Return of the Composite Model – Ep.551

A listener describes a hub semantic model with three to five composite spokes that DirectQuery back to it while importing department tables locally. It’s a reasonable design, and both hosts spend the episode explaining why they’d still steer somewhere else — plus the governance you need if you go ahead anyway.

News & Announcements

• Composite models in Power BI Desktop — Microsoft’s reference for the pattern under discussion.

Beat from the street — Mike’s PowerShell aha moment. He had Claude write him a PowerShell script to rename and sort the video files that pile up on his desktop, and the realization stuck: he should have been scripting his repetitive desktop work for years. The catch he names openly is that he doesn’t know PowerShell and won’t spend time learning it — which is exactly the point. Paired with One Commander (a power-user Windows Explorer replacement with a lightning-bolt button for scripts) and PowerToys’ PowerRename, it’s become a genuine workflow. He also flags AI Shell (start-aishell) and notes the new Windows terminal now has an agent built in.

Main Discussion

Topic: Whether hub-and-spoke composite models are the right pattern, and what to do instead

• The structure is defensible; the default answer is still no. Mike wouldn’t encourage central teams to use composite models at all. If a department spoke builds one for itself, fine — but it isn’t the pattern he’d standardize on.

• Master and mini models is the better story. Build one centrally designed model, then carve off domain-specific mini models from it. Manage one thing centrally, deploy the carved pieces repeatedly. Mike points to Michael Kovalsky’s master/mini model work and to Tabular Editor as the tooling.

• Cardinality decides whether DirectQuery hurts. Sending low-cardinality filter context across models — country, a handful of sales reps, a year — usually performs fine. Push day-level detail across many years and you’ll feel it.

• Two fact tables confuse users on their own. Mike’s firm position: give someone two fact tables where some dimensions work with one, some with the other, and some with both, and they’re already lost. Carving off a focused model solves a human problem, not just a technical one.

• Pro workspaces are underrated here. If your model is under 1 GB, you can run hundreds of them in Pro without worrying about capacity or noisy neighbours — do the data engineering in Fabric, run the models in Pro.

• Every composite model needs a data contract. Mike proposes it as a rule: a defined handoff, a clear owner, and change control over dependencies. Composite models are powerful and should never be built ad hoc.

• You can restrict them at the model level. There’s a setting on the source model controlling whether composite models can be built from it — relevant if you’re maintaining certified datasets, which Mike would not let go composite.

• Nothing tells you which models are composite. Discovery is the missing capability, and it’s why Mike built the lineage workload — measure drift across composite models is otherwise invisible.

Git as a governance tool

Tommy’s practical answer for anyone committed to the composite route: the listener already uses PBIP format with git and feature branches. So make monitoring mandatory — watch the repository for changes, pair that with usage data, and you have a real signal when a model starts drifting. His summary is that git is now doing data governance work, and doing it better than the alternatives.

The layer above the semantic model

Mike’s recurring argument surfaces again. A relationship between a fact table and a dimension is business logic buried inside a semantic model. That metadata should live one layer up in something generic — which is the case for the open semantic interchange, and the gap that makes composite model sprawl so hard to see.

Looking Forward

Before adding a composite spoke, ask whether a mini model carved from a master would serve the same users — and if you build the composite anyway, give it a contract and a monitored branch.

Episode Transcript

0:08 measures. It lighting up the sky. Dance to the day to laugh in the mix. Fabric and A. I get your feels. Explicit measures. Drop the beat now. Kings feel the crowd. Explicit measures. Hello everyone and welcome back to the explicit measures podcast with Tommy and Mike. Tommy, welcome back to the show. Hello. Hello. How are you doing? Hello. Hello. Good morning, good

0:38 Hello. Hello. Good morning, good afternoon. How you doing, my friend? I’m doing well. Thank you very much. Tommy, before we get into our main topic today, I want to share a little bit of maybe it’s a beat from the street. Maybe it’s just Michael just learning things. Sorry, I talked about myself in the third place, third person. I I think there’s something I just realized over here in the last couple days. And maybe this is just me. I don’t know if anyone else is having this epiphany as well., our main topic today is going to be a mailbag. We’re going to talk through a mail question that came in. And this question is the return of the composite

1:08 question is the return of the composite model. All right. This is going to be the the the topic here of at hand. So before we get into that main topic, Tommy, Tommy, have you ever felt like you were doing something fundamentally wrong [snorts] and you learned something, you were like, “What was I doing before? I think I didn’t understand it.” Are you talking about in the moment or afterwards? So like in the moment you’re like, “I don’t know about this or

1:38 like, “I don’t know about this or afterward you realized it because to me those are two different things.”, no. I I don’t know how to I don’t know how to answer that one, but it was it was more of like an aha moment. Like I I did something with claw on my computer and I was like, “Huh, I never thought of that. Maybe I’m actually using computers wrong my entire life.” life.” Oh, wow. Okay, that’s a big statement. Okay, so computers are good at doing repeated tasks over and over again.

