Backwards Ontology – Ep. 486

Who owns the ontology? In this episode, Mike and Tommy dig into the practical reality of building Fabric IQ ontology inside organizations—arguing for a “backwards” approach that starts from existing semantic models rather than top-down data governance mandates. They also unpack ChatGPT’s new app marketplace, the accelerating demand for AI in every application, and why the end-of-year consulting rush feels different this time.

News & Announcements

Ontology Overview — Microsoft Learn — Microsoft’s official documentation for the ontology item in Fabric IQ, now in public preview. The ontology lets you define business entities, their properties, and relationships in a machine-readable format that AI agents and Copilot can reason over. It supports generating ontology from existing semantic models—a key capability Mike and Tommy explore in this episode.
Generate Ontology from Concepts — Microsoft Learn — Documentation on auto-generating ontology items from existing data sources, including semantic models. This “backwards” approach—letting the system infer entities from what you’ve already built—lowers the barrier to entry significantly.
ChatGPT App Marketplace Launch — OpenAI launched its app marketplace, allowing developers to integrate services like Booking.com, Canva, and others directly into ChatGPT. Mike sees this as a strategic inflection point—potentially the “beginning of the end” for ChatGPT as a standalone tool, as it evolves into a platform play with monetization and advertising implications.

Main Discussion: Backwards Ontology — Who Builds It?

The AI Demand Wave

Both Mike and Tommy report an accelerating demand for AI integration from clients and their clients’ customers. Every application is expected to have AI capabilities—it’s becoming standard, not a differentiator. Mike notes his embedding work increasingly involves layering AI on top of Power BI embedded solutions.

Who Owns the Ontology?

The central question: who in the organization is responsible for creating and maintaining the Fabric IQ ontology? Mike and Tommy explore several candidates:

BI teams — They already understand semantic models and data relationships, but may lack organizational authority to define cross-departmental concepts
Data governance teams — They have the mandate for definitions and standards, but often lack hands-on technical skills with Fabric
A new “Intelligence Developer” role — Building on their Ep. 487 discussion, someone who bridges technical implementation and business concept management

The Backwards Approach

Rather than starting from scratch with a top-down ontology design (which requires massive organizational alignment), Mike advocates a “backwards” approach:

Start with existing semantic models — You’ve already defined entities, relationships, and measures
Use auto-generation — Fabric IQ can generate ontology from semantic models, inferring entities and relationships
Refine and extend — Clean up the auto-generated ontology, add missing relationships, align definitions
Scale out — Use the initial ontology as a template for broader organizational adoption

This approach is pragmatic: it builds on work already done rather than requiring a multi-month governance initiative before seeing any value.

The Organizational Challenge

Tommy highlights that ontology creation isn’t a technical problem—it’s a people problem. Defining shared concepts requires cross-departmental agreement. When marketing calls something a “customer” and sales calls it an “account,” someone has to mediate. This mediation requires:

Executive sponsorship to break through departmental silos
A facilitator role (the intelligence developer?) who can translate between business language and technical implementation
Iterative refinement rather than big-bang deployment

Practical Advice

Don’t wait for perfection — Start small, iterate, learn
Leverage what you have — Existing semantic models are your ontology foundation
Focus on high-value entities first — Pick the business concepts that would benefit most from AI reasoning
Document ownership — Even informal: who maintains which entities?

Looking Forward

Mike and Tommy plan to continue the ontology series heading into 2026, with deeper dives into implementation patterns and real-world examples. The backwards approach—starting from semantic models and building up—positions organizations to get value quickly without waiting for perfect data governance maturity.

Episode Transcript

Full verbatim transcript — click any timestamp to jump to that moment:

0:00 Good morning and welcome back to the Explicit Measures podcast with Tommy and Mike. Hello everyone and good morning. Good morning, Mike. How you doing?

0:33 I’m doing well. Thank you very much. , just been clipping along here. It’s gotten a little bit warmer up here in our Wisconsin area, so things are starting to melt, which has been great. Going outside is less painful. when you go outside and it’s so cold, Tommy, and your face just like sting hurts. It’s just it’s the worst. But, , things are changing a bit around here up up in our area. Still busy though. The end of the year for me. I don’t know about you, Tommy, but for me, for just work and consulting and everything that’s go, it’s busy. Usually December starts really ramping down and things start slowing down a bit, we’ve had a a good healthy amount of effort

1:07 Put in at the end of the year. Maybe it’s people getting budgets done. Maybe it’s people finishing up whatever they have, , for the end of the year before we get to Christmas. It just feels a bit busier this year. Does that feel the same way to you? , honestly, there’s been I’ve been having a lot of conversations with new people. , something’s got to be rescheduled because the holidays, but there’s definitely that interest. I think it’s all the fabric buzz that people are hearing about that they just want to get their hands on. I I definitely feel like there there’s a wind of that. I’m also hearing something too, Tommy. Anywhere I look now or anytime I’m talking to customers, it

1:40 It’s not necessarily just me, but it’s it’s my customers customers, right? So, I have a client, my client has customers. It feels like everyone is clamoring for any application to have AI in it. It’s like it’s becoming standard. There feels like a momentum here that is, , I’m building things in PowerBI. I’m doing a lot of PowerBI embedding. I’m getting a lot of questions around how do we embed more AI things on top of the embedding space. And I really do feel like there’s there’s a an accelerating need for

2:12 Companies to figure out where does AI fit because I I feel like there’s portions of organizations that are finding, hey, this is not that bad. We’re able to find appropriate places to use AI and the messaging on the internet, like my Twitter feed is just littered with everything. like every other day there’s another like app or dev or I’m constantly taking notes of like all these different products that are starting to exist around AI and how it’s changing workflows, how it’s changing what people build, how fast

2:43 People can build things. those the core concepts that I feel AI is affecting here is it’s removing grunt work out of things you do inside apps. It’s adding capability and it’s doing it faster than you could have done without it. It it’s just oh yeah it’s quite impressive. What do you think? Are you seeing the same? So the biggest thing I think we’re seeing with AI right now is we’re beginning to see the enterprise side like chat CPT just launched a marketplace for apps now because and I think there’s going to be AI is going to

3:17 Work great when it works with other services just like anything. PowerBI’s greatest accomplishment really was the building of the out- of-box connectors. So I think we’re beginning to see that now. Again, the biggest part is the prep. , it’s really like what are you trying to do with it? And then more importantly, , it’s just going through in terms of like how what problems are you needing to get solved? This is interesting, Tommy. I didn’t really understand what this apps in chat GPT. What does this mean to you? Have

3:49 You have you played with apps in chat GP? I know you’re a big chat GPT user in general or AI user in general. What are you what are you finding in apps? What’s what makes it Yeah. Well, and when I say it’s true, it launched yesterday, so like it’s really out yet. No, I what you build an app yet? I feel a little bad that I haven’t bought everything yet on the store because that’s usually what happens and something new comes out and I buy it and I have to try it out. But no, it’s just really the ability honestly they it’s just being able to integrate

4:22 With other services. , and they’re right right now they’re very much still in the we’re calling everyone to submit developers submit to the our app store phase. So very early in the game. , let me let me spin out some app people here that I think is very relevant to call out here. One one of them has been like so chatpt is working with booking.com right? So let me let me just say this. This is this is the beginning of the end Tommy is what I see this is for chatpt. So what so I’m listening.

