Name: AI’s Blind Spot: Why AI Fails Without the Right Data Foundation
Uploaded: 2025-06-17T16:01:09.979Z
Duration: 45 min 4 s
Description: AI’s Blind Spot: Why AI Fails Without the Right Data Foundation

Transcript for "AI’s Blind Spot: Why AI Fails Without the Right Data Foundation": Hello, everyone. Thank you for joining us today. We really appreciate you taking some time out of your busy schedule to join us for this Nasuni Unified session on AI's blind spot, why AI fails without the right data foundation. My name is Lance Shaw. I'm the director of product marketing here at Nasuni, and I'm joined by Stephen Catanzano. Stephen, you wanna introduce yourself, please? Sure. I'm Stephen Catanzano. I'm a senior analyst with ESG. I focus on all things data and, AI. Great to be here, Lance. Yeah. Great to be here. Great to have you here too. I mean, I I, you and I have have have worked together in the past, had conversations in the past, and, looking forward to getting into it and, and kinda discussing what's really critically important for any AI implementation out there. So we've got a couple of topics we wanna cover today. We wanna we wanna talk about what some of the needs are. We want to, identify what that blind spot is and and why it's important and why, how it impacts your business and why it's relevant to you. And then we'll, we'll wrap things up with kind of focusing on getting to one source of truth, which is ultimately our goal, and it should be the goal when we talk about AI. So, Stephen, let's kick things off. You working at ESG Research, you've done a lot of surveys. You've talked to a lot of companies and a lot of customers. I think, one of the things I wanna kinda just start out with is, you know, talking about the need. I mean, what are some of the big, factors that you're seeing out in the market today? Oh, sure. Sure. We do a ton of research in the market on data and AI and sort of the market landscape, sort of everything that's going on in the market. You know? One of the big things we see is is really the focus on unstructured data and the challenges that companies have around managing that data for AI. And, you know, one of the one of the stats that we have is that 64% of organizations are collecting data from a 100 to 499 different sources. So tremendous amount of data, and that creates complexity for organizations to think about the the data quality and, you know, the interconnection of all that data as well. Yeah. That is that is that that's actually staggering. I bet you there's some heads nodding out there with our audience right now. You know what I mean? A 100 to 500, basically, a day. I mean, that really speaks to not not only, you know, potential issues with quality, but all and just the volume. I've been trying to gather all that and and incorporate that where it's relevant. That's, that that's it's almost mind boggling. It's a lot. Yeah. It's usually complex for for every organization to do that kind of integration. And, you know, lots of companies do a lot more than that depending on their size. And we're also seeing that, you know, 65% of those organizations are using 21 to 50% of that data for AI. And so, you know, you can think about the complexity there of, you know, what data to use for which AI solution that they're trying to build and, you know, getting through that mountain of unstructured data and and structured data as well. Yeah. I think that's, a a big issue for a lot of companies that, you know, that we speak with in terms of just understanding what I can use and what I should not use and curating it and getting a handle on it, to to use the proper data. You know, when you when you combine that stat about using, you know, the 21 to 50% of their data to AI models and factor that in with the the the the number of data sources, there's a lot of filtration going on there. Yeah. You know? Massively complex. Yeah. Yeah. Absolutely. So, you know, I think it it might be it's interesting to see. I would expect I don't maybe I don't know what you think, but, you know, I would expect those utilization trends to increase that more and more data goes into the AI model. Yeah. It is actually increasing a lot because I think when people start to get their data right for AI, then they start to even look at third party sources of data to bring into that mix as well. You know, you might look at credit report data or weather data depending on your application. So the data this this data complexity is only gonna continue to increase for companies. Right. Right. That's a fact. Yeah. Certainly, I can see within that factors into the number of data sources even increasing as you bring in external data to, you know, factor into the equation. Yep. Yeah. Let's let's talk a little bit about the blind spot, and, and where the disconnects lie for most and for most organizations. I think, interested to see what you think about this. You know, we we, we we know there's a lot of pressure in the market. You know, there's there's the the c suite, the the board of directors, right, demanding at what are you guys doing? What's your AI plan? I wanna see it. You can you present that next week? Right? To the board. There's a there's a lot of pressure around data readiness, probably factoring into, you know, some of what we just discussed. I think, you know, customer expectations. What users expect is is is an interesting one, that, you know, as as consumers in this world, you know, what we expect in our in our daily lives, how that factors into the business, what and what business users expect. And then I I've seen