View Recording: From Insight to Impact: AI-Driven Decisions that Win Markets

Brian Haydin All right. Well, hey, everybody. Thanks for joining us for today’s webinar. I’m going to be talking about insights that we can get from data using a I, you know, and making real a I driven decisions. You know, when I came up with this idea a couple of months ago was before. 0:0:19.781 –> 0:0:34.861 Brian Haydin Months ago was before was before Ignite, and so a lot of announcements kind of changed what I wanted to talk about here today. So we’re gonna do a little bit of a pivot and I know it’s kind of close to. 0:0:35.421 –> 0:0:54.901 Brian Haydin The holidays and you know, glad that you were able to join us, but it’s going to be kind of a smaller group. So please feel free to drop any questions that you want in the chat. I’ll be keeping an eye out for questions that might pop up and I can give you a really personalized kind of walk through on some of this stuff. So a little bit about me. My name is Brian Haydin. 0:0:54.981 –> 0:1:10.701 Brian Haydin I’m a solution architect at Concurrency. By day, that’s what I do. And at night, on the weekends, I’m kind of an outdoorsy nutcase. And so mostly you’ll see me out in the woods or on my lake. 0:1:11.301 –> 0:1:27.861 Brian Haydin Or on the Great Lakes on my boat, but I’m really passionate about the Microsoft ecosystem. Low code, pro code, no code data, Samantha Kernel, AI Foundry, etc. Follow me on LinkedIn. I try to get. 0:1:28.341 –> 0:1:47.701 Brian Haydin Some sort of article out like every couple of weeks talking about some. of the latest things and love to connect with you on LinkedIn. But all right, so I use hunting analogies a lot and So what I was kind of thinking about is with this particular. 0:1:47.941 –> 0:2:7.541 Brian Haydin Another topic is something that happened to me over the last couple of years that our hunting property, you know, neighbors, we have neighbors and one of the neighbors put up an 8 foot fence, you know, around the property. And so it forced me to like, you know, not use like the normal things that I would do. 0:2:7.621 –> 0:2:26.541 Brian Haydin Or like the reports that I would run on a monthly basis or like last year’s like trail cameras and tracks to like figure out what I was going to do because the patterns are shifting, you know, like as as new capabilities come out, as new things happen in the ecosystem. 0:2:27.221 –> 0:2:42.461 Brian Haydin You have to shift the way that you think. So in today’s reality, markets are moving way, way too fast. Like things are shifting. Technology is shifting. Most organizations are drowning in data. They don’t even know what to do with it, but they’ve been told to, like, keep everything. 0:2:43.61 –> 0:3:2.381 Brian Haydin And So what I’m hoping to do today is help you understand what tools are available, what equipment that you have that can help you navigate the data to help you make insight-driven decisions faster, not doing the monthly reports, not waiting until the quarter’s over. 0:3:2.581 –> 0:3:18.101 Brian Haydin Or in order to like figure out what you want to do next. So what am I seeing? The first I’m going to do is I’m going to talk a little bit about the trends. This is the end of the year and I like to, I’ve been actually spending the last couple of weeks just sort of like reflecting on like what happened this year. 0:3:18.181 –> 0:3:36.741 Brian Haydin So if we start to look at the trends and data around a WS, Snowflake, Google’s BigQuery, Tableau and Salesforce, like what are they all doing across like the ecosystem that is enabling AI to help customers make better? 0:3:36.741 –> 0:3:53.701 Brian Haydin Insightful decisions. And so think of these kind of like as like the latest electronics that have come out this year that not everybody had a had a chance to buy them. But you know like we’re starting to get the reviews in and what people like what how successful they were fishing with this new fish Finder. 0:3:54.821 –> 0:4:10.941 Brian Haydin So the first thing, you know, big trend is using conversational analytics, you know, to get more information. And it’s actually becoming kind of like the default way that people are interacting with information. 0:4:11.461 –> 0:4:24.501 Brian Haydin So I think of like Microsoft Copilot in Fabric or using the analyst agent inside of Microsoft Copilot. It enables pretty much anybody to ask questions of the data as if they were like. 0:4:27.341 –> 0:4:46.661 Brian Haydin A knowledgeable consultant, you know that that that worked within the business. So here’s an example of of a question that one of our analysts asked a data warehouse that they built. It was, you know, show me the products by SKU that were the least profitable in the organization. 