4:54 The reason I say that is because the allowing you to open up apps inside chat GPT is giving you the ability to tie existing monetary like these are apps that have money like these these are like other existing financial systems that have are now going to start using chat GPG to make their stuff better which is great. totally acceptable but like booking.com right they’re making money some way and somehow chatp is enhancing that and now chatp can make money or advertise using along those lines canva right again

5:27 Another program you pay for corsera learning things another travel site Figma building design websites Spotify like Zillow these are big companies a lot of them are geared maybe around search or providing services to you about, , these different apps. I think this is the way that that chat GPT is going to start trying to monetize on certain things. So, does this does this mean like Spotify subscription goes to Chat GPT? Like where where’s the blend here?

6:00 I guess what I’m trying to figure out is what is Chat GPT’s ultimate monetization pattern here? And I don’t see anything else other than Chat GPT has to start advertising at some form. like they’ve monetization. Well, it’s going to have to happen, right? People want the thing for free all the time. And where do you get the money from? You get it from ads. So, where does that So, how does that going to display out in a chat GPT world, right? I I don’t This is where I’m trying to like struggle to figure out like how they’re going to like eventually turn their business around. They probably will never make money

6:33 Just because the cost of GPUs alone. And what’s going to happen, Mike? It’s going to happen eight years from now when they don’t have to be at this rapid rate of innovation. I was listening to something the other day and they were saying when you look at the iPhone or the phone right now Sure. we’ve made optimized it Yeah. because you really can’t do much more to a phone in terms of what it can do the features in terms of the size

7:05 The camera. So you’re not going to see like in the beginning every year like oh my gosh they added a camera here what now it’s it’s a retina display and so there’s always those rapid innovations which costs a lot of money to do and right now who the heck knows Pandora’s box of what the that looks like for AI but dude I don’t know if they’re ever going to make money but here’s the thing advertising I I’ll bet you a stake on that I’ll if If Chad TPTt starts doing

7:37 Advertising, I’ll buy you a steak. I have no problem doing that. Well, I don’t think we need to take bets on this one. I think it’s just inevitable that that’s that’s the direction they’re going to go at some point. It’s going to be some shopping thing. It’s going to be asking me for outfits. It’s going to be it’s going to be some integration that’s going to start advertising things for you directly inside the application. And so, chat GPT was is going to accelerate, , not turnover, it’s what’s it called? conversion rates, right? Hey, if you use your normal website and search the internet, your conversion rate is what, ,

8:08 3, four, 5%. Right? Use chat GPT with your with your program or app and your conversion rate is 15 to 25%. Right? All they’re going to do is just give better recommendations, I think, at some level. And that’s the reason I say this is is interesting to me is because I think this is a direct competitor to what’s happening in other spaces like Google or Microsoft or other spaces here that are already in the advertising space. Bing, I guess, for that matter as well. , I don’t know how big the other players are in the space, but if you if you look at that, I think that’s where you’re going

8:40 To see a lot of competition. And I think it’s also in the best interest of Google to supply all this AI information at the lowest cost possible. Gemini 3. I keep seeing people raving about it and just ripping and ripping chat JPT apart just because it’s like so much better performing so much better. It’s being given away for free. People are really liking it. It can do crazy 3D graphics and games and all kinds of other things that Chat GPT is not able to handle here. I don’t know if you saw this Tommy or the article. I thought

9:13 I heard or read an article that said Sam Alman, the president of Chat GPT wrote like a letter, a memo internally was like, “Hey, we got to really rally the horses here. People are passing us. We need to stay competitive. They have to redeote re reallocate resources to to produce some stuff that they can get out the door.” So, I think Chat GPT is under the fire here. And it’s this is good. I think this is good for the market. It’s good for competition. Yeah. Yeah. Gemini’s update too is you can create a web app that on

9:48 The fly right now in an incredibly easy way and like people are using that to do like their wedding invitations and you’re registering. So it actually like hosts something too which is insane. It it’s like a but this is what I was talking about earlier. Remember months ago, Tommy, I said, “We’re going to get to a place where you can just talk to it and you can just it’s going to be a throwaway thing.” Like building an entire website from scratch by talking to it about a wedding invitation. That’s a one-time event. Well, maybe two times event if you if you have if you have two married, but typically it’s like a it’s a one big event thing and here’s a URL, give it to

10:22 People, here you go. But after that, you’re done. It’s like the the event is over. You don’t need those invitations to hang around on a website forever. So, I think we’re going to get a lot of these new experiences. We’re going to start seeing new businesses pop up that you’re going to and build really rich, engaging experiences on the web, but it’s going to be a one-time deal. Build it once and then get rid of it, dude. Yeah, it’s going to be awesome, dude. All right. Well, hey, I think we got a ton of things to cover already. We do. Let me just quickly give the main topic and then we’ll go back to some news items. How about that, Tommy? All right, that sounds perfect. So today

10:55 We we’ve been talking about this thing called the ontology. This is something from Fabric IQ. And an ontology helps you visualize or or put together different relationships between tables and data and objects and things that make your business work. So ontology is this is this new item in fabric. You can go build it today. It’s inside fabric. You have to turn it on in the admin portal. So there’s an admin setting that you have to toggle on for your organization. And then you can make a new item in the workspace called an ontology. So as I was playing with this ontology experience, Tommy, it seems

11:28 Very new. There’s not a lot of buttons in it. It’s pretty straightforward. [laughter] Yeah, it’s pretty bare bones. And one of the things that’s not very clear to me, Tommy, and and look at this ontology thing is I don’t understand how the ontology item works. And where do you attach it to data agents? That’s where I didn’t quite understand like in the in the messaging of like the documentation. and it seems like they’re talking about it. I haven’t quite figured out where to add it to the data agent. Maybe it’s the alentology item is just added as a reference document to the data agent itself. I got to go look

12:01 At that a little bit more. That’s one thing I haven’t quite figured out yet. as I was thinking about this one, I came up with this idea and I think we talked about this very briefly, Tommy, was this idea of the backwards ontology. That’s going to be our main topic today. So, we’ll get into that in just a bit. Let’s go into some news. Tommy, what did you find for news articles here? All right, we got a few good ones and one I’m I really can’t wait to get your thoughts on. So, the first one is something new to dataf flows gen 2. From simple prompts to complex insights, AI spans the boundaries of data transformation.