2:10 repeated tasks over and over again. Okay. Okay. Tommy, I’ve been working with Claude and trying to streamline my workflows. And Tommy, we do a lot of podcasts, we do a lot of videos, we do a lot of things that just in general just we’re busy, right? We’re doing all these things. For the first time, I feel like I’m able to like actually get my computer to automate some of my stuff for me, do things in an automated way. And and maybe this is maybe this is an aha moment that other people have had. I have not had this. Maybe I’m too much of

2:40 have not had this. Maybe I’m too much of a business user. I don’t know. But on your computer, you can run PowerShell. PowerShell runs the entire machine. Can do anything you want with it. Like I and I knew this like I I yeah, I get it. But let me give you an example. Every like four times a week, Tommy, I’m doing a podcast. Two of them were with explicit measures. Two of them are agentic thinking. Videos are coming down. down. Every time I do that, I download three videos. It’s the audio file. It’s like the landscape version and then the port portrait version of those videos. They

3:10 portrait version of those videos. They always go down my desktop and I’ve got to like rename them, get them into a format that I like and then put them somewhere on my computer. This is a task that I do fairly frequently, but it’s pretty mundane. Well, I thought the other day I thought I’ll just have Claude write me a script for this. So, So, now Claude has written me a PowerShell script that when I have videos on my because I download everything to my desktop. I’m a very weird person. Whenever I download something from the internet, it goes to my PowerBI desktop or or PowerBI desktop. It goes to my it goes to my my Windows desktop, it just

3:42 goes to my my Windows desktop, it just sits there and then I am expected to do something with it. I move it to some place. I store it in a folder. that to me any files on my desktop are telling me I have to do something. So I download these files. They’re on my desktop. So they’re in the same place every time when they get downloaded. And I just wanted a script to just do the things, rename the files, move them to the appropriate location. So in working with Claude, I don’t have to write that anymore. It now automates that entire

4:12 anymore. It now automates that entire maybe and the Claude’s doing this anyways. I just had this aha moment like Claude’s building all these scripts for itself to do anything like all my skills and tasks that I give it. It it writes scripts to go search for things. It’s I think right now the harnesses that we’re using they’re getting smart like they’re not letting Claude just read files in a folder. It’s doing like okay I’m going to build a script that will then go read and search through the items and then return any results that I have. Right? So it’s building a lot more of these tools or mini tools that it

4:44 of these tools or mini tools that it uses to help you build stuff. And so Tommy this was my aha moment. I think I should have been building scripts for a lot more of the standard clicks on things on my desktop. If I’m moving files around, if I’m doing something that’s repetitive, if I’m cleaning up folders, whatever the the task is on my computer, I should be writing PowerShell scripts to do this. The trick is I don’t know how to write PowerShell. I don’t and I’m not going to spend the time to learn it. It can do a lot of things. I don’t want to spend the time to educate myself on that stuff. Unless you were in IT then

5:15 you were in IT then unless you’re in it I maybe sense yeah it would make but now me the non-educated PowerShell user just go to my agent hey write a script that does this I describe what I want here’s the format beep beep beep boom done okay oh you didn’t it didn’t work quite right right I couldn’t find any files great update it hey I want a little progress bar as you’re copying files because these files are large and I need to see when they’re uncopying and click to

5:45 they’re uncopying and click to exit the PowerShell script. Boom. Done. Got it in there. It’s all there. And I’m thinking to myself, Tommy, this is fundamentally shifting my understanding of what computers can do. Mike, Mike, welcome to the world of PowerShell. I went down What the heck, dude? So, there’s a few things with PowerShell. I I went down a rabbit hole initially when I think Ruie Romano had the audit report that was all PowerShell. This was we’re talking about like 2018 2019. Yeah. And it was incredible. And I was

6:16 Yeah. And it was incredible. And I was just like, well, this PowerShell thing can do a lot. You combine that with a Windows application called One Commander, which is basically replaces your Windows Explorer. It basically has always like two windows of what you’re trying to do. So easy. It’s awesome. It’s like the power user version of Windows Explorer. And what it has in it is this little lightning bolt that you can add scripts. So, I’ll give you a perfect example. I have a giant folder of cloned repos for skills and I have ones for fabric. Like

6:46 skills and I have ones for fabric. Like I’ve cloned them, but what if I want to get the latest version? I want to pull the latest commits that users have done. Well, I don’t want have to go into each one and do a pull. I can actually choose the main folder and it will go per each thing selected or within it and do a a get pull for each thing. But you can do this with any PowerShell script. to your example. You can go through all your files, select them, run the rename and move script,, for audio and set off a ton of things

7:18 for audio and set off a ton of things and it can do it. So,, dude, PowerShell is pretty pretty awesome. And you can get modules that you can download to really expand what you can do. So, I give PowerShell a lot of credit. If you have Windows, there’s a lot of reason to at least ask Claude to help you build some scripts to help your workflow. workflow. Yeah, I’m I’m I just felt like I have wasted so much time time clicking on things and renaming things

7:48 clicking on things and renaming things and do like there’s a couple tools that I found that was useful. So I use Power Toys from Microsoft on my Windows computer. Yeah. In Power Toys you have this one item called power rename. So you can like look at multiple files and you can find a piece of text and then you can rename it. You okay that’s interesting. Like that’s fine. But like I don’t want to do that anymore. I don’t want to type on the keyboard. I just wanted to like just do it. Just make it happen. Move the files where they need to go. And so now I’m I’m looking at this Tommy going okay this is interesting. Can I extend the script to upload these files to