this come up. One of the topics, then I'll ask you about some of these, is, you know, what what are the use cases? What are we using AI for? I think when we I'd be interested to see what you think. You know, when we first came AI first sort of rose to prominence, we saw a lot of companies trying to figure out how to use it or what to do. And and and, frankly, plugging the technology in and playing around with it, but with no clear course of action. I I'm curious what you think about the disconnects and and the factors that pressure organizations today. Yeah. I mean, it's been an interesting market over the past couple of years because, you know, AI kinda came on the scene very quickly, and and you saw it from the board level. Like, we need AI. We need it now. And then that kinda filtered down into the organization of of, like, data teams and IT and others thinking about the data for AI and, you know, what's what data do I need for each AI project? I'm building, you know, specific use case, a customer service application, for example. You know, do we have the right data, and how do we package that data to be used with AI? So it it's been sort of an interesting, you know, step back by many organizations to say, well, yeah, we'd love to have AI today, but we've gotta we gotta get our data right to get there. And then, you know, I see on the customer expectation side too, you know, we're all then very used to things happening very quickly and easily in our lives when it comes to data. And, you know, one example is I was in San Francisco last week, and you see Waymo now driving around these, you know, self driving cars with, you know, no one no one involved in it. And Yep. It's interesting when you think about the data for those cars and how important it is to have it right every time. If if they don't have that data, the freshest data, the newest data, the quality data that they need, you know, it's it's gonna go down, you know, one way street's the wrong way and lots of other things can happen. But that's the same take that same example into sort of every application within an enterprise, and you have that same challenge. You want the outcome to be perfect every single time, and you got what customers expect in their lives and now in business as well. Yeah. And, you know, you you made me think, this is, the example of the Waymo cars in San Francisco. Right? It data changes quickly. Like, who knew that, you know, Tuesday morning, the, Department of Transportation was gonna tear up the street and and and or there was gonna be some emergency constructors or water main break. Something happened, and it happened very recently. And there's no way that, you know, Google Maps doesn't know about that. Right? I mean, the day the the data flow from all of these different edges, all these different locations has to be quick and incorporated within the collective, if you will, that to to make use of it so that the driving car doesn't go down that road today. It went down it went down the road yesterday. Don't go there today. Yeah. Really good point. It's not just about the data. It's about the freshness of the data. It's about the trust of that data. All those things have to be factored into almost every use case that you think of for when you're building something around AI. Alright. Well, speaking of use cases, a little bit of a segue there. Stephen, what can you tell us, you know, a a little bit about what you're seeing with ESG Research in terms of what customers are looking to address? Sure. Yeah. Some of our research recent research to with company is talking about, you know, what are what are their thoughts on AI and where do they wanna implement it? Biggest one is customer service. We see that as one of the the biggest areas where they can get the most, productivity gains from. And then others are, like, marketing and software development, research, even cybersecurity, legal, finance. It's kind of across the board. It's it's really kinda fascinating. You know, I I was with, a group, a couple weeks ago and talking with a chief data officer and asked him the same question. And he said, we literally have a 100 different use cases. Every line of business coming to us saying, we'd like to implement this kind of AI within our line of business. You know, how do you how do you build this? So, there's no shortage of use cases that companies are looking at. But, you know, some of those first ones are on customer service and one and internal ones we're seeing gaining speed a lot faster than sort of creating an external, you know, customer facing one. You know? I think people need to get the first one right, and then they can continue to expand on more and more. Yeah. Yeah. I think it's interesting what you said about, the fact that they're coming to they're coming to, you know, the IT department and saying, I wanna do this and and I wanna do it with this model. Right? Or I wanna do it with this tool. I want so they're come they there's almost a, a requirement there to be open. Right? As the IT provider of of data, ultimately, I've gotta be able to work with just about anything. Right? Yeah. And we're seeing these kinda cross functional teams in there too. So it's you know, once they start a project, you know, there's finance involved and IT and data guys and even up to c level executives. That's how serious people are taking building out AI and and also wanting to look and measure the value of it over time. So those those teams are getting in early to be part of the whole discussion and not waiting till later to figure out, you know, where's