12:33 Again, I can guarantee you that title was made by AI. There’s no question about that. but that’s for another another conversation. So what do we got here is we actually have here a AI powered automation that for using natural language to do transformations on a new column. this is crazy at least the preview of it because what it allows you to do is simply choose a column and that feature in Power Query where

13:06 It’s like you create a column from examples. Yes. So that’s just basically what you want the desired output to be. Same idea, but except instead of you providing the what you want the desired result to be, you just put a natural query prompt in there and what you want to happen, which is crazy if it worked. so it’s really preview and I looked through the limitation. It looks like as there’s it’s not like a F six. It probably is going to be

13:38 Something like that, but I didn’t see anything in terms of who is it available for. You do need to be using a dataf flow gen 2 with CI/CD support. And besides that, it’s not co-pilot either. So, this is completely separate of co copilot. Mike, the idea of this is amazing. However, they are opening Pandora’s box and I was I immediately tried a few tests yet yesterday. Yep. Everything aired out with the worst error messages saying, “Look, we don’t

14:11 Know what happened.” So, it’s in preview, right? It’s in like Yeah, it’s it’s in like early early preview alpha alpha a right. yeah, I don’t This is a neat feature, Tommy. I think it’s good. people do want to chitchat with their data and transform it in interesting ways. , is this a feature? It’s rolling out slowly to all organizations. So, this is something I I do want to point out. Today is the day. It should be out in all regions. , but it may take some time to get there. Is this in only in the service or is this also in desktop

14:45 That is this is just dataf flows gen 2 cicd? So, it has to be in the service. It’s got to have to be the service based things. Okay. Right. Yeah. I think this feels to me I’m just giving my my gut feel on this one. This feels to me we’re just trying to throw AI at everything now for whatever reason. Just to make sure it’s there. Did the column by examples work for me? Actually, not too bad. I I felt like the column by examples was actually a pretty good feature all of itself and it did a good job of finding patterns and things and

15:17 It was pretty decent. I’m not sure why we needed to replace that with this other co-pilot agent thing. I would disagree in terms of if there’s ever a easy use case for AI, it would be with the, , hey, I don’t have to only worry about certain patterns. I actually need to clean, , the code here or hey, help me find everyone’s middle name, which is a lot harder to do with the combo. I I see what you’re saying, but this is more to me that I would see have a logical place to put AI

15:53 Yes, but I’m also I guess my argument here is why is that like most of the things that people are doing in Power Core like what is this going to do above and beyond pattern searching? Like what is it going to do? And so like that and that’s having just a button with AI that’s there like what are we trying to get it to work on? Like is it is it only focusing on just adding new columns? I feel like most of that’s common stuff. I don’t know. May I maybe I’m underestimating what it can do. it we’ll see. I I’m going to play with

16:27 It a bit more. I don’t have a major opinion on it. But one thing I will admit to you, Tommy, whatever it is, doesn’t matter. It’s got to work. Like, you can’t have it just send out very random error messages. This is this is one of my cases of things that seems to be happening fairly frequently with with Microsoft bringing out new features. a feature gets released, it’s in preview, it’s potentially too early or the error messaging isn’t very clear. Again, preview, great. We understand that it’s not really baked yet. Okay, but those first three or four interactions you’re going to have with a customer or user, if it doesn’t work,

17:00 People won’t touch it again for 6 months, long time. So, you’ve got to have it at least do something. It’s got to at least work to some degree and attempt even if it is wrong, it’s got to do something. Yeah. So, I don’t know. I I I really feel discouraged when we see features come out and they just error out immediately. The other thing I’m thinking about too because the cool thing is you can go over multiple columns. So it’s I think it’s going to be a lot more than just pattern but does once you create the prompt and it generates

17:32 Does it now have a certain like power query code that it’s going to do every time or does it evaluate for every row every time the thing refreshes? Right? So then your results could change every time your data is refreshed, right? What is So another question for you then around this then to make sure I understand when you use the AI column feature, right? Does it just make m code or is it doing something else? Nope. it and there may be m code on the back end

18:03 But what you see in the power queries little yeah the events editor and the formula bar is you you see prompt and columns that it goes over it’s not like the you select from examples where it would generate power query or you can see it it’s just literally showing you a prompt because but again let’s say you do something different modify it does then the results change because it’s evaluating it a different way. H interesting. Yeah, that’s going to be

18:36 A problem. I am I am not going to [laughter] I am not going to run want to run Power Query and have it load things and have it give me different results. Yeah, that’s going to be interesting. I don’t know how that would work. Now, if it was working at all, then I would I could tell you that experiment. However, yeah, we tried some very basic things on some very basic tool sets or data sets and still no go. H interesting. Okay. Well, let’s see how that turns out. Yeah. So, we’ll we’ll we’ll keep update

19:10 On that. But Mike, we got a few other things as well on the board. We have What else do I have? Let’s see. let’s do we’ll do it in order. I think that makes sense. Fabric runtime 2.0. This came out three days ago and what this is is simply looking at the now an experiment public preview which simply is built for largecale data computations which allows you to have a runtime that’s Apache Spark 4 Azure Linux 3 Java 21 and Delta Lake 4.0. , so it’s really just

19:46 About being able to go through and change your Spark settings and just try the newest and greatest things. So it is public preview. Yep. So experimental preview. So we’re not even in just preview, Mike. It’s we have to now say it’s a preview that’s also we’re highly experimental. this is what this is what happens. , Microsoft is usually behind the the industry when it comes to like these other Spark related items, but this is the stuff that they need to get to, right? Handling newer versions of Java, newer versions of Scala, Python, and Delta. This is great.

20:18 I love this. This is this is exactly what they should be doing. I this is a very expected thing. I’m very excited to try this out because I think this is going to be very useful for people. One of the things I observed when I work in data bricks, I do a lot of data bricks work as well is in addition to fabric when you work in data bricks, data bricks is constantly versioning themselves and every time they have like major releases. So up until this point we had like version 1.2 1.3. This is a 2.0 version. This is like a major

20:50 Version that Microsoft is releasing. typically that means you’re going to have like long-term support with the 2.0 version, whatever that means that they’re they’re building in there. Again, I don’t know their back end very well, but th these movements having these large development excite cycles, I guess I would call them, usually results in a lot better performance. Every time I’ve done this with data bricks, when they’ve made like major revisions and done some really big things, stuff got faster, tables got more efficient, I was able to, , use less compute but

21:23 Do more work. I I’m very encouraged by this and this is this is something I want to see more frequently is more versions of these different run times. this is also one of the reasons why I’ve I’ve gone down this route of try to use standard out of the box Python and Spark, right? So, you don’t want to do a bunch of like really crazy extra connectors and a lot of extra custom add-ins because when you use custom wheels or bring your custom code packages into

21:55 This Spark environment, you have to make sure that every new environment you put in place also works with the old packages. And this is something that I found that got burned on a couple times, which was we needed to connect to like a SQL server and we had this custom package that Microsoft was maintaining. Well, then they decided Microsoft said, “Look, we’re no longer going to maintain it.” And what that did is it stuck our data bricks environment into a certain version that would work with the package. If we went any newer, it wouldn’t work. So we had to slightly shift our architecture so that we weren’t doing, , Spark. We we didn’t like Spark directly connecting to

22:28 A SQL database. Instead, we we liked Spark reading and writing data directly from the lakehouse. So there were some patterns that we changed based on not using custom wheels or custom packages inside Spark. So that’s just one thing to be mindful of as you’re updating these things. But I’m very encouraged by better run times and faster run times. I’m excited about that. As always, make sure you look at the limitations, Mike, because any preview should always give you pause, but experimental well you you can not install library. You can install

23:02 Libraries directly in your code. you can set the spark settings. you can write to a lakehouse with Delta Lake 4.0. 0 but order native parquet writing those things are not supported features such in data science are not supported and and integrations with VS code are not supported as well that’s a good call out Tommy there are some things here that you’re getting by default so again a lot of this is when you get to this stage let me say another