8:18 the script to upload these files to SharePoint? Would that need to be uploaded? Can I can I use this script to also upload the video directly to YouTube using a URL and a token that’s on like on like there’s a there’s a SharePoint connector with PowerShell. I know I’m looking at this going like my workflows on my machine are greatly so Tommy we’re we’re not living in a very fun world., and I think for me, again, maybe this is an unlock mode because I’m able to use some AI agent

8:48 agent to just help me like be more efficient on my local computer. And I just wanted to share that with someone cuz I’m sure I’m doing something similar to other people that they’re doing the exact same thing. They’re they need to be using PowerShell, but you don’t know how to write it. you’re not sure where it fits in your workflow. Like you just don’t know what you don’t know. I didn’t know what I didn’t know. And I was like, “Oh my goodness, I now know this and think this is amazing.” So,, anyways, just wanted to share that that

9:19 anyways, just wanted to share that that tidbit of like knowledge or things that I was looking from., and it’s working like a champ now. I can I can download multiple videos from multiple different formats of of shows., it searches them correctly for each one of them. It has file name types. So it it only it only moves or searches for files that have like M4A or MP4 in the title. So it knows how which specifically items to reroute and and I Wow, Tommy. I’m going to warn you, it’s it’s a potential rabbit hole exploration with

9:49 potential rabbit hole exploration with PowerShell. Jeez, man. Well, and the part that the one thing I want to make sure that people understand is I don’t know how to write PowerShell. But you can you do now? But I can get it to do the the outcome is what I desire.

10:03 to do the the outcome is what I desire. The medium and what it writes it in. Not really interested in. I don’t really care to like learn it. I just want to I I want to be able to have,, describe in sentence to something and have it get out the output that we want. This is really useful. And also, Tommy, I’m using a lot more of the Windows H command. So, Windows H is where you can talk to your computer. Some people are using like Whisper AI on their machine, and it’s basically a your microphone’s here. You can talk. I’m finding again, you’ve been saying this, Tommy, you haven’t typed on your computer in like months. You’re

10:33 computer in like months. You’re continually talking to your computer more and more. There’s probably some typing. I can’t I can’t imagine there’s no typing at all., , but but this this world of talking to computers, automation of things, it’s getting much better. Now, one of the things I’m a little bit not mystified, Tommy, but maybe like leerary of or need to see how this is going to play pan out. PowerShell is really good when there’s like physical files, folders, and things on your computer that I can go look through and use. Makes sense. Totally get it. A lot

11:05 use. Makes sense. Totally get it. A lot of the work that I do now is just in applications or specifically websites. So, a lot of work that I do is like in my emails. Well, I don’t have the Outlook program downloaded on my computer. I just use the Outlook application in my web browser. And so, actually, I like a lot the Outlook new Outlook. I just use the new Outlook version that’s in the web browser. Yeah. I don’t even I don’t have it I don’t have it downloaded on my machine. And if I did have it downloaded, it’s only there because someone sent me an IC file and

11:35 because someone sent me an IC file and it needs to like load that file directly into the Outlook, but I never open it. It’s never open on my computer. I’m always in a browser window that has my calendar or my Outlook calendar there. That’s always there. Teams I still use locally as an application. There’s there’s a couple other apps, but most most of the apps I use now are just all websites. websites. Well, here’s the thing. If it’s if it’s a m especially if it’s a Microsoft product again PowerShell has a ton of modules and Microsoft is very serious about how important PowerShell is. So they’re constantly updating it. All those things you’re trying to do in

12:06 those things you’re trying to do in Outlook or Teams or SharePoint, there’s a package or module that has all these actions on it and it’s really cool. So yeah, there’s another thing I forgot what it’s called. It’s like P. It’s an extension that Microsoft created that has all these additional actions. You can basically do keyboard shortcuts in the terminal or like AIC or something and it expands it. So yeah, man, PowerShell is really cool. And they have an AI thing with PowerShell as well now. So you can actually use that

12:36 now. So you can actually use that directly. It’s okay. At at Microsoft Build, they just announced there’s a new terminal that you can go get. It’s called like the intelligent or terminal plus or something like that. you can download it. It’s a new terminal, but it allows you to have GitHub copilot like attached directly to the terminal. So, you can be in the terminal and then the agent, it actually basically brings an agent directly to your terminal when you need it. So, if you have a question, you can open up like to your point, there’s a split terminal that you were talking about where there’s like files and there’s like scripting or things that are happening with that Power Commander or Windows Commander., but the new

13:08 or Windows Commander., but the new terminal plus that they announced at Build, I downloaded it. It’s actually pretty useful. And if you ever get stuck or get if you’re running commands or doing something in code and you have a you get stuck, you can always just ask the agent, hey, what happened here? Can you help me fix it out? And it gives the the ancient context to what you’re looking at, which I think is useful. useful. I am seeing that now. That is pretty cool. Yeah. So, that is pretty cool. Is that what it’s called? Terminal Plus. Yeah. Windows intelligent terminal. Oh, that’s what Yeah. Okay. That’s the one. one. And then Yeah. And then the AI for