the benefit, where's the value, and everything else. So Yeah. Interesting. Do you get a sense when when you're talking to IT folks like that, Newt, and talking to the the the folks you speak with, do you get a sense that there's, like, an overwhelming amount of information they have to process? Just, like, just so many things to think about. Yeah. Tremendous amount. And it's, you know, you start to break down. It's not only getting the data right, but now okay. So how do we go to market this? What are we gonna build? Is it gonna be a chatbot? Is it gonna be an AI agent? You know? And then, you know, the interconnection of data with, LLMs, you need RAG and tools like RAG and MCP and vectors. So think about all these different things that they have to sort of plan for. So it's not just, you know, the data. It's what are all the other components and what is the outcome of what we wanna do with this application as well, and and how is that gonna act. So it's a tremendous amount for people to think about and and such a constantly changing market at this point too. AI just advancing at such a tremendous rate. Right. Right. You know, you mentioned agents, you know, six, nine months ago, most of us were not saying the word agents on a regular basis with regard to AI. Right. Now it's common place and it's it's it's accelerating rapidly. The same thing's happening with MCP. I mean, it was, you know, it's showed up as a great as a great, you know, standardized way for access. It's exploding. Right? So there's just an who knows what's next? Right? There's just always something else coming down the pipe. So it's a it's a lot to consider, a lot to think about. Yeah. Yeah. We've we've added a lot of new acronyms to our vocabulary over the past couple of years. Yeah. You know, that I think that's just what everybody, watching today, really wanted. They're looking for more acronyms. Oh, yep. There's definitely a shortage in tech. Right? Yeah. Yeah. MCP is the latest one. So it's, yeah. Another one to learn. We're gonna run we're gonna run out of letters at this point. Yeah. You know, so you're thinking about, you know, the all of this, that we that we're that we're looking at. I mean, I think the, you know, kinda wondering where the points of emphasis and where organizations then prioritize because there is all of this to think about. There is you know, do you have to factor in all these technologies, all the demands from various aspects of the business? How do I how do I prioritize that and and get that right? Yeah. You know, we we've asked those questions in our research as well. And what really which which is actually, I think, a really good point is that is that the highest that most people focus on is data quality. They they understand once they start getting an AI, they take that step back and they realize, okay. We gotta figure out what's the use case and where's the data and what's the quality of that data. And if you don't trust that data, you know, it's the same thing. The AI is not gonna respond the way you want to unless you have trusted data. And then, you know, the integration of those 100 to 499 data sources that we talked about earlier is is another challenge along with, you know, data validity. And I think one of the biggest things is the freshness of data. And and when we talk about freshness or real time data, it's not it's not streaming data. It's, you know, what's the most recent, you know, report that we wanna use? We may have 10 copies of a report. Well, what's the most recent one? What's the fresh data that needs to be part of this AI? So overall, challenges that data teams have to go to to sort of build that foundation for each AI use case that they wanna build. So lots of different challenges, but quality is top of the list and quality ends up being, you know, trusted data at the end of the day, which is really the most important thing for AI. Yeah. Totally. Totally. I mean, I I recently, was listening to another AI podcast and, they were talking about there are with AI these days, and I think most of us would probably agree, there are no breaks. It's just going and it's gonna accelerate, and there are very few factors that will slow you down. One of them is energy. Right? Which is why you see companies investing in electrical grids and nuclear power plants and and all kinds of things. And and the other is data quality. And if if you don't have the right data, you've got nothing. And and and that's, and I I I kind of agreed with that when I heard them. Like, yeah. I mean, of course, it's gotta be right. And by the way, how can it how can you be sure you've got it right when you've got, how many was it, 100 to 500 different Yeah. Data sources to deal with. Yeah. Yeah. I'm sure those are all accurate and up to date. They're all in sync. Right? Yeah. Right. Yeah. Hey. You know, one thing, about this data that you have, in in this chart, you know, there's a organizational culture and skills is way down there. It's the bottom of the barrel. I think that's sort of interesting. Is that just means, like, we don't care about people. We'll figure it out later. What what what's the what's the priority there? What do you think about that? I think it's just the priority today. I think that that whole, it it's all gonna flip around at some point. Right now, it's let's get the data right. Let's get quality right. And then they have to really start thinking about the culture of of within our organization too. Like, how do you build, a data driven culture is really important. You know, everything from the person entering