0:4:47.341 –> 0:5:6.181 Brian Haydin And you can get that information you know directly out of that. So if you want to have, if you want to ask you know your data platform like show me the sales trends by region or highlight any anomalies that were there in the last quarter, that’s the the kind of things that we’re talking about and. 0:5:6.461 –> 0:5:21.581 Brian Haydin Everybody kind of is starting to build this into it. Snowflake has Cortex, Tableau has Pulse, AWS has Q, you know, Google Cloud has, you know, has the Gemini that’s attaching to BigQuery, you know, being able to do that. 0:5:21.941 –> 0:5:37.541 Brian Haydin And what’s really happening in the background is taking that natural language and it’s kind of turning it into a SQL query and you know, getting the answers for you. So that’s one trend that we’re starting to see this year and I think most of the major platforms have adopted, have adopted that. 0:5:37.621 –> 0:5:52.581 Brian Haydin Taking that a little bit of a step forward, step forward is what if you had systems that aren’t really connected to your data warehouse and so RAG or like retrieval augmented generation is isn’t. 0:5:52.661 –> 0:6:9.701 Brian Haydin Is not just a feature anymore. It’s like your knowledge system. It’s your knowledge layer. And most business critical insights are trapped like within, you know, different types of PDFs. Or you might have images like receipts or documentation that is unstructured. 0:6:9.701 –> 0:6:28.301 Brian Haydin That you have to have to compute into something else. Or maybe you have something that’s sitting inside of an API that you can’t quite like bring into your analytics platform. So what we’re doing is we’re seeing people build off of those systems using RAG or using other pipelines. 0:6:28.981 –> 0:6:46.901 Brian Haydin To build more knowledge into the conversational aspect of what they’re doing, Databricks has Mosaic, Click Answers, you know, is able to comb through structured and unstructured data. And again with Amazon and Q, they’re starting to be able to do this. 0:6:46.901 –> 0:7:4.901 Brian Haydin And the cool part is that it’s not just like pulling the data out raw and throwing it in, it’s actually giving it somewhat of a semantic understanding about it. We’re gonna talk a little bit more about it, but in the Microsoft side, Foundry IQ is kind of like a knowledge layer for. 0:7:6.621 –> 0:7:23.301 Brian Haydin Agents and that’s sort of well that is built on Azure’s AI search, but it’s sort of building a commoditized wrapper around it so you don’t really have to have a deep understanding of how to build things in Azure. You can kind of one click it similar to what you do in. 0:7:23.381 –> 0:7:42.781 Brian Haydin copilot studio. So another trend that I’m seeing is the semantics and this is really something that I, you know, wanted to highlight after Ignite and LM’s are really good at like taking that language language part in in the first trend that I was talking about turning it into SQL. 0:7:43.61 –> 0:8:2.621 Brian Haydin You know, pretty good at at things like that. Even being able to like ask it and then have it figure out what to go and how to go and find the information if you have some sort of MCP that that it knows how to use. But it doesn’t really have. LLMS aren’t really good at understanding like your business. 0:8:2.821 –> 0:8:18.101 Brian Haydin Language like what a customer means in this context until you give it, until you actually explain it to them. So that’s that semantic or the business meaning that is necessary to really help like unlock the AI insights. 0:8:18.501 –> 0:8:34.261 Brian Haydin And I was really, really excited about Fabric IQ and IQ is going to come up quite a bit in the Microsoft ecosystem. And that’s exactly what that does is allows you to take all the different data elements and describe what those things mean to each other. 0:8:35.141 –> 0:8:50.141 Brian Haydin And generate some sort of a knowledge engine or a knowledge graph that can be used in order to interrogate or go through the data at a different level. So what you’re seeing on the picture here is Microsoft Fabric IQ. 0:8:50.661 –> 0:9:6.741 Brian Haydin But Google Cloud has like a BigQuery knowledge engine that’s in preview that they’ve started to, you know, expose to some customers. SAP, SAP has Jewel and then you know we’re going to dive a little bit more into the the fabric aspect of it. 0:9:7.381 –> 0:9:22.781 Brian Haydin But it’s like, you know, being able to tell the difference between if I’m looking at a topography, you know, topology map like is that a lake or is that a swamp? I can we know which one’s which so I can navigate things, you know, a little bit better. Another trend that we’re seeing. 0:9:23.221 –> 0:9:39.821 Brian Haydin In in the industry is that agents this year are really starting to move from answering questions to actually performing actions you know for for the you know on behalf of the users and we’re seeing a lot of. 0:9:40.261 –> 0:9:56.621 Brian Haydin New features coming out in these tools that that give it access to tools and allow it to do more meaningful things. So at the beginning of the year, most people were just doing chat bots and then copilot Studio started to be able to incorporate MCP actions. 