13:38 And then Yeah. And then the AI for PowerShell is a it’s called the AI shell which is simply you just do this command called start- AI shell and it’s I think it’s an older version of it because it only uses a chat GPT but still there’s a ton of things you can do. So yeah, I like it. So anyway, I I will say I do enjoy the intelligent terminal. It has some different formatting. It’s a little bit color more more colored well. so I do like that one. It’s a it’s a useful tool and they are improving it. the update was just 5 days ago. So they

14:08 the update was just 5 days ago. So they are,, regularly updating it,, and building something there as well. So anyways, really awesome. Thoroughly enjoy these tools that are coming out right now., very cool stuff. Anyways, that being said, let’s get over to our main topic so we stop yammering about how Michael doesn’t know how to use computers and is continually changing his workflows on every single thing. So that being said, Tommy, let’s go over to the return of the composite model., let’s hear the question. question. All right, here we go. This is from Austin. Hi guys, thanks for the recent mailback on bloated golden semantic

14:40 mailback on bloated golden semantic models. Could a composite model pattern help further simplify the one big bloated model issue? Our org has a few large semantic models for distinct business segments. As we research about the golden semantic model pattern, we are hesitant to consolidate to create a 100 plus table behemoth. We’re considering a hub and spoke approach. A core semantic model which they the hub with common facts dimensions and measures. Then three to five specific

15:11 measures. Then three to five specific composite models the spokes that direct query back to the hub and import department specific tables locally. Reports can pull from one or more semantic models with cross-domain data is needed. Additionally, this would give business users a cleaner analyzing Excel experience. For our context, our report consumers use port pro workspaces. We use some fabric premium capacities for data flows gen 2 pipelines and notebooks. We develop

15:41 pipelines and notebooks. We develop reports and semantic models entirely out of PowerBI service. Just kidding. We use PBI and what what Michael really use Git. Stop with a developing everything in the web. You can use Outlook. That’s fine. But my friend, if you come buy a Chromebook, go buy a Chromebook then. I can do it in the service. It’s so cool. cool. Buy a Chromebook. Just turn on get in the service. You’re good to go. Just use your iPad then. Not just kidding. That’s a true statement. Keep going. [gasps] We use our PB pip p format and get plus

16:12 We use our PB pip p format and get plus feature branches with local development. So good. They’re do all heavy developer here. Would love to hear your thoughts on this pattern. Sincerely, Austin. So, this is a great great topic. There’s lots of things to unpack here. So, we’re going to have to like really dissect I have a lot of thoughts already as you can tell with the why wouldn’t you build more in the service? It seems like it’s such a great such a great feature. So, what’s your initial reaction Tommy to this comment? Where do you want to take the conversation today? I I think there’s a there’s a few things, but I’m going to start with our

16:42 things, but I’m going to start with our one of our previous episodes. We actually talked about do we build agents for se or do we build semantic models for agents or do we build them for humans? One of the things that we mentioned on that episode was very straightforward, distinct and clean semantic models are great for agents, right? where it’s just the three dimensions and then one fact table that works very well for agent reading based on what anthropic put out and what snowflake has put out as well from

17:13 what snowflake has put out as well from what they are doing in their attempts to get AI to really do self-service analytics. However, we also know that’s not great for humans. If you if I’m building a onetoone pattern of for every semantic model, I’m only turning out a single report. That’s not really scalable for us. So there’s a lot here outside of the AI stuff, but I think especially incorporating the world we live in now that there is this potential of the semantic model. Our wish our

17:43 of the semantic model. Our wish our solution Mike was something that doesn’t exist where if you remember it was the P word perspectives but that that’s not possible right now to have an agent run on perspective. However, there is something here that I think is very potential from the AI side. But much more importantly, this is a if you s saw this question, this mailbag, and it was the year 2019, you would all it would make as much sense today as it does now.

18:19 2019. Yeah. Yeah. Yeah. Yeah. I don’t think there’s anything different in 2026. There’s not a lot of crazy fabricy things inside this story, which I I’m I’m liking. I do think there’s some interesting So there’s there’s a couple like minor notes I want to pull on here inside this one. Right. So we’ve talked a lot about composite models in the past. composite model is where you have two

18:49 composite model is where you have two separate locations of information. an import mode and a direct query. That’s basically what’s happening here, right? right? direct query can handle multiple things. You can direct query back to like a SQL server or some server side, but you can also direct query to another analysis services. So there’s import and direct query and then direct query handles many different sources. One of those sources is the analysis services. So,, , it it because we’re in PowerBI Pro for

19:21 it it because we’re in PowerBI Pro for this example, it seems like there’s not really large models, meaning they’re not the data models that they’re building are not larger than 1 gigabyte in size. That’s critical for what this story is meaning here., because as soon as you get above that 1 GB size limit, you’re no longer able to fit stuff inside the Pro model. Okay. I do hear there’s a little bit of note here around they’re using some fabric for like data flows gen two some pipelines and some notebooks. Awesome.