23:33 Thing I’m as an engineer of this right this is where my mechanical engineering or my testing engineering hat comes on right this is a lot I have a pipeline. I have a notebook that’s running right now on Python or run fabric runtime 1.3, right? So that’s something that’s been running for a while. As soon as 2.0 shows up, I’m always interested in like let’s try that same notebook. Let’s run the notebook again. What does the performance look like? and what you just called out there and some of the limitations. these features such as native parquet writing,

24:06 Autoco compaction, yeah, optimized, right? VO order, these are features that may have existed before and just you weren’t even aware of them, right? So having those things removed, you may find that there’s different performance when this using experimental. So just be aware of that. You may need to run the experimental. You may need to wait for Microsoft to figure out these other features that they’re trying to add in here and make sure that that vauorder and auto compat taking a place. Some of those are pretty integral to to your existing u notebooks or

24:40 Integration. So Mike, I’m really more and more especially in the fabric environment. I know it’s hard to do but preview features especially things that are getting released run it in a different sandbox because you can break things very easily. It’s one thing to try an open source GitHub repo on and create a virtual like your own environment there to test things out, but like the the vauorder is something that I know is a must. And if you have things that are production and you’re like, what, I really got to try this on latest and breaking.

25:12 Yeah, dude, you you can really break things. All right, let’s get into our last news topic here and we’ll get into our main ontology portion here as well. What’s your last topic you want to look at? So, we’re going to actually go to Gartner because we love Gartner and we love breaking it down. So, Microsoft is recognized as a leader in the 2020 five Gartner Magic Quadrant for data integration tools. So, this is not your normal PowerBI business intelligence one. I probably want to check the PowerBI blog for that. But Microsoft is

25:47 I this is the first time I think Gartner has actually looked at Microsoft fabric. I don’t remember another Magic Quadrant at least around the data integration and they’re up against AWS, Oracle, Informatica, Google. and still they’re a visionary and a leader. So if you don’t like Fabric, well tough cookies because it’s one of the best ones out there. Yeah, I like so one of the ones that I think is where Microsoft is stomping on some existing playground is Informatica.

26:22 I hear a lot about Informatica. It’s quite big. I’ve also heard Informatica is quite expensive for data integration services of things. , and to be honest, Tommy, I think there’s a movement here. I think Microsoft is very much trying to build provide a all-in-one software as a service portal web portal to build all these integrations across data pieces. I think this makes total sense. You have multiple different ways you can store data. You can bring in real-time data. You can store things in lakehouse. You have Spark. You have SQL. It’s like a Swiss Army knife of data stuff like and

26:56 To be clear a lot of this stuff existed inside fabric or sorry inside Azure before this. It just wasn’t as it just wasn’t as integrated as it is now. And I feel like the integration now is just way better. Yeah. I’m going to dive into the article because Gartner what they always do too unlike I remember we talked about the one of the news articles. It was like the not the magic quadrant but it was like the circle whatever that weird measurement rating that were like I don’t know how they’re rating these products. this one makes a lot more sense and Gardner actually does a great job of breaking down why they

27:30 Rated someone or measured someone a certain way so they get like the pros and cons. So I’m intrigued what they’re going to say with fabric that it’s doing so well and compared to the everyone else like where it needs to catch up. Yeah, I’m going to go to the art. So one thing I’ll just call out here as well for those who are making decisions around fabric or not fabric, right? the chat here was talking about fabric is fabric good enough to start replacing data bricks for your new projects. I maybe would argue if you don’t have data bricks and you’re deciding against data bricks and powerbi

28:02 It’s a much better choice to start with powerbi and not introduce data bricks but if you already have data bricks or trying to replace it I think I don’t think I’m there to replace data bricks yet at this point but this being said data integration is very much in line with this conversation. Are you doing data integration with data bricks? you can do it with just fabric. And I think this article points out very clearly that Microsoft is is in this space here. , and I’ I would highly recommend read through the article here. And so go to the Microsoft section and look

28:35 At what they’ve got. So I I just I just got that up. So Mike, if you were to think of what a company like Gartner would look at as strengths for fabric that’s now been out for two years, I’m curious what you think it would be. But the strengths that they outlined is a broad data management vision which I completely agree with. Microsoft vision for fabric as a converg data ma management platform is a key different key distinction of it. stable growth they’re have strong market momentum and then strong

29:08 Streaming capabilities. So wow there that just means more real time updates by Microsoft. It dedicate features a dedicated module called real-time intelligence and is a way that’s user friendly as reported by Gartner clients. I I would argue here Tommy I definitely feel the the messaging for Microsoft being streaming and real-time information becoming a very strong point for Microsoft and I’ll even note this it’s streaming and not in [snorts] high volumes lots of streaming data. I’m

29:42 Going to also argue around this idea of like event driven actions, right? being able to have information come in, a file lands, and then when the file lands inside the lakehouse, you can then detect an event and then process the data from there. So while we’re saying streaming capabilities, I think the first thing many people go to is streaming is a high volume, lots of data points coming into fabric, which it can handle, which that’s part of what they’re doing. But I also would argue it’s also this real time eventing based processing of data

30:16 Which I think is making a very good waves here as well. So I would agree with these three strengths. I do see stable growth. but I I will really lean on some of the precautions here. And the first one I think is also my message here as well especially when we’re talking like the difference between data bricks and fabric. The product is still maturing. Microsoft Fabric and Microsoft Data Factory and Microsoft Fabric is a relatively new offering even though it’s built on pipelines previously from Azure, but they’re still developing out continuous integration, continuous delivery. CI/CD is still being worked on. It’s not quite

30:49 Fully fleshed out. It’s not quite as smooth as you would like. And some companies like to use DBT in conjunction with fabric. So, yeah, it’s it’s getting better, but it’s not probably as smooth as we would like. yeah, the other one I will call here is they have limited on premise capabilities. And this is clear. I don’t think that I don’t think Microsoft’s ever going to get top marks in on premise capabilities with fabric. Their whole goal is to move your data into the cloud. It , you don’t want to have SQL servers on prem anymore. You want to move that to an Azure SQL server or bring the SQL server

31:22 Directly to fabric and have it work there and then everything works a lot more seamless and a lot easier to work with. Yeah. And I I think Microsoft’s looking at that going okay if you’re going to point out our cons is on premise fine. So we’ll take that. One of the So the other one here too so Microsoft is in the upper right hand corner which is a good sign. That means they’re they’re leading in vision and they have ability to execute. The other company that’s out there that’s just a little bit underneath of them on shoot I can’t see the graph here. Let me go back here. The

31:54 Other one that’s a little bit lower than them on ability to execute is Informatica. But for completeness of vision, Informatica wins, right? So they’re higher on the completeness of vision there. And when you go back to the Gartner report and look specifically at Informatica, again, I hear about them. I don’t directly work with a lot of Informatic Prodica, a lot of Informatic Informatica products, but I do when I hear things, this makes sense to me, right? Informatica is leading very strong in the AI and Gen AI capabilities. They’re

32:27 Trying to bring a lot of LLMs there. This is an area that I think Microsoft is trying to get in more, but they’ve they’ve watered down AI by using this co-pilot branding. And so instead of just directly integrating with models that exist that you can run, it’s not as seamless to me. And so to me, this is an area that Microsoft should do better in in the AI integration with existing items in fabric. It doesn’t need to be always whitewashed with this co-pilot experience. Okay, fine. It’s co-pilot, but I get co-pilot and VS Code and I can

