🦄 Making great presentations more accessible.
This project enhances multilingual accessibility and discoverability while preserving the original content. Detailed transcriptions and keyframes capture the nuances and technical insights that convey the full value of each session.
Note: A comprehensive list of re:Invent 2025 transcribed articles is available in this Spreadsheet!
Overview
📖 AWS re:Invent 2025 - From Data Chaos to AI Magic: Supercharging Customer Experience (BIZ209)
In this video, Rafael Flores, Chief Product Officer at Treasure Data, addresses the gap between AI hype and reality, emphasizing that AI is only as powerful as the data feeding it. He identifies critical challenges including organizational readiness gaps, disparate data issues, and the fact that 40% of AI-first projects will fail by 2027 according to Gartner. Flores presents five essential tips for AI success: identity-resolved memory, real-time context and inference, governance and trust, feedback loops, and extensible ecosystems. He debunks the "zero copy" myth and introduces Treasure Data's Marketing Super Agent, which connects to first-party data to create an "agentic brain" rather than just a chat interface. Flores concludes that AI should augment human capabilities through the 80/20 rule—automating mundane tasks while humans focus on strategic work requiring creativity and brand guidance.
; This article is entirely auto-generated while preserving the original presentation content as much as possible. Please note that there may be typos or inaccuracies.
Main Part
From Data Chaos to AI Magic: Understanding the Critical Gaps in AI Implementation
Hi everyone. Can you hear me pretty well in the back? My name is Rafael Flores, Chief Product Officer at Treasure Data. I couldn't be more excited to spend my afternoon with you all. It's going to be a quick 20 minutes, but we're going to have a little bit of fun. Today we're going to talk about data chaos to AI magic. Everyone here has been talking about AI, but they're missing a critical piece: nobody's been talking about data. I'm going to do that and bring it all together because at the end of the day, AI is only as good as what you feed it.
That is what we say at our organization and that is what I believe in as a technologist. You need to ensure that you upgrade the human brain along with how you design products. That's exactly what we're going to talk about. Today, there are a lot of tools out there. They're just interfaces. You go and you chat, you put a prompt. That's just an interface. To many folks who may not know AI very well or are just trying to learn, that seems like a very powerful product. But it's not. It's just a chat prompt calling on a model. It's not as powerful. What I want to talk about and what I want to focus on is actually the agentic brain that's underneath that. If you really think about where AI is headed, a lot of people talk about whether it's just hype or not, whether there are credible use cases or not. Well, some are and some aren't. Hopefully over the next 18 and a half minutes, I can help you understand what's going to work and what's not going to work so you can try to avoid some of the things that we have learned as an organization in supporting our customers who have embraced AI.
I wanted to start with a metric, which is actually pretty cool. Gartner published a report saying that by 2027, 40% of AI first projects will fail. There are a thousand reasons why they may fail, but it's just a reality. When you're thinking about investing into an AI native product, you want to ensure that you're making the investment worth the while. You don't want to be part of the statistic in 2027 that says 40% failed. So how do we make sure that you don't fail? It's all about identifying some of the gaps. There are two critical gaps that I do want to emphasize. Number one, there's a readiness gap. There's a readiness gap not just in terms of products, but also in terms of your organizations. You need to ensure that you're structured in the proper way in-house to even bring AI native products or AI first initiatives into those teams. You need to ensure that you have the right use cases. Use cases are still the same thing. A use case is a use case. It's there for a reason. You have to ensure that you know why you are bringing in those initiatives.
Second problem: unifying disparate data. That's not going away. You can use AI to solve some of that, but if you have messy data, your AI is still not going to be that strong. I'll walk you through an actual example. But you also have to think about the current market, and this is a reality that we all deal with across the board. It's something we deal with at Treasure Data, and I'm sure it's something all of you deal with. The reality is that budgets are not changing. Budgets are a little bit constrained, whether you're in IT or marketing. It's just a matter of fact. The price to compete has actually gone up. To be able to get brand exposure, to be able to land customers, whether you're B2B or B2C, it's just a tighter race. Everyone's trying to win this race, and it's just getting tighter because the budgets haven't changed very much and the price to compete has gone up.
So it's a little bit chaotic. And then if you think about macroeconomics and everything going on in the world, it gets even messier. Every day you wake up, you have to think about AI. You have to think about what is going on next. So how can I help you leave that world and get to a better place where you can actually find solutions to some of the problems that I have just talked about? That's where customer data platforms come in. Treasure Data historically has been a customer data platform. We have become an AI native marketing cloud with the Marketing Super Agent at the top. The reason why we have gone that route and the reason why I believe in that mission is because your AI is as good as the data you feed it, and there's very much value in your first party data. CDPs allow you to do that. If you utilize quality data, you can implement agents that actually allow you to be a superhuman.