19:52 notebooks. Awesome. Which that also what they didn’t note here is that’s likely pushing data down to the lakehouse. So I think there’s actually a lakehouse that was not

20:00 actually a lakehouse that was not described here that needs to exist in this architecture. Okay. So all this being said, I want to just hang on the quick note around composite models again that direct query portion of it. I honestly for it’s very good in small use cases when you need to supplement an existing model with a composite model because the composite model when it direct queries to analysis services all the filter context from one island. So each each each separate model is an

20:31 each each each separate model is an island in the in the the comp in in the semantic model, right? We call these things islands. As soon as you send data from one island to the next, right? I’m going from import mode which is in the current model and I’m going to a sales table that’s in a different semantic model somewhere else. Right? Any filter context I’m giving that query has to turn into a text string and that text string is then passed entirely to the other model. Marco Russo has gotten a number of really good articles on this one. And as

21:02 really good articles on this one. And as long as the number of text items you’re sending is minimal. Like if it’s a co if it’s a column with low cardality, you’re probably not going to notice an issue. It’s probably going to work fairly well. country might be a good example, right?, sales representative and you have a handful of sales representatives might be might be relevant, right? year filter context might be relevant. But if you go down to like the day and you’re doing like many years and you’re you’re sending day level context filters over to other models, the day level detail that’s

21:32 models, the day level detail that’s sending every single date directly to the other model. If you’re doing identifiers like product identifiers or gooids inside the model, every single gooid that you’re listing as a relationship between the two tables that is now getting sent across to another model using only text and it’s highly inefficient. So for that reason I don’t really love working with composite models. Kevin Arnold, did you see ever see his talk

22:03 Arnold, did you see ever see his talk about this Tommy? Kevin Arnold did a talk about composite models and he was like here’s when here’s good use cases of when to use a composite model. Yes. Yeah. Yeah. Yeah. That was years ago. years ago. Yes. Yeah. Yeah. Yeah. So I’d highly recommend go look up Kevin Arnold and then composite models because he gives you some really good use cases as to when to use them and when not to use them. Okay. So all that being said, I really like this pattern and I think before this conversation, Tommy, I would

22:34 before this conversation, Tommy, I would argue with you, Microsoft doesn’t have the tooling to do this. What you’re describing here is the enterprise semantic model. These are tables that exist inside a lakehouse and you need those lakehouse tables to get loaded into a large semantic model. And we could have a really big semantic model with many many tables in it wouldn’t be an issue except the users who consume that really large table that really large semantic model you get super lost

23:05 large semantic model you get super lost in what what measures can be used where with what stuff it gets so convoluted and it’s slow to edit. So people don’t build monolithic 100 table models even though that may be what how their business is described. Let me just pause there. What do you think about that? And I guess my biggest thing I to kind And I guess my biggest thing I to shoot back to you is that my biggest of shoot back to you is that my biggest concern here because you’re talking about to me I’m hearing a story of someone who’s beginning to start with composite models and the warnings along

23:35 composite models and the warnings along the way,, hey, if you’re going to take this route, don’t take a left here. You’re going to fall off the cliff. And to me, I’m hearing at least from the mailbag a different story that it’s like, hey, this is someone who obviously knows their stuff, right? They just based on the way they wrote this Austin Austin If you’re listening, tell your boss to listen, get a raise because you obviously know what you’re doing here to me., and something that’s been a lot of experience. We’re talking about also two very specific composite models. We’re not just building this monolithic,, composite model, making a

24:06 , composite model, making a monster or Frankenstein composite model, which to your point is very easy to do and then cause a lot of issues. However, if we have a directed approach, which what I’m seeing here, a very targeted approach on a very intentional way that we’re building composite models, I I don’t think that’s going to be my biggest issue that I would raise here, Mike, because what I’m looking at this is why would you do this, right? Let’s to me, I want to start there. Why would this be an approach that I would want to

24:36 this be an approach that I would want to take rather than the pitfalls that may occur? Why would this be a place that a lot of organizations do? Why does Microsoft still support composite models? And I look at this and this is the classic example, right? I don’t want to build a giant semantic model that does not allow perspectives. So, it’s either all or nothing for users. And also, yeah, I want to be clear on the perspectives thing. thing. You can use perspectives. They’re

25:06 You can use perspectives. They’re they’re in there. It’s just I think I think what you’re trying to say though is Tommy is you’re saying the model does have perspectives you can add them to a very large model but the only thing is when you’re building a PowerBI report connected to that model PowerBI desktop the service there is nothing that you can do to narrow down the fields that I’m looking at so that when I’m building I’m creating experience I’m looking at just a perspective right so until yes Mike yeah until I can you sorry it’s not that’s not to you it’s just my upset my anger at perspectives

25:37 just my upset my anger at perspectives not being a lot. I’m not mad at you, but until perspectives are available in most of the other features in PowerBI, then perspectives are not available to me. Sorry. Yes. Because what I use in a personalized visual dead to me. You’re dead to me. Exactly. Yeah. Exactly. I was like, you’re no longer my son. You may be alive, but don’t come for Thanksgiving. Basically, that’s that’s Oh, boy. Oh, boy. So, I get I get heated about this because I think there’s a lot of opportunity. There’s so many perfect use

26:07 opportunity. There’s so many perfect use cases for perspectives that we just don’t have today. Anyways, so going back to this and I would Mike make a bet that Austin is also in a managed self-service type and environment. Their culture, their organization is using MSS is using managed self-service. It’s not just the hub and spoke approach because they want other people to build. They’re not just doing composite models just for them. I’m making that assumption. Would you agree with that? just Yeah, I think so. I think what they’re

26:37 Yeah, I think so. I think what they’re trying to do is there’s the spokes are a smaller domain or a carving out it’s Michael Kowalsski coined this term. I think he actually made some scripts around this called the master and mini model. Yes. Yes. Right. So master and mini. Right. Master model is like the model that has all the relationships that has all the context has everything inside one large model. And that’s what we’re describing here. Right. We want a core model, the hub to have all the dimensions, all the tables, all the relationships, everything there.