33:02 Pick what model I want, right? Like why can’t why am I not able to pick models across the Microsoft AI things that they’re doing? Like I get it. Some of it they don’t want you to, but anytime you’re bringing a little chat window on the right hand side open, like let me pick what I want. So I I think I think that’s a good a very fair assessment there. And I’d also agree Informatica has a does seem to have a very broad integration system, a lot of partnerships, right? It works with everything. It bolts on to anything you want. Microsoft, Data Bricks, Oracle, Snowflake, it’s all everything’s

33:35 There. AWS, you can integrate Informatica into all the systems, which makes sense. that’s what they’re they’re known for. So, it’ll be interesting to see where this is going to play and where Microsoft is going to continue to push themselves in this space, but it’s encouraging to see. I like attaching my learning to products that are winning, right? Yeah. To me, fabric is a winning product. It’s growing. It’s gaining adoption. It’s getting high marks on Gartner. It’s

34:07 Getting high marks with reporting. is getting high marks in data engineering. I like that. I like being on that winning team side of things. So, I I feel very comfortable tying more of my career into the fabric and data engineering space because I think Microsoft’s doing a good job here. I completely agree. All right, dude. [clears throat] Ton of good things going on. But speaking of new experimental and innovation, Mike, I think it’s time we talk about ontology a little more. Yeah, let’s go into some of that stuff. I’m not sure where to really go from

34:40 This one. Do we need to do like a little overview of what ontology is? I guess maybe that we should probably start there first. Yeah. So our ontology, what is ontology? It’s really ontology, it’s part of this new feature or product called Fabric IQ that was recently announced by Microsoft. And ontology is simply a new item that can be created in the fabric environment. And it’s simply a business definition or fabric IQ ontology simply creates a governed semantic layer with what’s they’re

35:14 Calling reusable entity types like customers and orders and the properties and the relationships that bind them together. So think closely to a normal semantic model. However, this is that married with Microsoft Graph in terms of it’s not just a technical, , unique columns is simply the the human understanding of how things are binded or related to one another and also their definitions. then it helps standardize business vocabulary

35:46 For AI quering and PowerBI consistency. consistency where the ownership is, , usually rests with the creator of that. And again, ontology background, the definition, we’ve said it a few times already on the podcast. there’s a from a philosophical point of view, is simply what makes a thing a thing. Like what makes a chair chair? It’s the the metaphysical. A more structured way of looking at it is simply how something can be, , what are something’s capabilities and how is it

36:19 Confined? what’s that’s purpose and in terms of really what is that definition of a thing and that’s more of we’ll call it the business way of looking at the word ontology so ontology is something I can create I can use existing data I can use a semantic model but I think one of the things we said on a previous podcast Mike is consider our PowerBI semantic models as a 2D painting that we’ve been doing creating our Mona Lisas. What fabric ontology is is making that 3D.

36:56 It’s simply a third dimen or you got another dimension to what a semantic model or semantic model does. Yeah. , yeah, I’m going to slow down here a little bit on ontology pieces, right? I I I I I think it’s when I look at the word ontology, it’s basically just relationships between things is how I see it, right? You’re already building an ontology inside your semantic model. And my mental model of this one is we have dimensions, we have

37:29 Factual tables, right? That’s that’s some level of an ontology. When I go into the model view, it’s showing me the relationships between my data tables. When I when I think about this concept of ontology for my whole business, I I my mental model goes to I could have four, five or six models that describe things about my sales team. So I have sales data, right? And inside those things, you can go even deeper and say like what products do we sell?

38:02 How often are things being sold? What are the sales channels? there’s all this information that relates to that that product area. So I think really what’s the ontology piece here is when you step back we have a lot of knowledge in our head on how things relate together. This is a way of representing information for AI agents or things to help them bind themselves to different data elements that you’re across your your business. And so the example that Microsoft gives I think makes a lot of sense. It talks about like airplanes, airports, and customers. And so you step

38:36 Back and just say, how do these different objects relate? And I know in working with airlines, there’s a lot of really rich data that comes from multiple things. , for example, if we’re just talking the airline example, we have on the plane, just the plane itself generates gigabytes of data every day. just sensors and maintenance records and all these other things. So the plane itself has a whole bunch of data points attached to the plane. Then people who

39:09 Are sitting in those seats, every person sitting in a seat has a whole bunch of data with them. when did they buy the ticket? are they flying with their family? Is this a business or pleasure trip? Like all these different other options here. And then you want to track what did that user buy throughout the course of the trip. Are they always buying the the premium seats? Are they always buying business class? What does that look like? And so to have a better picture of like knowing your customer and what they do, there’s all these data points and they all they all relate together, right? It’s the the

39:42 Whole idea of the business is these are related data points that link together at some level. And that’s how I think about the ontology. I’m thinking about ontology is this is a way of describing relationships between your data. And so I’ll just pause right there. What do you think? talking about that. I don’t hear you. , we’re good. We’re good. Not a technical thing, a Tommy thing. No, I I agree with you. And I and I think we’ve definitely really u when you think about it, you assume that something of

40:17 This nature would be owned by leadership and from business leaders or the leaders at an organization right because you’re why do you assume that the initial assumption let me talk let I’m I’m working on it. So the in natural assumption to me would be okay if you are defining for example with the airplane example well if you’re going to define those order those are things that just like leadership does define usually what the targets

40:50 Are who to hire these are it’s the core crux of that is making sure that we again have aligned definitions aligned understanding of in a sense what things are where is gas something we’re buying now or in in the future, etc. So, if you’re going to have those definitions and how they’re related to each other, you would assume that this would be something that would be coming from higher up because that’s going to again it trickles to every part of an organization.

41:24 Does it and I but I think the question really is because of how they’re how Microsoft and how Microsoft works with data does it demand my question though is as I think about it more is does ontology demand an executive or domain subject matter expert? probably because you have to again understand the entity and the relationship accuracy. making sure that we have a single definition for things like our example of leads can be different in marketing

41:56 And sales and ensuring that those definitions go across the business. we have this new feature, Mike, in ontology where I can actually be in my semantic model in the semantic model homepage and there’s a button that says create ontology from semantic model. Autocreate. Yes. and this is pretty crazy because this really allows ontologies to be create generated automatically. understand

42:30 That tables are entity types, columns are properties, relationships, etc. and at least enables teams to prototype and begin to scale upward. So my question for you Mike is how how successful do you think it could be? Because unlike I created a data set that just begins to grow with maturity when you see ontology, do you see this is something that needs to be initiated by leadership or is this something that in a sense you want anyone to take

43:04 Organically? So, let me let me go I’m going to go to that question first and then I’m going to maybe dive in a little bit more of like how do you build an ontology from a semantic model because that’s something I think a bit more interesting to me and that’s more where my idea comes from like this idea of like the backwards ontology. So, let me first address your question. Who needs to own this stuff? We don’t know yet Tommy. It’s all in preview. I don’t I don’t think this has to be centrally managed. I don’t think this has to be centrally led. , the way I see ontology working out right now, especially starting from the semantic level up, right? If you’re