0:9:57.61 –> 0:10:13.21 Brian Haydin You know, and other type of things that you could do. So inside the industry, AWS has Bedrock agents, SAPP has dual agents, Databricks has Agent Bricks, Microsoft has Copilot Studio and Copilot agents. 0:10:13.141 –> 0:10:32.141 Brian Haydin As well as being able to build your own, your own agents in like more of a procode kind of way as well. So using AI Foundry and you know a bunch of other mechanisms as well. And so one of the cool things that we’re going to really dive in today is is around these different. 0:10:32.661 –> 0:10:52.381 Brian Haydin Layers of how you build agents and what that actually means, the difference between like just specific data agents or like action agents or something that’s hybrid or like kind of in the middle of that. Another trend that I’ve wanted to notice this year is or actually it really hasn’t taken on. 0:10:53.501 –> 0:11:12.301 Brian Haydin This year it’s starting to take off right about now and it’s gonna really snowball. It’s gonna start to really snowball throughout 2026 is trust and governance and the amount of conversations that I had in Q1 were, hey, tell me about how I can keep my data safe and. 0:11:12.581 –> 0:11:27.341 Brian Haydin We talked about Purview, we talked about this, we talked about that. Copilot has all that kind of built into it and that was great. But now people are like, how can I really trust it? How do I know that the permissions are working? 0:11:27.861 –> 0:11:47.261 Brian Haydin What are the some of the the next level governance features that I can incorporate into into my workflows as I’m building these agents? Evaluations you know are something that if you’re not familiar with the term evals in in this context, it is the way for you to be able to measure. 0:11:47.461 –> 0:12:3.661 Brian Haydin You know the the response that comes back from an LLM versus what you actually expected. But all these tools basically have RBAC controls built into it. Microsoft Fabric has, you know, basically from the very beginning. 0:12:3.861 –> 0:12:19.941 Brian Haydin Has had that Google BigQuery has RBAC controls and you know, so I think you’re going to see people start to go back and say, hey, wait a second, this stuff has got a lot of dangerous content, dangerous from the standpoint of like. 0:12:20.61 –> 0:12:36.981 Brian Haydin Data leakage that I don’t want to like, let this just go out into the wild. So people are going to start paying a lot more attention to that, you know, as as this year starts to this next year starts to happen. I’ve talked a little bit about evaluations and observability, so here’s kind of a screenshot I was talking about. 0:12:37.941 –> 0:12:57.21 Brian Haydin And inside of Copilot Studio, this feature was released at Ignite where you basically could say here’s the question that I want, what the test method that I’m gonna use, here’s what I expect the answer to be and it’s non deterministic to a certain degree. So when I ask. 0:12:57.141 –> 0:13:11.501 Brian Haydin I’m not gonna get Apple as the answer, but I do need to know that it’s like kind of loosely related to that Apple thing that I’m expecting it to come out of. So I have seen Databricks making some investments in that. Copilot Studio built this as part of it. 0:13:12.181 –> 0:13:31.861 Brian Haydin Foundry’s had this for quite a bit and and so I think that you know you as an organization should you know start to reflect on this trend and what that means for how you’re going to build agents you know throughout 2026. So those are some of the trends that I’m thinking you know that I I’ve seen happening over the last. 0:13:31.861 –> 0:13:47.741 Brian Haydin You know specifically accelerating over the last quarter, but throughout 2025 we’ve been seeing the the momentum build in some of those areas. But I wanted to talk a little bit about like when we you know how we’re going to build something that allows us to make a I driven decisions. 0:13:48.301 –> 0:14:3.461 Brian Haydin So that’s kind of like the vendor neutral approach and now I’m gonna talk a little bit more into the Microsoft implementation layer aspect of this, but it’s important to kind of break this down into like. 0:14:3.461 –> 0:14:21.101 Brian Haydin The three different areas of three different layers that we’re going to that we’re going to explore. First, we’re going to talk about the unified data platform. This is making sure that I have a single source of truth against all of my structure and my unstructured data in a way that has got the governance. 0:14:21.781 –> 0:14:41.701 Brian Haydin Compliance and security kind of built into it, you know, from the beginning. Without that foundation, it’s going to be really difficult for us to go to the next layer, which is building the semantic layer on top of that and having a giving the LLM’s an ability to actually understand the relationship. 