Why do increasingly capable AI models struggle to produce reliable answers inside large organizations?
In this episode of Tech Talks Daily, I speak with Misti Vogt, SVP of Engagement at Orange Logic. Her career spans military intelligence, data science, and enterprise content technology, and she also teaches in the DAM and AI program at Rutgers University. That combination gives her an unusually practical perspective on how machines interpret information and why business meaning cannot be assumed.
Misti argues that enterprise AI reliability depends on the context surrounding company data. A model may be technically impressive, but it needs to understand relationships, rules, metadata, rights, and intent. Without that layer, it reasons over information originally organized for people rather than machines. The result may sound convincing while remaining disconnected from the way the business defines accuracy, trust, and permitted use.
We discuss three forms of context. Static context reflects accumulated knowledge. Transactional context develops through projects and outside information. Semantic context helps systems interpret meaning and relationships across large collections of information. Misti compares this with human conversation. When an answer misses the point, we add information until the other person understands what we mean.
Digital asset management sits at the center of this discussion because DAM platforms already organize master data, metadata, transactional data, governance, relationships, and usage rights. Misti believes those structures can give AI applications a stronger business foundation. She also argues that content should become self-aware, carrying information about when it was created, how it was produced, its intended audience, where it has appeared, and how it has performed.
Natural language search provides a useful example of why this matters. An employee might ask for creative assets suited to a campaign and welcome a broad set of suggestions. The same employee may then ask for assets licensed for the United Kingdom and United States, with print and web rights for the next 12 months and no use in another campaign during the previous six months. That second request carries business consequences, so the system needs deterministic rules alongside creative choice.
Misti also shares an Orange Logic customer example involving a conglomerate with several brands. The company consolidated seven platforms, including three DAM deployments and local storage. Orange Logic then supported shared governance across the group while preserving autonomy for individual brands. Misti says the early results include time savings, improved efficiency, richer metadata collection, and lower costs, although no quantified figures were provided in the recording.
The wider question is whether businesses are spending enough time on the information surrounding their content before expanding AI use. Could better metadata, rights management, and business logic produce greater value than another round of model upgrades?
Listen to the conversation and share your thoughts with me.
[00:00:03] Why can an increasingly capable AI model still produce unreliable answers inside a business, even though it has more data than ever to refer to? Well, in today's episode, I'm going to be speaking with the SVP of Engagement at Orange Logic. And together, we're going to talk about the missing layer between enterprise data and dependable AI.
[00:00:30] And my guest brings vast experience spanning everything from military intelligence, data science, enterprise content technology, and teaching digital asset management, or DAM as it's often known in the industry, and AI at Rutgers University. But today, she's going to explain why models need business context. And by that, we include everything from metadata, relationships, rights, rules, and intent,
[00:00:59] before they can reason usefully over company information. And I also want to bust a few myths today and learn how digital asset management is becoming part of the data foundation for enterprise AI, and what natural language search requires behind the scenes, and why every piece of content may need to become self-aware. So if your AI project looks impressive, but struggles with very real business decisions,
[00:01:26] hopefully today's conversation will explain exactly what is missing and what you can do about it. But enough from me. Let me introduce you to my guest now. So thank you for joining me on the show today. Can you tell everyone listening a little about who you are and what you do? Thank you so much for having me. So I'm Misty Vogt, SVP of Engagement at Orange Logic,
[00:01:53] where I spend my time working with the biggest brands in the world, understanding how they can best speak authentically to their audiences. I have a background that spans data science and military intelligence, and then as of the last 12 years, the enterprise content technology stack, all the way from the workflow and creative side through publishing. And lastly, I am an educator in the DAM and AI program at Rutgers University,
[00:02:21] and I always like to look at myself as a lifelong student. Absolutely. Love that. Feels like a book or maybe even Netflix CV series waiting to happen there. What an origin story. There's a lot going on there, isn't there? There is. Love it. And of course, fast forward to present day, we hear constantly that AI models are becoming more capable.
