name is Joeri Billast. I'm your podcast host, today I'm excited to be joined by Daniela. Hey, Daniela, how are you?
Daniela BragaHi Joeri. It's a pleasure to be here again
Joeri BillastGreat to have you again on the podcast. not as planned, Daniela, because we were going to meet at the Web Summit in Media Village, but the nature had other plans today, so we will do this online. Guys, if you don't know Daniela, Daniela Braga, she's the founder and the CEO of defined.ai, the largest marketplace for ethically sourced training data for AI. She's a World Economic Forum Global Innovator, and she was appointed to, the NAI RR task force to advise the US president on AI strategy. Daniela, it's our second podcast episode. So the first time we spoke at the Web Summit three years ago, and we talked about, the power and the responsibility of generative AI. now looking
What Changed Since Three Years Ago
Joeri Billastback, what did you believe about AI then that you would change or qualify today?
Daniela BragaYeah. well, first three years ago, I was convinced that AI, would consolidate, or the AI builders would consolidate the efforts, given the enormous amount of, funding required to build frontier models... Well, they were not frontier back then. but all these, frontier models that we are-- that we see today, I thought the world was gonna co-consolidate, just like it consolidate when we were talking about, search engines and cloud providers. But for my shock, we have a lot more competition and, and good. that's a great thing. We have a lot more competition. I think the open source world has done a lot for this. we have geopolitical changes. I mean, we have other powers in the world. the emer-emerging, I mean, not emerging, they've been there. We just don't know about them so much. In China, in the Middle East, we have, companies like Humane and, Humane is one of these, king- powered by the Kingdom and funded by the Kingdom companies. G42, which was already there, but still very, humble. So we see-- the only place we don't see, frontier model labs, emerging is Europe, except Mistral, which still not enough, but good, good, good. But we have Canada with Cohere. We have a lot in the United States, and we don't have Latin America, Africa, or Europe really well represented. But so that was one of the things I wasn't seeing coming, is this expansion, which has been great for our business, of course, but it's this expansion of players. The other one is the agentic, aspect. I must say that this, the rise of these agents, I, I didn't expect them to rise so fast. and this really happened in the last, twelve, eighteen months. I believe probably-- I probably gave a talk, a panel last year where I, I remember saying this is not gonna be s- it's still not here. It's still something we wanna, we wanna get there, but it's gonna take us another two, three years and, and it's here. So this, this speed of, innovation, that we're w-witnessing. Same thing with speech, speech-to-speech. I mean, look at now, we-- there's several companies already offering, real time... I mean, we can have this call and be translated. I could be speaking Portuguese, you could be speaking in German, and we can have a real time translation on the side. these are all things that, yes, they've been in the works for really two decades, but suddenly they came together because of the power of compute, the power of data, and the competition
Private Data And The Synthetic Trap
Joeri BillastYeah. So you mentioned that, okay, AI models have become dramatically more capable, yet you continue to emphasize high-quality data. So as models get better, does trusted data actually become more important rather than less?
Daniela BragaTrusted data and closed source data, because everybody else already, despite all the efforts from content providers and legislators, and people like myself who has been advocating for copyright payments and IP, the reality is, we know that all the frontier labs crawl the web daily, and even Elon Musk a year and a half ago said that, we already exhausted all the knowledge of the world. What he meant wa-is we already crawled the entire internet. That's what he meant, not the, the knowledge of the world. The knowledge of the world the differentiator lies in the private data, in the closed source data, which is much bigger than that, much bigger. This is what we... that's our value proposition. and a lot of companies are starting to understand, that they can use that kind of data, to train their own, smaller models and custom models. there's a third source of data that's been really on the rise, especially in the last year, which is synthetic data. needs to be well-balanced. Can be amazing in areas where you cannot get data otherwise. But, we need to make sure that we're not now consum-- especially in a world where, models keep getting updated with data that is out in the web, and a lot of the content out there is now synthetic, or it's AI-generated.
Joeri BillastMm-hmm.
Daniela Bragathis is actually what's causing, the other ugly side of AI, and the model collapse is now continuing to ingest open data or publicly available data with the rise of synthetic data. That needs to be, taken in consideration
Why Leaders Should Care About Data
Joeri BillastYou explained, data can maybe sound like an engineering issue, right? why should the CMO, a founder, a business leader care about the data behind the AI systems their company uses?
Daniela BragaWell, first because it's what makes them different from, everybody else. everybody's now... And if you are building in a world of agent creation and everybody building agents, you- your agent is basically a custom model on top of a generic large model. how you're fine-tuning your agent is with the data and the qual-quality of the data you're building it with. the other thing is, depending on who you are, and if you care about brand reputation, you cannot rely on data that has broken an IP, that has not, copyright clear, that is not anonymized, properly anonymized, es-especially if you are in finance, healthcare, and other industries. So you absolutely need this level of governance, and this is why also more and more, standards and groups are being created. The Linux Foundation is building a new standard for agent creation. and there's folks in the United States trying to work on an AI constitution precisely to also to limit the guardrails of what AI can do, where does it get their data-- training data from, and what level of agency should the AI have?