43:39 Starting at a single semantic model, that’s that’s a very narrowed window of data that you existing organization. Your organization is going to have thousands of semantic models at some point, right? Tons of models across the organization. The ontology part is just looking at one of those models. Very rarely do I recommend to a company to build out a monolithic semantic model that has every table, all the relationships, all the data into one large semantic model. A lot of times we

44:11 Work with customers and we talk about domain specific, right? What are what does this model need to have in it in order to service the reports for that model? And an organization needs potentially multiple domains and even departments within a particular department, the finance department, the sales department or the marketing department. They’re probably going to need multiple semantic models to describe their entire domain of stuff. So even there, so I think of it as these models become like little

44:44 Data libraries for what’s what’s happening with these organizations. So that being said, when I look at what ontology is doing, this is where the concept I think of the reverse ontology comes from, right? I have all these different semantic models. I’ve spent time building relationships between tables. But what I don’t know is how many times have I used dim product? Is there product master? Where was that used across different semantic models or other systems? How similar is that table to like so for

45:18 Example if I have a lakehouse table that’s master products how many other teams have used and consumed that have they taken that shortcut from that table and transformed the data and done something different with it so there’s some lineage there that goes with that table and I feel like that’s something that’s part of the ontology right an ontology would describe here’s our master product data and here is derivatives of that work that maybe be used in other semantic models so when I look at ontology across whole organization. Maybe we have the product the product master table, but

45:50 There may be multiple semantic models, multiple data tables that exist that support that. So, this is where where I’m thinking from like we’re already spending time, we’re already building these really rich integrations across our data layers, but they live in the semantic model. They’re stuck there. And so what I’d like to see is in the in the ontology piece is like bringing more things of the ontology back up across multiple semantic models. That that seems to make sense to me. Multiple semantic model. Yeah. So okay layer there. Yeah.

46:24 So let me add just one note here as well. Have you done have you gone through and have you done a reverse well call it reverse an ontology? Right. I’ve made one ontology from scratch. Mhm. It’s difficult to be honest. Yeah, it’s not fun at all. It’s a lot of clicking. You have to add like every single detail. , you can make the item basically like like the object, right? This is country or whatever that is, but you have to add all these different like properties and bindings and every column can become a property. , and there’s just a lot of

46:59 Extra things that can be added to the the ontology graph here. Yes. Interesting. useful, but like I look at it going, “Oh my goodness, this is not like yes, it adds some layer of capability here.” , but it it’s confusing in if you start from scratch, it’s like what are we doing here? It doesn’t make sense to me. Does that make sense what I’m describing? Oh, 100%. Because you and to your point, Mike, when things are demo or preview, they are preview for Microsoft because this is bare bones. like you said the

47:33 Whole like that top ribbon what is there one button. there’s like three buttons maybe. Yeah. When you start you click on something if you click on something you get two more buttons you and that’s that’s about it. And in terms of creating that link between the relationships. There’s really no in a sense if you don’t know what you’re doing or the purpose understanding there as well because it’s just talking about those empty like yeah it’s weird in terms of those bindings how things are binded together yada yada yada but

48:06 The it’s really hard to start but at least when you’re dealing with a semantic model again you already get the table everything’s already basic things like date the name of the tables you don’t have to pick the it would be excruciating almost Mike to start doing this by from scratch. to your point if let’s say you a team was like hey we’re going to build our ontology layer and we’re going to do everything manual that’s a tedious effort however Mike the the thing is I’m not

48:43 Sure to and to your point there are multiple scenarios here where a single semantic model is not going to cover what an ontology layer would need to Right? Because again the purpose of a semantic model, right, is is for from the business logic definitions for a reporting solution. It’s so it’s not going to necessarily cover or filter, , include everything. But ontology, well, you’re gonna want more or less everything about the the, , anything that has to

49:16 Related to do with a plane. You’re going to want that graph that those binding those links from vendors, etc. Things that you would not normally need in a dismantic model. And that’s where I think there’s an additional layer of complexity. But I don’t want to get too ahead of ourselves because let’s just take what we got today and just that hey how can I get started type of attitude does the semantic model feature and right now I think it’s that or bust personally like is that a kickstarter for you for

49:50 Organizations or for teams to go let’s just try this out or is that just a need to have cool feature which where does that align for you what you’re seeing so My initial impression is if you start from ontology the the item itself verbatim just the way it is without any without any starting it’s confusing. I don’t know what it’s there for. You don’t really understand like how the relationships work. how do you add relationships here? When things start making sense to me is when I start seeing it using the

50:22 Semantic model to build out the ontology. There’s like some scripting that Microsoft’s got behind the scenes here where they are using like the relationships are the relationships in the semantic model are bound to the relationships inside the ontology. Right? So for example I have a very simple star diagram. It’s a star model. I understand what star diagrams are doing. You have a fact table and you have a handful of dimensions that filter the singular fact table. And in this example, when I load this semantic model directly into the ontology, it says, “Oh, here’s my fact table. Fact sales from the fact table, it has the fact

50:58 Table has dimensions. Here’s the relationship to the dimension. Here’s the relationship to the dimension for date. Here’s the relationship for the dimension to dimension for segments. And when I look at this in my fact table, I have a handful of measures. So each measure becomes a property of the fact table. Makes sense. That’s is what I would have initially have viewed inside the PowerBI semantic model. But there’s this other thing called bindings. And I don’t really understand where that fits. What’s the purpose of

51:30 It? How does a binding get data in? So it looks like bindings are trying to attach an actual data source to bind factual data from like a lakehouse or some other source. to to bind things. There’s also this idea of a entity key type or an instance display name. So there’s always other extra things that are in here that I wish it was a little bit more useful. one other thing that’s a bit of a complaint here for myself is when you’re creating this entity inside the ontology, the name of

52:04 Everything here has to have underscores in it. Even though my semantic model has a proper name for the the measure, , sum of sales with spaces in it, for whatever reason, the ontology thing makes you have underscores. Okay, , I get it, but like I don’t the reason of having an ontology is to make it human readable. Like, yeah, the reason it exists is to make it understandable to to individuals that are looking at this and exploring the graph of this. So why are we forcing ourselves to have

52:36 Something with no spaces in the properties of these different objects? It seems like self-defeating. So yeah, okay, fine. You need to have something the item may need to be named some of sales with underscores, but there needs to be like a pretty name of it. Like what is what is the actual name, the display name? Where’s the description? Right? These are things that are already inside this that should exist there. Right? If I’ve made again going back to this like I’m already adding metadata into the semantic model this has been I think Tommy our guidance for a number of years

53:08 Now has been the documentation of the semantic model is the semantic model that’s why there’s description columns that’s why there are relationships describe the data that you’re producing and how those tables got there because when you go back to it and you use it in one lake catalog when you go look at it in the report building side that business logic appears when you hover columns and measures. Makes sense. So, anyways, I I think we’re a little too early of this, Tommy. I still don’t

53:39 Understand why I’m trying to use it. That’s maybe a a something I don’t really grasp yet. , and this is where I think the backwards ontology thing makes sense to me because I’m already looking at hundreds of semantic models across my organization. We’ve already figured this logic out. We know what works. We’ve built some reports that are actionable. So on the items that we build across our PowerBI ecosystem, let’s pick the best ones. Let’s smush them all together.