0:14:41.781 –> 0:15:1.301 Brian Haydin Between all these different pieces of information. Once you have both of those two key pieces in place, then we can start to talk about the actual agents themselves. So at the ground level, that’s the base camp, our middle layer on the semantic, that’s kind of like. 0:15:1.301 –> 0:15:17.21 Brian Haydin Having our trail map that shows us, you know, where everything is and how to get to it. And then finally at that top layer, that’s our trail guide, like the guy that we hire, you know, to help us get through to the mountain or get to the top of the summit, et cetera. 0:15:17.381 –> 0:15:37.101 Brian Haydin So let’s take a look at what we’re going to do to get there. So some companies you know are still stuck in like the data in Excel, manual on premise. I do kid you not. We work with a bunch of customers that still have the data, 99% of it on premise. 0:15:37.141 –> 0:15:56.901 Brian Haydin And very, very little in the cloud. But other companies have built out traditional data warehouses that are maybe sitting in the cloud and can leverage some of that. And we’re now starting to see organizations use things like Snowflake and Databricks and I’m actually finding. 0:15:56.981 –> 0:16:14.301 Brian Haydin Very few companies that haven’t invested in in a strategic analytics platform. So most of them are probably sitting in that diagnostic space, but they’re really trying to get up into this wisdom and insight, you know, area and that’s what we’re going to do is teach you how to get there. 0:16:15.501 –> 0:16:34.421 Brian Haydin So as that base camp, Fabric is the Microsoft data analytics platform and it’s built on this foundational layer of One Lake. And you could think of like One Lake as OneDrive for your data. So everything goes in there and. 0:16:34.501 –> 0:16:53.981 Brian Haydin It helps to eliminate the data silos because everybody’s reading from the same from the same map, and so it gives you a really unified view of all the different data that you have, but it allows you to set up separate workspaces for each individual or different product groups or different use cases. 0:16:54.501 –> 0:17:14.421 Brian Haydin To allow them to make modifications to the data that help that are really aligned with their business goals. It’s a unified experience. So what I mean by that is that all the traditional tools that you would use are included in the Microsoft fabric. You’ve got your data movement tools. 0:17:14.501 –> 0:17:34.341 Brian Haydin Your ETL tools. You do all your data engineering. Data science work can be done inside of the Fabric ecosystem as well as like the reporting and the analytics as well. So Power BI is built into it. Another cool part about this about Fabric is the openness of the. 0:17:34.461 –> 0:17:53.221 Brian Haydin The ecosystem. So you might think it being in Azure that you can’t use a WS or you know other non Microsoft centric data data sets and that isn’t the case. Microsoft’s built in this concept of shortcuts that allow you to bring in data. 0:17:53.301 –> 0:18:11.861 Brian Haydin Or leverage data you know from these other systems and so and then they also support like the open formats of of data like you know Parquet files. And then the last but not least that I want to talk about or just call call attention to is that. 0:18:12.301 –> 0:18:31.821 Brian Haydin Microsoft does, you know, take the governance and security layer very, very seriously. It’s at the very, very foundations of, you know, Microsoft Fabric. And what I mean by that is that, you know, as I bring data into the ecosystem, I can tag that data, you know, right from the beginning. 0:18:32.21 –> 0:18:51.941 Brian Haydin And that tagging, that data security, that posture, that governance follows that data from the point that it enters the system all the way through. So matter if I’m building a medallion architecture and I want to transform the data, I I still have the same data protection and security because the data that fed into it. 0:18:52.21 –> 0:19:9.741 Brian Haydin That lineage, you know, brings the data security with it as well. So it’s a very, very important concept. And then it allows you to actually work with the unstructured data inside of the data lake as well. So I can, you know, start to look at what the taxonomy of this data is, how should it be used? 0:19:10.501 –> 0:19:26.501 Brian Haydin What personas should be tied to this and how am I going to tear, you know, different information about it? Next, let’s talk a little bit about those, the trail maps. So we had the base layer for for Fabric, just the Fabric ecosystem in in general. 