I don't believe that AI and agents are here to take your job away, and I kind of chuckle at that because I'm sure some people say that or some of you may believe it. This is my own opinion, but as an organization, we also believe in that because we believe that making you a superhuman is giving you the right tooling or the right AI products to make you successful.
We believe that making you a superhuman is giving you the right tooling or the right AI products to make you successful, regardless of what functional department you sit in. Ultimately, what we want to do is help IT and marketers become superhumans and drive customer experience. That is the job that many of you do. It is not just about what marketing wants or what IT wants. It is ultimately what drives your business forward. This friction or this dilemma has been around for a long time. Marketing wants this, IT complains. IT does not want to re-pipeline, marketing complains. How can we ensure that both teams get the right tooling to ensure that they can succeed? Ultimately, their success is joint. It is not disparate, and their data also cannot be disparate.
Some of the common pitfalls that we see as we work with a lot of global 2000 brands include scale. If you have limited memory or limited context in your AI, it is hard to scale it. Scale is not just in the sense of performance, but also in the sense of your use cases. What are you trying to serve? Number two, there is a common dilemma: do we build or do we buy? It depends on how you structure in-house. If you have highly technical teams and you are ahead on the AI ballgame, then you should probably continue down that path and find ways to accelerate or augment that with vendors. If you do not have those teams, then you should probably be buy first, because it is going to be a heavy investment and a big uplift to try to do it yourself.
Data quality is another critical factor. I cannot emphasize that enough. Data quality matters. AI is not just going to solve every single one of your problems. If you go into ChatGPT today and you give it a bad prompt, that is bad data quality, and that is also going to be a bad response. That is not a bad model. It is just that your prompt was bad. If you are also giving that to your autonomous engines in the background, bad data means bad output. Then there is AI adoption, which is a real challenge. Every company that I meet with, every CMO that I talk to, it is not just about embracing AI. It is also about how do we make the best of it? How do we ensure that we consume those credits? How do we make sure that we actually engage and our teams are going to chat with those interfaces? That is a real pitfall, and it ties to the fact that you need to ensure that you have use cases at the top. You cannot talk about adoption if you do not talk about the use case for the business.
Five Essential Tips for AI Success: From Identity-Resolved Memory to Composability
There are five tips that I will give you to ensure that your AI actually succeeds and works. You can see them all here. I will go through them one at a time. Number one, when you look at an AI vendor, you need to have at the top of your mind the concept that AI and data equals proper personalized experiences. You have to ensure that they can do identity-resolved memory. What does that mean? That means giving you a source of truth profile. The golden record era is what that used to be. Today, you are more entering an era of the diamond record, something that you are more in control of. Something that is not just deterministic or probabilistic in nature, but agentic in nature. You have full control and you can fully trust it, so that when you target that person, it is actually the person you need to target. It is not just wasting precious marketing and IT dollars.
Number two is context and inference. Every interaction matters more than ever. The concept of real-time matters now more than ever. Why? If you have the latest behaviors, forget about attributes and forget about just the profile. If you have the latest behaviors to the milliseconds, you actually ensure that anything you put on autopilot is being delivered at the right time to the right person across the right channel. If you do not have that, you could be missing a moment. Many of you have seen that. How many of you here in the audience have made a purchase and then you get an email the next day or the same day saying, hey, you should go and purchase X? You just bought it. You already bought it. You are not the only one that got that email. Imagine how many of those emails went out. That also means you spent some money sending those emails out or somebody did. It cost the company something. It was not free. Because it is not context and inference, it is not in real time. You need to ensure your AI vendors work and operate in real time.
Number three is governance and trust, which is extremely important. You do not want to be that automotive company that put a chatbot on their website and suddenly somebody was able to go and get a free car. You make the news for free brand dollars, but not for a good reason. You have to ensure that you are built on the concept of trust. It is not so much just the data. It is also the AI.
If you can trust the data and you can trust the AI, you can actually use that vendor. I always tell our customers when they come to us and ask why they should use our AI product versus another option. Why should they trust it? My answer to them is the same answer I'll give you all, which is that you already trust us with your most critical resource—all the IP or PII information you have of your consumers, your shoppers, or who you're trying to target. If something were to happen to that data, there would be major lawsuits, but they trust us because we have a proven record with it.
So if you trust us with your data, you can trust us with our AI as well. You have to have this at the forefront. One of the reasons we're here at AWS is that we work very closely with them. We use Bedrock within our architecture because of all the capabilities it comes with from a trust and governance standpoint to ensure that our brands can actually operate and they're not worried that they're going to get a lawsuit because their AI or their data broke.
Number four is feedback loops. It's not just getting data in real time, it's also iterating on that data. If you go and build a journey—how many of you here are familiar with journeys? There's a few. If you go and build a journey, a journey is a path where a specific profile, like Rafa Flores from Treasure Data who likes to buy a certain car, goes down a journey of how we get him to actually purchase the next vehicle across different channels based on different signals.