27:07 all the relationships, everything there. So from a management standpoint, I can centrally manage a single item. I can manage that one item and then when I need to carve off a portion of this, five tables, a specific domain., one of the things I will I’ll highly stand ground on are if you give a user two fact tables and some dimensions work with fact table A and some dimensions work with fact table B and some dimensions work with both tables, users get confused already, right? So if

27:39 get confused already, right? So if you’re going to build a multi-fact table model, model, right, right, we need to be very intentional as to how do we communicate which dimensional tables are available to be used with each factual table inside that model. In my mind, when you spend time to really simplify this down and carve out things you just don’t need, it really makes it a lot easier for users to understand what’s going on the model. We don’t really have that capability. And this is where the mini model, I think, comes into this. So one other area I think I’m

28:09 into this. So one other area I think I’m going to I’m going to read another requirement here just basically projecting based on Austin’s question here. here. Sure. Sure. So the other question here from Austin is talking about composite models. models. If we’re doing this correctly the lakehouse has tables in it. These pro workspaces I believe are importing lakehouse tables using import using compute in the pro workspaces. So there’s a pattern here Tommy. I’m doing

28:41 there’s a pattern here Tommy. I’m doing the data engineering in fabric but I’m doing the doing the pro models importing and running and compute all that’s inside the pro workspace. Pro workspaces are amazing. If the model size is below 1 gig in size, dude, you can build hundreds of models and never worry about capacity usage, the noisy neighbor. Microsoft just figures out how to manage it for you and handles all the stuff. Done. Like it works. It just

29:12 stuff. Done. Like it works. It just runs. And so when I look at this, I go, “Oh, interesting.” Well, let’s spend our time defining what is in the lakehouse. Now, and Tommy, I’m going to go another layer here. here. We build bronze, silver, gold. We, you We build bronze, silver, gold. We,, we bring in the raw data. We do know, we bring in the raw data. We do some transforms to it. We basically get to a table that’s gold tables or tables that are ready to be consumed by a different team. Those gold tables, I would fully expect them to be completed dimensional tables, completed factual

29:42 dimensional tables, completed factual tables. They’re verbatim copies directly into the models. I like this pattern because it now enables me to have mini models programmatically created from the master model. And here’s where I want to

30:00 model. And here’s where I want to I’m so convinced this is the way people should build things. I’m going to say it again. Tom, you’re going to you’re he’s going to roll his eyes. I can see it already. He’s already getting ready to like bulk at me a little bit here. Start with the W because I’m I’ll just hang out now. Does it start with No, no. it I we have built the lineage tool. So there is a workload I try to leave lineage but it’s a workload. there there’s a workload that that we have created that’s called lineage and lineage is

30:30 that’s called lineage and lineage is allowing you to do exactly this. You can build the larger semantic model. So if we think about what’s happening in our world Tommy right we have tables in the lakehouse. Those tables inherently have relationships to them. You’re building the tables in a lakehouse that has a factual table and you have a dimension table. They exist in the lakehouse. There’s nowhere to define the relationship. In the lakehouse, you don’t define the measures. That information is carried in the semantic model. So, think about the tooling we’ve got. We’ve got one piece of tooling which is the lakehouse which carries all

31:01 which is the lakehouse which carries all the really rich like that’s the data area. That’s where all the data is being stored and and kept and and updated. Great. Love it. Then the semantic model comes in and adds what we’re calling the semantics layer to it. Now totally get it. Makes makes a ton of sense, Tommy. But in the semantics layer, we’re required because of the way the semantic model is built. In the semantics layer, we’re required to load physical data into the semantic model. We need something bigger than that. We need we need a semantics thing that does

31:32 need we need a semantics thing that does not have data in it is what I think we need. No, if this may shock you, but I agree with you on this. I think if you are doing composite models and deploying that, if you don’t have anything like the lineage view that you have in your workload or you built your own but some capacity like that, you’re going to get measured drift. You’re going to get into a lot of complications. I that’s essential. I and again I I may roll my eyes but to this day I don’t see a

32:02 eyes but to this day I don’t see a better view to be able to view all the measures if I can choose certain measures. There’s lineage view and PowerBI but it’s very standard in the box out of the box feature. What I I like your lineage view because I can choose the different semantic models that I want to actually see where the crossover is. And for me, this is a governance thing because you cannot do composite models without having some way to audit what’s actually out there in the in the wild. So, let me extend this idea a little bit further. So, one, I love your idea,

32:33 further. So, one, I love your idea, Tommy. Tommy. By the way, so we’re trying not to push it too much, too much, right? Okay. Understood. Understood. When we think about these composite models, things that are being built here, right? So, let’s go back to our example here. We have lakehouse tables. We have a semantic model. Let’s just say we did build one monolithic semantic model, right? Well, I don’t really want the data to be in that because that model is going to get really big. I can’t really publish it to power. com, right? I do want what I do want though, Tommy, is I want the relationships. I want to be able to establish like