0:19:27.581 –> 0:19:45.581 Brian Haydin But on top of that, we had to build our trail maps and that was really talking about the semantic layer. And so there’s three different IQs that Microsoft rolled out. Some of this is branding, some of this is is actually new features that are coming out and we’ve got Fabric IQ around your business data. 0:19:45.941 –> 0:20:4.141 Brian Haydin And Foundry IQ was released at Ignite talking about the knowledge retrieval and knowledge IQ. And then work IQ is in the productivity context, you know, bringing it all kind of together for you to use within the Agent 365 ecosystem. 0:20:4.261 –> 0:20:20.61 Brian Haydin So what is Fabric Fabric IQ? So at at its core, it’s really a way for the Fabric ecosystem to build up the ontology and have a shared understanding of what these different entities mean. 0:20:20.101 –> 0:20:37.141 Brian Haydin So your fabric implementation really doesn’t have any understanding of what your entity means, what properties it should have, how it relates to other entities, and so it’s a way for you to do discovery. 0:20:37.981 –> 0:20:57.701 Brian Haydin Inside of the ecosystem and build those entity relationships. Some of it is automated and then other parts of it allow you to be able to work through that yourself if it’s a little bit more complex. And a cool part is that it doesn’t have any additional licensing, it’s just included. 0:20:57.781 –> 0:21:14.421 Brian Haydin Part of your Fabric subscription. The next layer is this knowledge layer and Foundry IQ is a new service that’s coming out that bundles up AI Foundry AI search. 0:21:16.61 –> 0:21:34.181 Brian Haydin And some other resources similar to how you might have built copilot Studio agents that sit over the top of like a SharePoint document set. But it’s since it’s using AI search as a backbone, it’s getting much better semantic understanding of the data itself. 0:21:34.581 –> 0:21:54.381 Brian Haydin And gives you more control over it inside of the Foundry ecosystem to be able to fine-tune the responses as well. So it does do automated, you know, indexing at a much higher level than what you have done with Copilot Studio. But you can also like, you know, connect it with a bunch more data sources that would have been problematic in the Copilot. 0:21:54.501 –> 0:22:9.981 Brian Haydin Pilot Studio area. So Compilot Studio out-of-the-box allows you to do SharePoint documents, some APIs. You’ve got a lot more options when you work inside of Foundry, and then it builds all of that into the same semantic layer as well. 0:22:10.421 –> 0:22:25.261 Brian Haydin And like all the other copilot studios and copilots, you get your intro ID based RBAC security so that for any of the information that it might be trying to query through AI search. 0:22:25.861 –> 0:22:43.421 Brian Haydin That’s bound to your entry ID. So if you don’t have access to that information, it’s not going to answer the questions about it. And then so the third layer that we’re that I wanted to bring up was work IQ is another concept that was brought up and that really is just the organized like the productivity context. 0:22:44.221 –> 0:23:3.381 Brian Haydin So this is, if you think about Microsoft M365 copilot, you know this is connecting your organizational and personal data and teaching it all the different relationships of what that is. So it starts with your work data. So all of the emails, your files, your meetings and your chats that you might have access to. 0:23:3.461 –> 0:23:23.61 Brian Haydin Individually, it brings all that together and does analytics around it to build up the semantic graph or the knowledge graph of it. It maintains memory for you, which is something that is really helpful when you keep asking the same questions. It doesn’t have to go back and try to figure out where all those different pieces. 0:23:23.141 –> 0:23:39.541 Brian Haydin Is the puzzle fits again again, and it learns from your unique habits and your personal preferences and how you typically interact with it. And then the inference aspect of it is combining all that together to build those. 0:23:39.541 –> 0:23:56.901 Brian Haydin More permanent connections so that it can answer faster and more efficiently and effectively the next time you ask the questions. So now we talked about the base layer, we talked about our semantic layer. Let’s talk a little bit about the copilot agents, you know, the agents themselves. 0:23:58.861 –> 0:24:16.221 Brian Haydin And so we’ve got a couple of different ways to look at this. I’m going to split this up into like agents themselves and like the tooling you know that that can help you build this you know as well. So I’m you know agents themselves you can kind of look at it as like I’ve got co-pilots which are. 0:24:17.61 –> 0:24:35.261 Brian Haydin You know, lightweight agents that sit next to me and answer questions. So copilot and Power BI was is a really good example. I can ask, you know, I can ask questions about the data. I can take my natural language and you know, have it transform it into a query that I can run. 0:24:35.781 –> 0:24:53.541 Brian Haydin And you know, just basically sit, you know, next to me to to get me there faster. Same thing with like your M365 fabric, M365 co-pilots like ask questions about like, hey, I missed this meeting, can you tell me what was in it, you know, et cetera. 