54:11 Let’s see. Let’s bring them all to the ontology. And like that to me that feels like, hey, I shouldn’t have to be creating an ontology or maybe there’s a feature that should exist here. Maybe I should go into the ontology piece and say, select these semantic models and load them in. And then once each model has been made and part of the ontology, I can then start relating and merging different elements together and say, “Oh, actually these two tables are the same thing.” So we have a to your point Tommy earlier in the conversation you’re talking about like these like there’s like specific details which is like the actual name of the

54:45 Fact table fact sales and then there’s higher level objects which is sales data, right? Or or let me see if I can come up with a better name of this one. I I’m trying to look at one of my semantic models here. We have products, right? So products may have multiple tables that describe them potentially. you might have a product table that describes what is the product right now. It’s a it’s a fact it’s a dimension table that is the current record the

55:18 Current values of that item exist right now. You may have a separate a second product table that has slowly changing dimensions in that table. So, what was the product description over time? Every single description that you’ve ever edited to what was the did the product change in dimension or weight or did you change something on the product? It’s still the same product, but features of the product are changing. So, when I look at this ontology piece and I look across my organization at the base level, we may be describing products different ways. And remember, Tommy, we we we always beat

55:51 Heads about this one around like who defines these things. Yeah. Who defines what something is named? I think another one that we argue about a lot is client. This is one that’s , who is the client? What is the customer? How do we define that? Right? I think the ontology has the ability for us to take two different semantic models and say here’s how Tommy defines client and that there’s there’s an overarching object called customer or client but now I can have two or three tables inside this ontology that describe okay here’s how sales looks at customer

56:26 Here’s how finance looks at customer they might have slightly different definitions but we’re still saying the word customer. To your point here, this is where we at least can put those two tables of information under the same umbrella and now we can have a conversation. Right? Now we can add descriptions and information and metadata about it that describe what does customer mean to these different business departments. And then I think at that point now you can have

56:57 And now you can have the conversation that says do we actually need to align on definition? Is it okay to have two definitions? And you might be okay with that. That’s a business decision you’d have to make. So, and I I think the hard thing for us right now, Mike, is there is so new that it’s not like outside of the technical barriers here is like to your point, the binding and like the purpose of it, right? And I like to me I feel like if I was bringing this to leadership saying hey we’re going to build

57:29 This ontology and it’s going to be our definitions around all of our information in the organization and they’re I have to start asking them questions that to related back to the the the technical structure of this they’re going to start getting really overwhelmed because let’s say like just if you were to say it’s not what this is not is simply a glossery and I think that’s the misconception here because it may have some parts of that right but then

58:02 There’s okay the binding like what data is something binded to I can add data to an entity type first off I’m not sure why I would need to in terms of if I already have a source data and then I think the biggest thing is like all the things that you need to define in order to get one entity created. This is where but more importantly like what’s the point like what’s the purpose of it right Mike? Like so even if you do create an entity

58:35 Type and you have it for dim product and you have the main data the source data is coming from and you bind other data to it. Great. Sure. Okay. Now what what do you do with your ontology right like does it what does this now affect in the business that we’ve established different definitions etc that and I think seeing that output is going to help basically how we structure and define

59:08 This too because this is something I would not bring right now to leadership and doing like a workshop because that is just going to I think confuse the heck out of because I don’t even know what the entity type is for I think when you go through like the dialogue box here and you try and figure this out right so semantic model has tables and relationships but you you bind like real data when you go through the binding exercise it pushes you directly to the lakehouse and inside the lakehouse it’s going to automatically

59:40 Allow you to select a binding type of static or time series and then you basically pick a table and the table says here’s the columns. Here’s the property name on each of those items. And you can preview some of the data if you want to see it. And you hit save. So it takes a semantic model and basically goes back to the lakehouse and says, “Here’s the lakehouse table. You can bind this data to.” [snorts] Bless you. Had a sneeze. , so when I look at this, I’m going like, okay, this is interesting. I can see some of the data bindings here. Again, I think this is a play for Microsoft to start

60:16 Adventuring into data governance, right? I think this is a this is an early play into like what does data governance look like? Back to our point earlier, Tommy, we’ve been touching around this topic for a while now, which is like, hey, let’s define customer. Let’s pull in multiple items. What does customer look like to us? How do we define that? Where does this data live? When we say the word customer, which lakehouse, which lakeouses are is it in, right? , I can bind a a lakehouse a specific table to one of these things called customer, but I could bind other things. I could have

60:49 Multiple definitions for that same stuff. Does that make sense what I’m describing there? Yeah. No, it totally does. And I I think once you have those definitions, again, it’s going to be that the the data agents. Mike, as you were talking, I’m going through configuring a data binding and it’s interesting. I don’t know if you did that as well, create it manually, but if you took one of your I’ve tried both. I’ve tried manually. I don’t find much value from it because I have to define every little thing about what’s going on. I liked having at least a semantic model to start from

61:23 Because I I again, let me go back to my mental model here. My mental model was I understand what a semantic model is. I understand the relationships, the tables, and the measures, how all that stuff binds together. it it was a good I because I I’m starting from a reference point of understanding and then going to the ontology say oh I see what it’s doing relationships are reflected this way tables are reflected this way measures are reflected like this okay interesting so that that I think is useful to me from just a mental standpoint but to

61:56 Your point Tommy I I also have done the start from scratch and you can build anything you want and it’s it’s funny though too because one of the things I try to do I needed a time series and that’s the the binding types the time series I’m like wait what why like so and this is where it gets interesting where I think a lot of the this is yeah right now I’m leaning with the semantic model it’s a great starting point but and honestly I would recommend right now for anyone I would start with the semantic model

62:30 Because that’s your best business you mapping of metadata in your organization for certain data. especially compared to the the manual creation here. and this then by default answers that question of [snorts] we are going to be creating these definitions from more from the ground up right now unless the tool changes drastically and we have the better purpose of those entity types and the links. , and then like like that purpose of it. I think

63:04 You’re what we’re going to be seeing is a ton of semantic model ontologies created and they’re going to be less than perfect. what ? Because they’re not going to have everything that you are going to need. But it’s the starting point. , now it is great that I can use the semantic model as a starting point and I can bind data to it from lakeouses, anything in fabric. That’s just part of the beauty of fabric. Okay. But again, Mike, I need to see the output here. I need to see the for what

63:36 Side of this, right? I’m going to create this semantic model. And I think for people, they may test it out. But what do you do with this is going to dictate, I think, how what we create in it. And that’s where I’m leaning here. And I’ll say that again. what this is going to do like so what the output of this is what this feature or an enhancement or how it’s actually making an impact is going to dictate how like how extensiveies are being created and how much

64:08 Governance they have I I think you’re right on this one Tommy I think again I think the idea here is we don’t have a good framework I I Yeah, I’m not sure if the documentation illuminates to me a clear purpose as to what the ontology is there for like I I like the idea that it’s building like this graph thing. It’s trying to build relationship between stuff, right? Is let me so let me let me project some things here based on what I