Specialised Data And Humans In The Loop
Joeri BillastYesterday, I attended your press conference and, I learned that the Defined AI has just expanded its catalog across areas ranging from multilingual data and multimodal AI to healthcare and physical AI. that tell us that the challenge is shifting from simply finding enough data to finding the right data for a specific context?
Daniela BragaAbsolutely. every three months there's a new wave of, trends, and we see what the market, wants by the request of our clients. And the last 12 months, we saw not only these trends, changing faster, but also getting more specialized. So we now have more specialized and a lot more human in the loop required. So this is the interesting thing. Humans are not going anywhere, even though everybody's worried about the future of work, and there's more and more, human curation on the data is required as we are moving towards, specialized, mechanics, electricians, type of data, surgeries, dental, and all sorts of beauty. we're seeing a big, and the... So the healthcare, the physical AI, the legal, the finance, we're seeing a huge, specialization of re- of data requests that need, less data, but much more curated, augmented and, and, and with humans on it to make it, to make the data sets perfectly accurate and machine learning ready. So this is, again, I think this is good, good news as well.
How To Prove Data Improves Models
Joeri BillastYeah. I have another, question about that. So your new strategic access program allows companies to evaluate data before committing to commercial use. what should an AI team actually measure to know whether better data is improving the model rather than simply adding more data?
Daniela Bragathat's a great question. Usually, you need to pick a data set, and it's the old school way of breaking the data set in 90% training, 10% testing. And this testing piece, which is obviously we're not gonna use the testing data sets into the training model, is what's gonna give you the challenge of the model. there's evaluation of models that we also run, not in the strategic access program necessarily, but, it's in our Nevo platform, which is our original data collection, data curation, and data model evaluation. everything that has to do with, probing the model, people in the loop synchronously or asynchronously, testing the model, trying to break it, red teaming it. this is what, the engineering teams should be doing. they're building their custom model, and they need to test it. we can see already small instances of that when we build something on Claude, a little agent on Claude, and Claude starts giving us answers, and we say, "No, that's not right." This is a small sample of how model evaluation works, except that we do it at scale. We do it with a demographic breakdown, or skills breakdown and much more, relevant, data points
LLM Search For Data And Metadata Risks
Joeri BillastYou're also using an, AI assistant to help c- customers describe what they need and, and discover relevant data sets. As AI increasingly helps us choose the data used to build AI, what new opportunities and risks does that create?
Daniela BragaWell, our AI assistant is an LLM to help people search. this actually explains the future of search. The future of search is now an LLM really, and it will be an agent. So obviously we- when you're dealing with so much data, we have 10 petabytes of data today, we probably grew this 10X in the last 12 months. it's really hard, and it's a good way for us to actually constantly red team our own, catalog because a lot of times we know that the quality of your results depends on the metadata quality as well. a lot of times, the LLM is a complement to other more traditional ways of search, which we have. the traditional index searches in our catalog, but also allows us to red team the quality of the metadata. and yes, it shows also that we are already navigating the wave. The future of search is an LLM, and next will be
What Trusted Data Means In Practice
Daniela Bragaan agent.
Joeri BillastYeah. everybody talks about trusted, responsible, and high-quality data, but what does trusted data actually mean in practice, and where do companies most often take shortcuts without realizing the consequences?
Daniela BragaI would say that for us, trusted data means data that is consented, it's paid for, And whose metadata is, accurate. So for example, if I have a conversation and I have a transcript, and if the transcript doesn't match the conversation, there might be many different reasons for that, including actually still errors in speech recognition, technologies, which is fascinating being working in speech recognition for the last twenty-six years, it's still not there. It's very close, but it's still not there, compared to the rise of agents, like I just said. So th-that's what it is. Trusted for us means that, But I d- You see, I started with consented. Consented may not be paid. That's true. I mean, we know that we are giving away data in the, you know,
Joeri BillastMhm
Daniela Bragaand the social medias. We know we've been giving away data and some how-- somehow, even maybe we don't know exactly, we consent it about that. But consented and paid for is another one. That obviously gives you access to a much more curated, type of data, data that you can't find in the open web. so that's what it means.
Responsible AI In High Stakes Worlds
Joeri BillastYeah. Now, as AI moves into healthcare, robotics, physical AI, and other environments where mistakes have real world consequences, how does our definition of responsible AI, needs to change?
Daniela BragaThe application itself and the uses of AI, need to be taken into account in the word responsible. Responsible means many things. In the data angle means making sure that everybody is paid in the chain, and that we have, document traceability to prove the provenance of the data and, the rights of usage of the data. But responsible AI goes a lot further down. It's basically what can, At what do we rely on AI and what needs humans in the loop? should AI be completely, unleashed Or should there be limits to certain applications? Just like we stopped human cloning before, not that we didn't have the technology to do human cloning. We, as humans, came together and thought that was probably irresponsible, unethical, and we stopped it. So there-- I believe that there are limits to AI, to what we should build AI for, and a lot of people are concerned about that. I don't see action on it yet
Speed Versus Provenance In Marketplaces
Joeri BillastYeah. I also wanted to ask a question about your partner acceleration program. it is designed to get new datasets into the marketplace faster. How do you reconcile the pressure for speed and scale with the need for quality, provenance, and responsibility?