0:24:54.301 –> 0:25:11.821 Brian Haydin Then we have like actual agents that are going to do things on on behalf of me as well. So I’ve got like operations, autonomous agents, I’ve got virtual agents that can do things, you know, for me and that’s giving it the tooling and the skills, whether that’s through Power Automate. 0:25:12.261 –> 0:25:30.21 Brian Haydin You might make some like workflows that that can you know when I get an e-mail from this individual, catalog this or do this this next step. Or I could say when I get this angry e-mail from a customer, please open up a ticket in ServiceNow for them on my behalf. 0:25:31.901 –> 0:25:49.661 Brian Haydin So co-pilots, you know, typically are just like things that are helping you do the work within the context of that document, that PowerPoint, that e-mail. And then agents will be, you know, things that I’m I I’m more think of of agents as things that actually have tooling that can get things done for you. 0:25:50.181 –> 0:26:7.301 Brian Haydin And then how do I, how do I actually build these things? So in the Microsoft ecosystem, there’s really going to be two avenues that you’re going to go go through. So on the no code side or low code side, Copilot Studio is a great way for you and the organization to get started. 0:26:7.701 –> 0:26:24.181 Brian Haydin It is. It’s it’s no code to a certain degree, some low code when you really get into some sophisticated things. And I I just did a a fantastic talk about, you know, when you hit that boundary of low code, what to do next. 0:26:24.181 –> 0:26:39.941 Brian Haydin But allows you to take, you know your standard knowledge sets like SharePoint. You can connect to Fabric, you can get to some APIs and then readily deploy those to your teams environments in just a single click and maybe you know a. 0:26:40.141 –> 0:26:59.621 Brian Haydin Call to your IT guy to say, hey, can you approve this? But it really safe, you know, user entra, your entra ID bound, you know, kind of chat bots are very quick and easy to build. I mean we’re talking an hour or two hours to get something up and running that can actually get something meaningful done for you in the organization. 0:26:59.621 –> 0:27:19.461 Brian Haydin And then the other tool that we use predominantly is Azure AI Foundry. So this is more of a developer procode kind of an approach with the caveat that the Foundry IQ is more tell me what you want to build aspect but. 0:27:19.541 –> 0:27:39.101 Brian Haydin But once it’s built, you have the ability to go in and kind of fine tune it. So much more customized. You had the ability to use many different LLMS in order to answer the questions and you can build MCP over the top of it pretty easily as well to expose those as. 0:27:39.221 –> 0:27:58.821 Brian Haydin Tools to other, you know, agents that you might be able to build. Not that you can’t do that in Copilot Studio, but a little bit different, a little bit more mature in AI Foundry on that aspect of it as well. And then the memory capabilities are something that would be a differentiator between Copilot Studio and. 0:27:58.901 –> 0:28:14.221 Brian Haydin In the AI Foundry at this point. So, all right, so those are kind of like the how we get things built is by paying attention to those three layers. And so now I’m gonna like walk you through like 3 different kind of scenarios. 0:28:14.741 –> 0:28:31.741 Brian Haydin That you might you want to explore, so things that you probably can do right now so I can build out. First one that we’re going to walk through is like doing like kind of a guide bot like it’s standard like chat bot get insert instant answers to like some internal documentation that I might have. 0:28:32.181 –> 0:28:49.141 Brian Haydin Then I’ll kinda explain a little bit about like rapid analytics with copilot and the last one will be like a proactive kinda agent that like does some other you know does some things on the guide bot. So the the the scenario here is that you know I’m a. 0:28:49.341 –> 0:29:8.581 Brian Haydin Consulting company. I’ve got a whole bunch of internal knowledge, you know, best practices, documentations, SO, PS, things that people don’t necessarily use, you know, all the time. But you know, it always takes a long time to find it. Some of it’s structured, some of it’s unstructured, you know, and I want to be able to like, get access to this information a lot. 0:29:9.181 –> 0:29:28.541 Brian Haydin Quicker. Very common use case. You and your organization might actually be doing it already. The business impact is that it’s gonna save us hours of hunting for documents. And instead of me having to ask three or four different people, like how do I change the spark plug on this? 0:29:28.821 –> 0:29:45.701 Brian Haydin 1970s machinery that