[00:02:45] But I was reading before you joined me today that you argue that this doesn't automatically make enterprises more reliable. So where is this disconnect coming from? What are you seeing here? So this is a great question and something that I get from customers all the time. They're trying to understand the value of AI and how to implement it. And I think there's been a lot of experimentation over the last couple of years now,
[00:03:12] and it's starting to get to the point where we need to see measurable value. And it's boiling down to the context that we're able to give the models. The models are very good. I don't think anybody would argue that. The models are getting more and more sophisticated. But a really sophisticated model without good business context is still not going to provide value.
[00:03:38] So you have to have the combination of both of those along with governance in order to get a reliable output. In every conversation this year that I'm hearing around AI, especially agentic AI and bringing agents into the workplace, is context. And I know you've pointed out that context is one of the biggest missing pieces here. So for people listening, though, what does context actually mean for an AI system inside a business?
[00:04:05] And why do things like relationships, rules, metadata and intent, why do all these things matter so much? So I'm going to make this very human because I think it's most understood easily that way, because it's something that we've all done pretty much our whole lives. So we all share context and context is kind of going around. I've even seen context as a service nowadays.
[00:04:29] And really, it's what we do as human communicators and what we've done our entire lives is sometimes we'll ask somebody a question or we'll have a conversation. And then we might not get the answer that we were looking for or it might not drive the conversation in the direction that we're looking to drive it. So what do we do? We add additional information to help guide that conversation. So with that, I like to think about context in three layers, static context.
[00:04:58] And that's kind of just the memory that you've built over time. Maybe you went to elementary school and then you went to high school and then you went off to college and you've experienced a lot of different things in your life. And that kind of becomes your core static context layer. And then there's transactional context, which is built and evolves over time and adds on to that. And that can come from external sources.
[00:05:25] So you might be working on a project and you're leveraging all your experience and your education. And then you get external information that can influence your decision making or how you might proceed. And then lastly, there's semantic context, which is one of the biggest values of AI because it can stands and helps to translate meaning and relationships across a very vast array of information that it would have.
[00:05:52] A really easy explanation of that is translation. So translation is really just semantics. I mean, in the business world, we have a lot of semantics and specific industries. There's a lot of semantics and AI can help to do that translation to simplify the action and the outcome. And we will have people listening and many organizations that just assume that they have enough data. In fact, they're swimming with data.
[00:06:18] But of course, AI will not necessarily be able to make sense of it. So again, for the people listening, what goes wrong when all that data that they have was never really structured for interpretation by an AI agent, for example, in the first place? I think you nailed it right there with the second half of your sentences. It was never structured for AI-based interpretation in the first place.
[00:06:42] And that's one of the first recommendations that I make to customers that are trying to build this context layer is to start to think about the way that you collect data a little differently. Because you're collecting it previously. Because you're collecting it previously and originally you're collecting it so humans can make sense of it and humans can find what they're looking for. And now it's that and, so machines can make sense of that information.
[00:07:09] And it has to build upon the context, adding the layer of intent on top of it or natural language. So we're updating traditional schemas to add additional context in a natural language and also using JavaScript because Java is very machine readable or markdown files. I'm seeing a lot of that as well to simplify the information so machines can better make sense of it.
[00:07:39] So Gartner has also suggested that stronger semantics could actually significantly improve AI accuracy while also reducing costs. So win-win there. But what do you think this tells us about where companies should be investing before they go out there with their shopping basket, buying more models and more infrastructure? It's the data layer that we really have to look at now.
[00:08:04] So building upon what we just talked about and thinking about how we structure the data a little bit differently, it's a little bit more than that. So speaking about content, I like to simplify the concept and say every piece of content that exists, whether it's a fragment or published piece of content, it should be self-aware. It needs to know when it was created, how it was created, who it was created for, where it's been published in market, how it's performing in market.
[00:08:33] And there's a different pathway to collecting and organizing that information. And I'm also seeing a lot of consolidation of context, and that includes consolidation of different technologies. So before we would see a lot of data silos across an organization. And I think that's still OK. It's hard to change from there.