64:41 See so far, right? I have this mapping of how things relate together. I can add relationships between tables and fact tables and dimensions, right? We can have that exist. I can bind data to specific areas. This is something that I think Purview was trying to accomplish, which is let me scan your system and show you all the things that you have and then you can add relationships between different tables and data and where does it exist and and mapping it out. Right? So this is basically a way of mentally mapping

65:15 Your organization and and specifying across semantic models, lakehouses, data warehouses, right? Where do all these tables exist? Right? It’s a mess right now. How do I link this stuff together? So I see the ontology as being somewhat this glue of how we start documenting what we build. Now great interesting interesting point. Where does this start? Like my question to you would be is why does this add value? Why would I do this other than having a pretty graph that I could explore and click on different tables to

65:48 See what’s there? I I’m I’m not sure the val the lemon is not worth the squeeze for me at this moment to go in and say, let’s spend a ton of time building out ontologies for my organization because I’ve got to figure out what does the value what is this saving me time in doing, right? Is it saving me time in like discovery? Is it saving me time in like data quality? So, I fully feel like, and this is maybe something I could maybe go in here, but if as I’m using this experience, I feel like I would have data rules or data something

66:22 Constraints on top of this data. Hey, I’m going to bring in this sales table. what’s the count of records today? Okay, every day I wanted to go count the number of records tomorrow. Okay, that starts adding value to me. Now I’m starting to watch data drift over time. , hey, I’m going to build this sales table and this column is the primary key. It should never be blank. , or this data in this column should have this criteria, right? So there’s like business rules I think I would apply to

66:54 This. So if we’re talking like proper data governance, I think a value for me and this may be where this is trying to go. Maybe it’s going to get there eventually, but I’m not sure yet. But to me, adding like constraints against this data, adding rules about this data on top of this ontology piece makes a lot more sense to me because now I can say, okay, let’s get a holistic view of my organization data. Let’s have quality data quality scores on top of all this stuff. Like that

67:26 Makes sense to me, right? Here’s all the sales data that we have contained in our sales tables. We have three tables. We’ve got some lakehouses. Here’s all the data. These are the rules that we’ve applied. Here’s the data quality scores for all that stuff. Like I need a holistic view of my company and the data that we have everywhere. And I need scores and percentages and is it valuable? Is it not valuable? How many people are using this? , how many times has this data source been accessed in the last month? Are there usage metrics that we tie to this stuff? To me, this ontology thing, I think,

68:00 Gets much more powerful when we’re able to add extra metadata to it that enhances like the quality and and the usage or discoverability of our data. Now, maybe that’s where Microsoft is going. I may be projecting a bit too much here in the ontology space, but to me, that’s what’s going to add real value and that’s when I can start putting the, , working on building this and pushing out to my organization. Right now, I don’t really see a lot of value at the initial stage other than making a pretty graph and for some very lightweight discoverability. And even that, I think is like minimal at best. Does that make sense what I’m

68:33 Describing there, Tom? No. 100%. Because if the purpose of this right now is you create an ontology to create a data agent, well, how much better of an experience is that data agent with an ontology layer source rather than a semantic model or a lakehouse? You’re projecting as well. And I I like I like this, right? Where where do we attach this? Like can you go to a like the other question I had I’d have for you Tommy is if we make a data agent inside a workspace that has an ontology in it, are you able to add the ontology to the That’s the main purpose from

69:05 What I’m reading. That’s the primary reason an ontology exists is you can do like one of three things with an ontology. You can connect it to a data agent. You can graph use it in Microsoft graph to query and you can use it with an operations agent in real time and those are the three things right now that an ontology in a sense is a engine for or fuel for. So, but the big one is the data agent which again if I already have a semantic model with

69:39 Definitions why would I need to create a ontology layer like how much better is that data agent going to be using ontology? because I can do in with a data agent the same queries, right? So that other side of it again is the Microsoft graph, but I don’t know where in the organization you’re going to be doing that. Tommy, have you loaded an ontology into the micros the the data agent yet? No. Why? It looks horrible. There’s definitely

70:12 There’s a major bug. [laughter] So if you load if you go so while you were saying that Tommy I went back to my Microsoft workspace. I have a number of ontologies. I made a star model ontology and I pulled that ontology directly into my data agent and oh my word there is it’s a mess. The the icons are all over the place. I have a major bug by pulling in ontology into the so we are in really early stages here. , [snorts] and maybe and may and and maybe this is

70:44 Where it’s going to go. Like maybe the ontology is helping you describe data to make it better for your AI agent to go find data sources and relationships and reports and things. It’s traversing that knowledge graph of like what’s inside your business. But dude, man, it’s it’s rough right now. It’s it’s definitely yeah, it’s it’s really difficult to use at this point. So, we’re definitely not there yet. , it feels like we’re very far away from effectively using ontologies at this point, but I think I’m starting to get the concept about it. This is a weird concept. There’s a lot of weird things going on here,

71:16 Tommy. That’s encouraging. I get Gosh, that’s frustrating. Microsoft, if you had to wait six months to get this released, I’m okay with that. I really am. Well, I I guarantee you it was Come on, everyone. we need to get this out the door for fab for Ignite or whatever the whatever the next big conference was and so we rushed some things out the door to get it done. So yeah, it’s definitely quite buggy. definitely not digging the experience so far. It needs a lot more work here to get this to work better. Anyways, okay. I think we’ve beat up this whole

71:50 Ontology. I think the the idea of the backwards ontology does make sense. I think we need to go for more semantic models and get more elements into the ontology. , we’ll see how well Microsoft can make that easy to integrate and easy to pull data in. I still feel like the story is not I don’t understand the story yet. From what I read the docs, from what I read it, it doesn’t quite fit for me yet. I need more vision for where Microsoft’s going with this. Anyways, I completely agree. That being said, thank you all for listening us ramble

72:23 About more ontology pieces and your organization where data lives. these things might be useful to you. I’ I’d give it a couple months. hold off on the ontology pieces. We are early testing these things. Maybe give it a couple months here to refine a little bit and add some features here and and fix a couple things here because it’s a bit rough around the edges right now. So I would say that being said, thank you all very much for listening to the podcast. We do appreciate you. topics and information you guys give us is a wonderful and super valuable. So make

72:55 Sure that you let us know what you’d like to discuss in the comments of this video right now. Let us know. Are you experimenting with Ontology? What do you see it being useful for? Where would you want to see ontology being value added in your business? Do does this even make sense? I don’t know. Let’s unpack this together. We’re really unpacking like what does this mean and is this something that we will find to be fundamental as we build things in the future. That being said, Tommy, where else can you find the podcast? You can find us on Apple, Spotify, wherever you get your podcast. Make sure

73:27 To subscribe and leave a rating. It helps us out a ton. Do you have a question, idea, or topic that you want us to talk about a future episode? Head over to powerbi.tips/mpodcast. Leave your name and a great question. And finally, join us live every Tuesday and Thursday, 7:30 a.m. Central on all of PowerBI tips social media channels. Thank you all so much for participating. We really appreciate your ears today. have a wonderful week and we’ll see you next time.

Thank You

Want to catch us live? Join every Tuesday and Thursday at 7:30 AM Central on YouTube and LinkedIn.

Got a question? Head to powerbi.tips/empodcast and submit your topic ideas.

Listen on Spotify, Apple Podcasts, or wherever you get your podcasts.