we built, you can get the answer in a few seconds. So and then you know it can be built so that it filters only the latest information. So you have multiple revisions, documents that can kind of go through that. 0:29:45.701 –> 0:30:5.301 Brian Haydin And then a lot of use cases, you know, these use cases are tied directly to a customer service department where there might be a high amount of turnover in that that part of the organization and the ramp time to bring people up to SME level is significant. So this gets them up to, you know, up to stuff like. 0:30:5.541 –> 0:30:20.661 Brian Haydin Much, much quicker. So how we’re gonna build this is that we would index a company SharePoint or a teams and we might throw in some layers, you know, you know, in Foundry if we wanted to with, you know, the AI search. 0:30:21.61 –> 0:30:40.461 Brian Haydin Or if we’re using Copilot Studio, we could, you know, we’re just letting it build it how it wants to. And then in Copilot Studio, once you have this this agent built, one of the benefits of using Copilot Studio is like I can really control the topics and the flows and have that semantic understanding. 0:30:40.501 –> 0:30:59.901 Brian Haydin You know, quicker so that I can get people to the part of the agent that I want them to get to faster. That’s really efficient at doing in Copilot Studio. So I might build that conversational aspect of it and then I need to make sure that I have the right user authentication and permission. 0:31:0.541 –> 0:31:18.261 Brian Haydin And aware responses for to control who has access to what kind of data once you’ve like. One thing about Copilot Studio is if you take a document and load it into Copilot Studio directly, it just becomes part of that ecosystem. 0:31:18.741 –> 0:31:35.21 Brian Haydin Versus using SharePoint where it still has that entry ID attached. So you know, kind of being careful about, you know some of those things. You wanna make sure that you’re looking at the security and the permissions and then click, you know, I just say deploy that to my team’s environment. 0:31:35.781 –> 0:31:54.901 Brian Haydin And that’s usually available in the next 10 or 15 minutes for people to add and be able to use. So that’s one scenario. The next scenario is more like, can I get insights into this data? Like how am I gonna do that? And so a couple different tools to really open that up Microsoft. 0:31:55.581 –> 0:32:11.901 Brian Haydin Opened up the Fabric Power BI sort of like copilot experience to all Fabric SKUs early this year, which I think is really, really a fantastic thing. It was when it was originally released, you could only get that with an F64 SKU. 0:32:12.341 –> 0:32:28.421 Brian Haydin Which is like $5000 a month plus your Power BI licenses and all that stuff. So I I think that’s a really, you know, fantastic addition and it doesn’t take a lot to really be able to get some insights, especially with Fabric IQ being built into it and having. 0:32:28.421 –> 0:32:43.981 Brian Haydin You know, a good understanding of standard ontology for the organization, sales orders and sale orders, headers and lines, etcetera. So you can typically get some pretty good answers, you know, out-of-the-box and so. 0:32:44.181 –> 0:33:3.581 Brian Haydin One of the benefits of this is being able to like I’m sitting in a meeting and the question or the topic comes up like how do I handle this particular sales situation and you can get those answers like on the fly. Another tool that we use is something you don’t actually have to build is if you have a. 0:33:3.781 –> 0:33:19.661 Brian Haydin 365 Copilot license. I would strongly encourage you to take a look at the analyst agent that’s available, similar to like the researcher agent. The analyst agent can. It’s really geared. It’s fine-tuned to answer questions about like very specific data oriented questions. 0:33:19.981 –> 0:33:37.261 Brian Haydin So you’ll find that you’ll be able to get good insights out of that as well. So Fabric, copilot, Power BI, Fabric IQ, building out your semantic models, making sure that you have trained Fabric on the ontology. 0:33:37.621 –> 0:33:53.581 Brian Haydin You know, and then you’ll be able to get really good insights. I’ve had several customers just tell me mind blowing stories about the ways that they were able to uncover questions and answers and they’re finding that to be a really effective tool in the in the ecosystem. 