33:03 I want to be able to establish like these two tables exist. Here’s the fact table. Here’s the dimension table. And I want to build semantics between table A and table B, right? This this key column in the dimension table matches to the fact table. That’s a that’s a piece of metadata that we need to have. and hold on to. And so that level of information is relevant. Okay, that being said, once you establish the relationship between two tables, think of it this way. You can abstract away even the semantic model. Let’s say

33:35 away even the semantic model. Let’s say I have multiple business users across my company. Again, I’m going to talk really conceptual here for a second here, but just bear with me. I have a table in the lakehouse. One is the fact table. One is the dimension for customers. Once I establish a a relationship that’s not physically coded into the semantic model, once I just put down the idea that this column is related to this column and it has this one to many relationship, once we define what that is, you can actually run SQL against that because now SQL has context that these two tables are related,

34:05 these two tables are related, right? And then the tables that that relationship means if you wanted to talk to an AI agent or say hey agent build me a table that has customer dimensions and this factual data it would say oh I know what relationship to do the left join on and I understand that this is a one to many relationship. So that relationship can be interpreted down into SQL the same way you would build the relationship in the semantic model. In the semantic model, you physically build

34:35 the semantic model, you physically build a connection between the two tables. It’s stored inside the semantic model. So all this to say is when you get to this common understanding of what the model is, model is, what we’re doing is we’re actually burying business logic directly into the semantic model. And we need to pull that semantic model a layer up into something that’s a bit more generic, generic, right? right? And and this is where the open semantic interchange, which is an OSI. This is where this OSI comes into play and you

35:05 where this OSI comes into play and you can do a lot more definition like I can define or hang or attach additional data objects onto something that allows me to describe a SQL statement, a DAX statement, a measure, a KPI, a relationships between tables. And so really what we’re really I think what we’re doing here Tommy and I think everyone would agree if you just took the data away and just described how the data is related and what measures you’re building that’s something that’s relevant and then this

35:37 something that’s relevant and then this OSI standard and this is something that’s also in the lineage view. We have the ability today to to basically build an OSI standard out of a semantic model. but the the thinking here is if you have a semantic model that has just the descriptions of information about the semantic model or other models when you bring in many different models you can track all the changes between them. You can see which measures are the same or which measures are different. Let’s back up a bit because I’m going to ask I’m going to ask you a question here because as you’re talking I’m thinking about this. I mentioned earlier that I

36:09 about this. I mentioned earlier that I would be very conscious about what composite models I built. They are very powerful. But I don’t think this is something you just do ad hoc. I’m going to introduce a concept to you or a maybe a process and you tell me if you agree that an organization should do this this and I’m going to not necessarily I agree with this. So I’m going to I’m going to I’m going to proclaim out loud if that makes sense. I would state that every composite model needs a data contract. It needs a data handoff. and needs to have a clear place

36:40 handoff. and needs to have a clear place for some a user or the person building to go to look at the definition of that. Where is this coming from? What is the chain control that data contract must include when changes occur, who gets notified. It’s going to basically include all the things of what are the dependencies. Again, this is user specific for consumers and there’s also one for the builders to know how what do what the heck am I actually using here? And more importantly, because Austin

37:11 And more importantly, because Austin said this, I may build off of a golden semantic model, but I want to add my own business data, right? Well, I don’t want them doing that whenever they want just because they have a random thing they downloaded from Mailchimp unless they are in a very controlled managed self-service environment. So my statement here and again I’m not necessarily I agree with this wholeheartedly but I would imagine or I would want there to be some type of contract anytime a composite model is created over change control and

37:41 created over change control and dependencies. So what would you say there? So again just backing up from the even the lineage point of view just the creation of management of this there’s really nothing today Tommy that lets us lets us No but agents yeah or discover what models are composite right if if you think about like if you start opening that ball of wax around like so composite models let’s think about this way there’s there’s certain things in your organization that are certified there’s

38:12 organization that are certified there’s certain things in your in your organization that are promoted and certain things that are not tracked at all. all. Right? Certified stuff, I would argue, should never be composite models and only in very rare instances would you need to have that because it needs to be performant. I think you’re going to get slowdowns. I think users are going to do very interesting things with models. Unless you unless you really look at the use case for composite models, I don’t think a composite model is a good story for a certified data set. Except Mike, we talked about in 545, how can agents

38:42 we talked about in 545, how can agents help with data governance? If you want an easy way to scan with a composite model, I know. Yeah. So, so that’s a different note, but yes, that also assumes that you’re getting like you’re able to like you already have set up some APIs. You’re using the fabric unified admin monitoring. You’re getting models down like some you need to start having some way of absorbing all the data that’s happening inside fabric, which that’s a separate workspace. That’s setup time. I separate workspace. That’s setup time., okay, but I would argue I’m not mean, okay, but I would argue I’m not really interested in

39:16 really interested in managing a composite model at at the at that level. Like I’m not I’m not interested in managing a certified data set that is composite because I think there’s going to be too many slowdowns. You’re going to get bad performance issues. So, I’m going to say from there I’m going to say no. Yeah, Yeah, that’s a standard. But to your point, Tommy, if I’m in a department, if I’m in a a spoke of this hub and they’re using it, I might allow them to use it, but if the user base grows, I’m going to I’m going to very carefully watch the performance of that model. Yeah.