Daniela BragaAbsolutely. This has been our challenge of the whole year. because of the pressure of the market and the pressure of the offer too. I believe that this year everybody woke up to, "Oh my God, I'm sitting on data that could be monetized." Now, not all of it can. That's the truth of it, and so I wanna make that very clear. And we have a lot of times, partners coming to inquire, if this is even an opportunity they should consider, and we are not in the business of consulting, that kind of thing, right? I mean, we could, but we're not. So, we are really streamlining because we want to be really the Amazon of data. We want, at some point, we are still in the middle, but we want to automate a lot of the access from clients and partners. we want to have at some point, a much more automatic process for people to come together and do a transaction themselves with our seal of, legal vetting. So what we're doing, a lot of it, is we're putting a lot of AI on it. we're introducing a QYB for all the businesses, which was something that, it sounds obvious, but it's not obvious in certain, parts of the world and not easy to get as well. we are streamlining the quality or actually automating the quality checks of the data and the metadata that comes with. A lot of the problems with the data we receive is metadata. It's not consistent, it's missing, it's messy, and that needs to be completely, validated. and we have no capacity neither, we should anymore to check everything. So again, part of the... A lot of AI on it, a lot of, AI automation here. But not less of the concern on the legal aspects and the compliance aspects.
Human Accountability And Indemnification
Joeri BillastYou mentioned already human in the loop, as AI moves from generated content to making decisions and taking actions autonomously. where do you believe human accountability absolutely needs to remain?
Daniela BragaAbsolute human accountability needs to remain. Absolutely. And that comes down to the companies that put an agent or a model out there. I cannot, understand when OpenAI says that just because someone built on my model, I have no responsibility on it. The people who build variations of models on someone's models, or their own, they absolutely need to be accountable for it because, and we shouldn't put a price, like some other recommendation from Anthropic on, yeah, we should put a price on the indemnization, and it should be 50 deaths. anything that impacted 50 lives, 50 lives and I guess was a billion. My God, the, the fabric, the business fabric of the world is not built out of billionaire companies. And why? So what about one life lost? Isn't one life lost enough? How can anybody put a value on a li- on a life, a human life, right? So this is the part that we all need to come together and, and say, "Wait a minute, people. W- it's, it, it's a criminal action if someone dies, no matter how many, and it's also criminal action if you don't... It doesn't matter if it's 50,000 or a billion. Everyone... It's just, it's another... An indem- indemnification needs to be in place. It... There should be no limits or no ba- no baseline
Trust, Sources, And The Future Of Knowledge
Joeri BillastAbsolutely. Daniela, I know your time is limited today. We are coming towards the end of the podcast episode, but I wanted to ask a question which links to my book, "The Future CMO." I argue in that book that, trust becomes more valuable as AI becomes more capable. So what will separate companies that genuinely earn trust with AI from those that simply communicate that they are responsible?
Daniela BragaIt's a great question because unfortunately, the, our... In a world of misinformation and where sources are no longer validated and where the media, I mean, I was just in the future of media here at the Web Summit on Monday. that whole class of, industry is, needing to... not being valued, really at this point, and almost like it doesn't matter the sources, and it does matter the sources. Well, I think the response, we may potentially face because, until these sources come to the table, and discuss these matters, we live a world where obviously brands, brands make their statements. it's a race. they don't necessarily use the right sources. agents are coming out or these, LLMs are coming out with information that is not paid for because they are not making the proper deals with the media publishers and the, historical sources. brand reputation has been historically what keeps a company alive, and I do think that, we may leave, a threat to knowledge if we don't bring universities and publishers back to the table to continue to feed the big frontier model labs.
Joeri Billasta
Where To Learn More And Final Requests
Joeri Billastlot of food for thought, Daniela. Thank you for sharing all of that with us. If people, want to know more, about you and what you're doing building with Defined AI, where should I send them?
Daniela BragaWell, you should send them to defined.ai. they land in our main page and you see a big button say Browse our marketplace, and that's how you log in. You create a login and a password, and you're in. And you see, you have a feeling of what the world looks like today in the closed source world, and how can you contribute and how can you buy data e-ethically
Joeri BillastYes. As my listeners know, Daniella, there is always a blog article with show notes, so, the link that you mentioned will be put in there. really a pleasure to have you on the show again, Daniella
Daniela BragaThank you. Always a pleasure
Joeri BillastGuys, what a brilliant episode. I am sure that you know people around you, that should also listen to this episode, so be sure to share it with them. If you're not yet following the show, this is a good moment. Hit the subscribe button. If you haven't given me a review yet, if you give me these five stars, this really goes a long way, I would love to see you back next time. PS, also a word of thank you to my podcast sponsor, RYO.