Any pivot in me as a potential buyer should be accounted for in that journey. Which means you need to account for feedback loops, because any change that I make in my actions or interactions needs to be accounted for in those feedback loops. And it has to be in real time. If you missed the moment, you missed the moment. You won't get it back. You have to capture the signals and make sure you iterate on it.
Number five is that it has to be an extensible ecosystem. What do I mean by that? It sounds like two big words, but in the world of data platforms, there's a concept of where your data should reside. Should it sit in your data warehouse? Should it sit in your activation channel? Should it sit in your CDP? It doesn't matter where your data sits. You need to have a vendor that can pull that data without charging you extra for it so you can use it.
The concept of composability has really risen in the past few months because companies are trying to adopt and embrace AI. When they're trying to get an agent out the door, they want to feed all the data they have, and that data could be in a lot of systems. The Martech stack or Ad tech stack is pretty broad. You should be able to pull that data. You need to make sure that you have an open ecosystem.
Here's a great example. I love this visual because when we talk about composability and where your data should reside for your AI, everyone talks about zero copy. We need to have zero copy. We need to ensure that we can just move the data freely. But the concept of zero copy doesn't really exist. If any vendor tells you that they are zero copy, that's actually not true. There's always some sort of copy for a specific set of use cases.
If you want to do things in real time, you have to have a copy of that data in your activation channel because you have to have a trigger that actually goes and engages that person or engages that brand. That's a copy. It's not zero copy. That's a copy. When you really think about it, historically, if you go back many years ago, you had four copies in your systems—your data warehouse, CDP, email service provider. Then we moved away from that and said, well, composable CDPs. Now you have the channel where the data is coming from to your data warehouse to your activation layer.
In this new world, if you really want to be in a zero copy type of environment, your data is always going to come in from a specific touchpoint into your data warehouse, and it should be able to flow anywhere. It shouldn't have to sit in your CDP. It shouldn't have to sit in your email service provider. You should be able to really pull it from anywhere you want. But again, there are two copies. So if there's anything you take away from this other than the fact that we're cool and we're at Treasure Data, take it that there's no such thing as zero copy in the world.
The Marketing Super Agent: Building Agentic Brains Powered by First-Party Data
Where does that bring us? I want you to actually meet something that's really cool in this new era of AI and intelligence. We talked about earlier the interface versus the brain. Every prompt, every chat you go on—that's the interface. But you need more than that. If you bring in the right data, you make it intelligent by design. This is our latest product. It's called the Marketing Super Agent. What it does is connect to your first party data. It's not just going and calling on models and pulling in deep research information or doing deep analysis from anywhere on the web, which is powerful, don't get me wrong, but you have valuable resources.
You have first-party data in-house today. This allows you to pull from that plus everything else. So if you want to create a nurture journey campaign, you can. It's doing everything—campaign programs, looking at content, creating digital assets, looking at competitive landscape, and doing the data analysis within your CDP or within your data warehouse, wherever your data resides. This is a 30-second clip of it. If you want to learn more, you can go to booth 235 where we'll show you a full demo. That is what we mean when we say we need to ensure that you use agentic by design products. The agentic brain—because the brain is as good as you train it. Your model is as good as you train it with the right data set. It's very simple. Every company is going to have agents. We have them. Many vendors here have them. It's nothing new. The ones that are going to win are those that actually have brains powered by data that's contextual. It goes back to being able to give context and inference. It's going to be governed, which goes back to trust and governance. And it's going to be in real time, because you need to account for every single interaction to make it meaningful and impactful.
I said earlier that AI is not going to replace your team or you, and I do believe that. I've heard that the whole era of BDRs is going to disappear. This has been a dilemma for the last 20 years, and poor BDRs are still around, and they serve a good purpose. Right now there's a lot of that going on, and there's a lot of fear that AI is raising. Don't be scared of AI. My biggest advice to you is embrace it. Embrace it to your advantage. Make the best of it. Use it as a tool. I use it as CPO of this company. I love this quote from a head of product over at Google. He said the 80/20 rule, which I agree with. If you can put 80 percent of things on autopilot—the mundane things, the things that you don't want to do, the things that you clock into your job and just say, "Oh man, I have to do that report again"—but there's that 20 percent that's always going to require the human element. Focus 100 percent of your time on that 20 percent. That'll make it so much more impactful because in the current era of AI you need to be able to focus. You can't put everything on autopilot. Some things are going to be co-pilot, some things you're always going to have to do.
Brand guidance reviews and the feelings you want to evoke as a brand or company come from you to a human. So there are humans in the loop. I want you to take that away from this session as well. With that, I have two minutes left. Thank you so much. I can take a question or two before they boot me off. Any questions?
; This article is entirely auto-generated using Amazon Bedrock.
Top comments (0)