data and that's needs to realize, well, the accuracy of that data that they're entering is incredibly important because down the road, that ends up in AI. So there's a culture shift, and then there's the whole shift of using the AI for for organizations and the and the employees as well and and, you know, education and training and all those things has to happen too. But it seems like a big focus on the on getting the data right first before they focus on the people. Yeah. I I I think it makes sense. I can see where where to your point about this is gonna you you would expect to see this flipped on its head in a year or so where then the real focus is on getting people to make the best use and create and make sure they are they do have the most accurate data. Right. Yeah. So what are some of the, you know I mean, so we know we we know we know data quality is important. How do how how do you see companies, you know, ensuring that's the case? And what are they doing to make sure that happens? It's a it's a challenge. You know, one of the tools that's out there that really is the glue between your enterprise data and large language models, and and that's using RAG, so retrieval augmented generation. And it's really you know, it can help with accuracy of data, and and this sort of combined with using vectors as well. So you can vectorize your data. It gives it, you know, another level of of quality and resilience, you know, so that you can when you run a search, you can find more accurate data. So RAG and vector kinda work together. And, you know, it's really about improving the accuracy, accuracy of the data and accuracy of the results. And, you know, RAG's sort of that real real piece because, you know, like, large language models, they they have a lot of intelligence and natural language processing, all these things. But what they don't have is the context of your data. And that's what RAG ties it together in a secure way. So now it can be about your customer data and your history and all the other things without, you know, taking your data and integrating it into a large language model or having to build them yourselves. So really two important tools that companies need to be thinking about as they're building a solution. Right. Right. Right. So there there's a developing these solutions and and then these have to continue to scale because we're just getting started. Right? So where there's where there's five today, there'll be 10 tomorrow, and so on and so on. And before you know it, you've got a lot a lot to a lot of integration points and a lot a lot to a lot to consider. Right. Yeah. Yeah. Really important part of, the whole story of bringing your data, your enterprise data to, build out an AI model, whether that's chatbot or AI agent, whatever it is, RAG and vectors are a key part of that. Absolutely. Absolutely. You know, you were talking about use cases and using Rag. You know, one of the one of the most common scenarios that we see comp companies maybe maybe, you know, starting out with whether one that makes the most sense is leveraging institutional knowledge, whether that be on an external basis or an internal basis. Just because, you know, I can do things that you you you showcase, customer service. Right? I would like to make it easier for my customers to find the information they need, preferably without even having to call anyone or send an email or any of that old old school processes. Right? So I'm I'm I'm looking for ways to expose that data. Are you seeing that as a as sort of the, you know, the kind of a similar similar scenarios where customers are or organizations are starting out with using the institutional knowledge either for internal or external uses? Yeah. It I think it's the the top sort of use case that people can focus in on because it's it's it's utilizing information that the company already has, and then that becomes your data foundation, your file data, all your unstructured data. You know, if you're building a knowledge base, it might be all your marketing materials and your customer records, sort of, all those things can be combined to create that foundation. And then then you're going through those steps of indexing that data using vectors to to make it more more relevant and searchable. And, and you're tying that into a large language model, so and you have a choice of all the different models that are out there depending on, you know, if this is a multimodal solution where you have images and graphics or they're just text based. You know, that's another challenge that companies have is, you know, which LLM to really use. And then, you know, like we said before, it's building out one of those solutions. It can be a chatbot. It can be an AI agent or or something else. But that's sort of the the progression that companies are working through right now. And institutional knowledge database, I think, is a great one for many companies to be starting with. Yeah. You mentioned agents. I mean, I that seems like that would be the next big thing. As I said earlier, you know, that that's that's on the rise and a lot of activity now around agents. Is is that, something you're seeing a lot more of? Yeah. I mean, it's it's it's, you know, very short time. Like, in the past year, agents just came on the scene, and and now everyone is all about agents. And, you know, our research shows that 67% of organizations are planning or or for, you know, considering agents now. And, you know, the difference is, like, if you if you look at the example we just we gave us a knowledge base where a chatbot is