0:33:53.941 –> 0:34:13.501 Brian Haydin And then the last scenario they may want to look at is building like a proactive agent. So like a lot of times this comes with like manufacturing organizations where they might have a bunch of IoT sensors and that data is being fed into, you know, some sort of observability platform and I needed to be able to. 0:34:13.741 –> 0:34:30.341 Brian Haydin React to things and let me know not just that there’s hazards ahead, but like help clear the path for that and you know, so that I don’t have to stop or that I don’t have to, you know, fall off, you know, into the water, put my waders on or something like that. 0:34:31.421 –> 0:34:47.701 Brian Haydin So proactive agents really are kind of built around like an like some sort of autonomous loop. So it’s going to detect, you know, things that might actually be happening in the ecosystem. It’s going to build a context around that, like what does that actually mean when I get a reading like this? 0:34:48.341 –> 0:35:7.21 Brian Haydin And so it might go and look at some of the documentation that it’s built out as part of its rag, you know, pipeline. And then it has to decide what it’s going to do. And before it acts, there’s usually some sort of a threshold, like, do I need a human in the loop for this kind of decision or is this something that I’m going to do automatically? 0:35:7.741 –> 0:35:26.221 Brian Haydin And once you have all that kind of like scaffolding put in into place, you kind of let this loose and it it becomes a pretty good agent. Microsoft released one that’s called the SRE Agent. So like a site reliability engineer that you can plug into your Azure ecosystem. 0:35:26.861 –> 0:35:43.861 Brian Haydin And it’ll watch things like App Insights, you know, take a look at like, hey, we did this deployment and two hours later, like I’m having latency issues with the database and it’s causing timeout issues. And so it can automatically say, you know what? 0:35:44.61 –> 0:36:3.381 Brian Haydin We need to do a rollback and it’s gonna do that and then notify, open up a ticket or something like that for the next day and then people can handle that in a less stressful situation. So building these kind of proactive agents is something that doesn’t necessarily take a ton of time, but depending on the. 0:36:3.461 –> 0:36:11.661 Brian Haydin Complexity of what you’re looking at doing. I still think it’s something you can do within like a week, you know, to get some proof of concept up so. 0:36:13.621 –> 0:36:30.741 Brian Haydin Kind of wrapping things up here. So key takeaways here. I want you to think about those three different layers you know that that I laid out. So the unified data platform, Microsoft Fabric using these these new IQ layers. 0:36:30.741 –> 0:36:49.181 Brian Haydin That I talked about, you know, as like your trail maps and then that allowing you to build your AI powered insights, your agents, your co-pilots that are going to sit on top of that after the first of the year, you know I’m going to have had some time to actually play with these Iqs, you know, under the hood. 0:36:49.661 –> 0:37:9.461 Brian Haydin So we’re going to do some deep dives on each one of these. So you know, I want you to follow concurrency, follow me. I’ll be doing some webinars, you know, pretty early in January and February, especially on like the Fabric IQ and what that means. But for you as an organization, what are some things that you could be doing right now? 0:37:9.461 –> 0:37:24.861 Brian Haydin So I would say like, you know, put together like a 30 day, you know, adoption plan like what am I gonna do week one, week 2, week 3 and week 4. And so I would probably start with just identifying a good use case, one that’s going to provide value to the organization. 0:37:25.781 –> 0:37:45.141 Brian Haydin Science experiments are great, but if they aren’t going to teach us anything, then what’s the point in doing them? And more importantly, is the organization going to, you know, give me the licenses or allow me the flexibility to do it? Then I would take it in week two, take a look at like, am I ready to do this? What permissions do I need? What data foundations do I have to clean? 0:37:45.261 –> 0:38:4.741 Brian Haydin Anything. What governance controls should I put in place or what should I think about? And then week three, it’s pretty easy to build a prototype in copilot Studio. Even AI Foundry only takes a couple of hours to kind of get used to some of the mechanisms and build out like a lightweight LM kind of question. 0:38:5.821 –> 0:38:22.61 Brian Haydin In an answer bot. And then last but not least, in the final week, make sure you’re taking a look at like how effective this is. So measure the outcomes, get some feedback from some of the users that you might have using it, and plan on what you what you might want to do as sort of a next step. 0:38:22.501 –> 0:38:42.141 Brian Haydin So that being said, I think that Amy probably dropped a a link in you know in the chat here. So if there are you know any opportunities or any questions that you have, let’s connect and let’s talk about how how concurrency can part with partner with. 0:38:42.181 –> 0:38:58.821 Brian Haydin With you and the organization to build some of these agents to help you gather more AI driven insights out of the data that you already have. So I’ll stick around for a minute or so if anybody has any questions. Otherwise, thanks very much and happy holidays.