[00:08:58] But we need to better connect all those systems into a semantic layer that kind of hovers above any of the content technology within the stack. And digital asset management has traditionally been seen as a content operations tool, but I think that has evolved now. So tell me more about why you think it is becoming part of the infrastructure. And that infrastructure needed to support enterprise AI too, because it's quite a big change, isn't it?
[00:09:28] It is. And I think one of the coolest things about DAM is DAM was initially designed with a lot of these foundational building blocks. So it's been uniquely positioned to be an enablement engine for any large language models or applications of AI, because the way it's always been structured is within the different layers of information.
[00:09:54] So you've got the master data, metadata, transactional data, governance that exists over the top of the data. And that's structured in relationships and organized with rights. And then that in itself is already organized information. And AI works really well with that. When we look at everything that you've talked about today, including metadata, rights management, taxonomy, for example,
[00:10:22] how do all these things collectively influence whether an AI system produces something that is useful, trustworthy, and legally safe to use in a business environment? Right. The metadata provides the data, the structure, and the way that that data is structured provides the governance to enable the guardrails that give you a higher level of confidence when you're outputting any type of information.
[00:10:51] So when we're talking about useful, trustworthy, and legally safe, I mean, those are all contextual, right? What does it mean to be useful? What does it mean to be trustworthy? What does it mean to be legally safe? And that's all in the business logic. So the model itself, it doesn't make sense for the model to hold the logic to determine whether or not something's useful, trustworthy, or legally safe. That needs to exist within the business itself around the content so it makes sense for that information to be in the dam.
[00:11:19] And if I was to go into any search engine and type in something, it will give me a list of results there, and then I can select which one of those results I think has got the relevant information I'm looking for. But when we go to AI and ask AI a question or search for something, we're almost at the mercy of the model. We're not sure what it's going to give us. So natural language search is often presented as, yes, simple convenience,
[00:11:44] but what needs to exist behind the scenes for that conversational search to understand what an employee is looking for, rather than just matching words, for example? That's right. And I think I have been spending a lot of time actually this week thinking about the value in deterministic outcomes and the value in indeterministic. And there is value in both depending on the context.
[00:12:11] So if I want to log into the dam and I need to find a group of assets that could be used in a campaign, and I say I'm working on this campaign, provide me a group of assets that I could use or that could be good collateral to build on this campaign, it's nice to have the creative liberty within the machine
[00:12:36] and the indeterministic nature of it to provide me some creative options. And then I can iterate upon that. Where the guardrails really need to come in is when you start to ask more detailed questions that have business consequences downstream. So you might say I'm working on this campaign and I need assets that I can publish in the UK and in the US.
[00:13:03] I also want print and web rights for the next 12 months and content that hasn't been used in another campaign in the last six months. So now you have to combine that creative indeterministic nature with the governance, the deterministic side of whether or not this content can be used for those scenarios. So how do you do that? You put that in the prompt layer within the system.
[00:13:30] So there's kind of like a middleware where the agent can operate within certain guardrails while it's interacting with any particular model. And then that can help to give you higher confidence in any of the outputs that you receive or as you expand and adopt across the organization. Scale your business with agentic AI with limited risk.
[00:13:58] With Denodo's AI data layer, your agents are provided with real-time company data and guardrails for company protection. Create the business you always dreamed of with Denodo. And you can do that by simply visiting denodo.com to learn more. But now, back to my guest. And I always like to try and give people listening a few valuable takeaways. So if we have a leader listening or an organization preparing their content and data for wider AI adoption today,
[00:14:28] what are they most likely overlooking here? And what should they be fixing now if they really want to get those reliable results that we're talking about from AI systems that they've implemented? Do you see any trends in the kind of things that people overlook? I think, yes. We're looking at content differently now, as we mentioned earlier, because we're building content sometimes now primarily for machine consumption.
[00:14:55] And we're interacting with the customer journey at different phases. So we have to think about the information that we're collecting kind of backwards. So if you look at the interaction point and then work your way back, you can ask yourself from this interaction point what information might come in, what inputs might come in from that interaction point that I need to address in the metadata so it can be contextually aware when that comes in.