great and chatbot can communicate with the knowledge and answer questions and do all those things. An agent takes it to the next level that can take action on that as well. So someone might be asking, you know, your, your solution about, you know, their history of orders that they've taken or a contract or an agreement or any of those things that relate to them in their business, and they can get information back if it's a chatbot. But an agent can go to the next level and even be able to say, okay. Would you like me to send you this email, you know, email this contract back to you or take steps and actions? And that's where an agent really comes in, and that can be fully automated without any people being involved in it. And so that's a huge step in the industry, and you can think about the cost savings that companies can have when they start implementing agents that are gonna take those steps and reduce, you know, manual intervention and and some of the, you know, mundane tasks that, we kinda go through on a regular basis on a, every day. Yeah. You know, when you when you talk about, you know, the automation of this. Right? Because it's one thing to have a a multi step workflow that an agent is walking through. We'll call it that Right. For the sake of argument. But the as soon as that is automated and a lot of things just happened behind the scenes that I didn't necessarily see. I only see the outcome or the output later on. That data better be right. We talk about Yep. Data quality, any along the way, every little step. Right? The you know, it's the, you know, send me an email, automatically went to the wrong person. You know, maybe that wasn't a great idea. You know, any I know that's probably a terrible example. But but, you know, just the the the idea of having your you know, the the data that the agents are drawing off of and referencing, whether that's using MCP and other tools and and other services or or whatever the case may be, that's, that is really gonna be critical. Yeah. Even even that example, right, you know, if it's sending off the old contract and not the latest contract when you requested that contract to be sent to you, Right? That's that's not gonna go very well for customers customer satisfaction, and you're sending off data that's old. I mean, that's it comes back to that foundation of data. Is it the freshest data? Is it the trusted data for that agent to be using? And really, really critically important to make that right. Yeah. Oh, absolutely. Absolutely. I think we're gonna see, see a lot of folks probably run into that, and that is really kind of the blind spot for folks for most organizations is not really seeing the full picture, not having too many data sources, not being able to ensure data quality. I like the way you you you call it you call it data trust. And that's really what that's that's a great way to put it because because that's that's really what this is all about. Am I am I really feeling confident that I'm I'm delivering the right data to all my different processes? Right? That's, at the end of the day, if you're not feeling comfortable you know how it is. If you think something's wrong, it probably is. Right? You probably should go check that out. Yeah. I've never seen that. Yeah. So, I guess that's that's really what we, you know, we wanna encourage all of our viewers today, that are watching. You know, this is what you need to think about in your organization about the data quality, data trust, the confidence that you have the right information at the right time, the right place, and delivering it to your whatever the case may be, whatever AI related service or prompt technology you're using, whatever the case, you need to be able to make sure that's true. And and so there's a couple things there that are, I think, ultimately required. I don't know what you Stephen, if you agree. I mean, I think there's first, there's the, you know, there it has to be accessible, right, around around the globe, around for everybody. That that kind of that one source of the truth. Right? The the one place to go, not 500 places to go. At least get that number way down would be, would be a goal. But, you know, the and the and the seamless movement from the edge to the cloud because there are a lot of cases where there's something recent that happened over in San Francisco, but, you know, all of my data is stored in, you know, in, Topeka, Kansas. Don't know why I went with Topeka, but we'll we'll go with it. And and, but so then that has to be delivered. That's gonna be gathered at the end. That's not I don't think the San Francisco to you know, maybe they're they're not probably sending that data into my headquarters. I need to gather that locally or bring that data in, incorporate it within the collective so that I have that full most recent freshest data available to my to my for my AI purposes. You know, the other thing that I think is is is we haven't really talked about it a lot, but it's part of the equation is resilience, that touches on, you know, business resilience, AI data resilience, cyber cyber resilience, you know, and, in the previous session UNIFY session, my colleague, Jim Little, mentioned this in in as well. You know, if if if if anything is down behind this, if the services are unavailable or if the data has has has, gotten lost or damaged or anything like that, that radically impacts all of the agentic processes we were speaking of earlier, everything. Right? And so that cannot happen. That you has to be reliable and constantly available. And if something does go right, it's a very quick recovery process, so everything's back up and running. And, you know, and being able to retrieve, you know, the the the right data, the contextual data across, you know, what what could be different locations within your realm, but all collectively managed under a single umbrella. And and these are the kinds of things that, that that are really required when when a company is thinking about it. Is there anything that you would add to that list, Stephen? No. I think it I mean, I think it's a great list. I think it really fits. And, you know, it comes back to, resiliency, I think, is incredibly important. I think data governance is important, security. I mean, all the all the things around your AI that you need to make sure we're right, you know, protecting PII and all all the things that might show up in data, is is another, you know, big aspect of it. So it's it but it comes back to that I think it does come back to that trust. Right? Do I trust this data? I trust that it's governed right. I trust that it's secure. I trust that it's accurate. You know, when you start thinking about trust as a term, I think it really just, you know, brings it home to say, yeah. This is I trust it. Now I can use it versus, oh, it has good quality. Well, good quality is great. But, know, trust is a whole different level. Or or or is it? Is it great? Yeah. Right. Right. It's okay. There might be some errors with that data. I think it's probably alright, but wouldn't worry. Right. You don't you don't wanna feel that way, but you wanna feel confident right at the end of the day. Yeah. I think that, you know, that as as we, as we're we're talking about the blind spot, it's really about most organizations, a lot of organizations will say that, have not considered how they are unifying their data, how are they bringing everything together. We called you know, this session is called Unify. Right? And that is at the heart of any AI initiative, being able to ensure you have the right data. It's incorporated in one location. It's not scattered to the four winds. You don't have all that, you know, 300 different versions of a file that you, you know, you kind of alluded to earlier. You have edge access and you can really, you know, you can you can you can really make sure that everyone is working off the proverbial same page, in a in a global way, with with an infrastructure that allows you to incorporate all those previously siloed datasets into a single global name space, into a single location so that, you know, that all the users are working out the same data. Also, by the way, when that when that happens, you are it's much easier to find out where you have a discrepancy. I do have two different versions of the same file from some that were incorporated from different silos. Now they're now I can reconcile those and make sure I have when I'm curating my data, understanding what I have, eliminating that redundancy. And, and there's actually some huge cost savings that go along with that when you do that sort of unification. But having that, that data foundation in place is is essential. So, you know, certainly here at Nasuni, that's what we do, but I would encourage anyone that's looking into this and considering your AI strategy going forward to really give a hard look at how you maintain data quality and ensure data trust. So that's what I had to say about that. You know, is there Stephen, is there, you know, you have some closing thoughts or anything else that you'd like to, to to bring to the audience here? Well, I think the unity of data goes a long way because when you unify data, now you can much more easily incorporate data governance and security and resiliency around that data. So that's a really important part of making sure it's right. And, you know, I think one of the big things that people need to think about is, at the end of the day, we're all gonna have access to AI. I can get to every LLM in the market. I can get to all kinds of tools and all those things. But the the big differentiator for me is my data. That's the context. I can create competitive advantage. I can create lots of different value for my customers. And and so that's really the most important thing that people need to focus in on, and and it's important to get started. Even if it's one use case of building a knowledge base and then learning and then building off of that, I think that's where many companies are. It's, you know, important to take that first step. Work on one project and then get good at it, and then you can continue to add more over time. Yeah. Yeah. Great point. I mean, the urgency is the urgency is there now, and it's only gonna wrap up. So it's definitely time to move for sure. Competitors are are building something right now. So No doubt. No doubt. That's true. No matter what business you're in, that is absolutely true. Well, Stephen, listen. Thanks for joining us today. Thanks for having us. It's always a pleasure to work with you and have conversations with you, and, it's always enlightening. Really appreciate you, joining us today. Yeah. Thanks, Lance. It's yeah. It's always great to be here with you, and, I think it's such a great topic and really love everything that's happening with the SUNY and, and Unify at this point. So excited to be here today. Alright. And to our audience, thank you for joining us today. We really appreciate it, and we hope this was educational, informative, and implores you and gets you thinking about what you can do to improve your stance and your data foundation. Thank you so much for joining us.