[00:15:25] And how do I build a framework to collect that information over time? Because the way folks interact with content is constantly changing over time. You have to have that feedback mechanism in, in order to make sure that you stay relevant. And finally, at Orange Logic, obviously you're an agentic content orchestration platform. It's built and used by some of the world's biggest enterprise brands,
[00:15:52] especially brands where content is mission critical to revenue, compliance, and customer experience, and so much more. And for people listening, to give them an understanding of what we're talking about, and maybe a before and after picture here, you don't have to name any names. Are there any use cases or examples you can share of how you've took somebody on this journey, unifying content rights and workflows into this single-governed foundation and the context AI agents needed to act at scale?
[00:16:20] Do you have any big success stories that would really bring to life what we're talking about? We have lots of success stories, but a recent one that I think is really interesting and really quite incredible. And Enterprise Dam Solution, oh gosh, I think it's been maybe seven months since they joined us. So a relatively newer company. We consolidated seven different platforms, three different dams. So they had multiple deployments of the same dam.
[00:16:50] And then a couple instances of local drives, like a LAN or a SAN drive. And this particular company is also a conglomerate. So they own multiple brands, which has historically been tricky for them because each of the brands have their own governance as well. So within Orange Logic, we have the ability to allow a super admin or the conglomerate
[00:17:15] a level of governance that is across all of the brands. So they want to build some consistency because with that comes efficiency. And it also speaks nice to the overall conglomerate that owns it. So then within each brand, we want to give the admins of each brand some autonomy as well. So we were able to do that across all of their brands. And when we launched just a few weeks ago, we're already seeing some really,
[00:17:44] really exceptionally exciting results in terms of time saved, efficiency across the business, the metadata that we're collecting, and overall cost savings. What a great story. And I think that really brings it to life because every day on this podcast, I try and demystify an area that many people may know about but not fully understand.
[00:18:09] So I think today you've done a brilliant job of demystifying digital asset management or DAM and why DAM is emerging as a critical infrastructure for AI, not just content operations. I think that is one of the most important points we can make today. And for people listening wanting to learn more about you, Orange Logic, where they can find out more information and maybe look at some of those other success stories as well. Where should they go?
[00:18:35] I always like to steer you to our website, so orangelogic.com. And please also feel free to reach out to me on LinkedIn. And I'm constantly posting. And you can follow the conversation there, both with myself and with Orange Logic. Awesome. Well, I'll include links to everything there, including your personal LinkedIn. It'd be great to keep this conversation going. Some big changes going on in the industry. And it is an incredibly exciting space at the moment.
[00:19:04] And I think it's only going to continue to pick up. So it'd be great to get you back on next year, see how things are moving or how fast things are moving. But thank you for sharing this today. Thank you so much, Neil. I appreciate it. I think Misty's message gives leaders a useful test for their next AI investment. Before going out and buying another model, maybe ask whether the content already knows when it was created, who it was created for, and where it can be used.
[00:19:32] And also how it can be performed and which rights apply. These details might sound much less glamorous than an AI agent, but they determine whether that agent can act with confidence. I'd also love Misty's distinction today between creative freedom and deterministic rules. Yes, a machine can suggest campaign ideas, while publication rights, geography and usage peers require very firm boundaries.
[00:20:02] So a massive thank you to Misty for showing how digital asset management connects metadata, governance and business meaning. Remember, you can learn more at orangelogic.com. And don't forget to continue the conversation with Misty on LinkedIn if something resonated with you today. And speaking of which, over to you. Is your organisation giving AI enough context that it needs to make decisions that people in your workforce can actually trust?
[00:20:30] As always, you can find me, techtalksnetwork.com. I'm on the road a lot and I run up to the end of the year. So wherever you are in the world, have a look at the events page over at Tech Talks Network. Maybe we can meet in person on a chauffeur somewhere. Other than that, I commit to you. I will be here every day, same time with another guest. And hopefully we can solve a few of those big problems together. So keep in touch and I'll be back in your podcast feed tomorrow. Bye for now.
[00:20:59] Bye for now.