39:46 Yeah. Because odds are something’s missing and that composite model is not doing what we want. I’d rather build a master model than a composite model. My preference, you’re actually explain you’re actually explaining a awesome story of managed whole service, Mike., I

40:00 managed whole service, Mike., I love this and this was actually one of my bigger projects. I got I was had the opportunity to work on was very much similar to this where we had these gold semantic models that business users can use because they need to add their own data, right? Or they need to do some other manipulation to the model. Well, that handshake was very simple. It was if you’re going to do that, you have to create it in these specific workspaces. They had basically two workspaces for report building a certified data sets and then they had ones off for the composite models. Now the data team was

40:32 composite models. Now the data team was very very but there’s no way to block it though you can’t you can never block a composite model right you cannot block it yeah you can say I think you can actually say though how many no you can there is a setting in the actual model yeah in the real model you can say don’t build composite models off this one or how many you can build off exactly so okay yes okay yes that’s at the model level that we’re building at the certified level which gets away from manage self surface once you enable it yeah so Okay. So, what would they do? Because the data team did not like this. They’re like,

41:02 team did not like this. They’re like, well, no, we don’t know what they’re going to build, what they’re going to add to this, and then they’re going to start using that. So, the handshake involved usage to your point, Mike, and I think this is a huge also growup story for data and adoption where we would monitor because we had the audit log like, okay, we know that all these reports are being created, people are viewing like crazy in the business specific workspace report X. Now it’s time to investigate is met a threshold over a certain period of time that this

41:34 over a certain period of time that this is relying the department’s relying on or that set of teams we investigate it’s a composite model then we talk about okay we’re going to bring this into the data teams environment now because this is obviously essential what data are you using with this is it excel why is it excel this becomes a great story then also for let’s attack the things that are manual effort and that people are doing that are taking up a lot of time. Well, Mike, one of the great use cases we found with this was we had this

42:04 cases we found with this was we had this composite model like well what what’s the Excel thing? It’s like well we have the sales quota and everyone has to input this manually. We’re like oh no well we built an app we built an app that actually allowed people to manage the sales quota in the team. So it was integrated in the data system and became something that actually became certified. So there is a story here with composite and I however I would back up just to say there are specific stores there are targeted stories with composite models. I would not just generally allow it because to your point

42:35 generally allow it because to your point Mike you’re going to run into the bloating issues and more importantly I think you’re going to run into trust issues. I’m not sure I’m not sure if I’m not sure if I’m bought in on the trust issues. issues. I need to hear this. Well, I don’t what trust is I’m looking I’m worried more about performance, right? I think that’s kind performance, right? I think that’s where I’m going with this composite of where I’m going with this composite model piece of this. trust issues maybe because I’m fully

43:05 issues maybe because I’m fully assuming like if anyone is building something from that model and it’s slow it it that to me is like supersedes some of the trust pieces. And if you’re building composite models and you’re on the spoke of this we’ve already got agreements in place to say look this is the data we’re going to give you. You’re up you’re building your own thing., if you’re if you’re running queries on stuff, that’s on you., that’s the only thing I would be concerned about is my main compon

43:41 goes really really slow and starts really getting bogged down because of all the direct queries from other mini models, then I’m I’m actually looking at that going that’s probably not the best solution here. I don’t I probably don’t want that to happen. I I agree with the bloating issue, but let me try to emphasize the trust here because if I generally allow composite models without a story for them or potential story for them, it is very likely and something again just like you’ve been in the trenches, but when I worked internally, I saw this firsthand

44:12 worked internally, I saw this firsthand as well where these composite modules composite models all of a sudden it became very ambiguous what people were actually looking at because we knew it was off of a golden semantic model, but then it’s integrated with their own data. So like, okay, you’re saying sales here, but I can’t find where that number is actually coming from because that’s a sales measure that we used, but you’re using your own Excel file. So what the heck is that? We can’t we can’t check this, right? We can’t we can’t compare this to anything. So now you

44:42 compare this to anything. So now you have people looking at a composite model without that in a sense layer of verification or that layer of we know this is a composite model. it’s it has a potential story. They look at the actual universal,, reporting doesn’t match up. So, which one’s right? So, to me, I think trust is a huge part without a story. I guess that would be a better way to say it., not just composite models in general, but let’s actually move in because I want to actually go to how do you actually standardize?

45:12 how do you actually standardize? Let’s say your team’s like no matter what we are actually going to go down the composite model route right and these are always the scenario like unless you had something else to go through first okay what’s your take time or what’s your scenario time you are now in a situation Mike where like I don’t care what you say we are going to go down this composite model route we have things in lake houses but we have things that transforming there

45:42 things that transforming there Where would you actually start with a decision checklist to say are we ready to do composite models? And I’ll jump in here too as you think about this because stop it though Tommy there’s nothing that lets me stop composite models. I I again like are we are we ever really ready for composite models?, you can use them, but be aware of like you’re until until I have at least a

46:13 you’re until until I have at least a conversation like, “Hey, be aware. You can use composite models. They may be slow. be slow. [snorts] [snorts], that’s that’s it may be slow and and it may not perform the way you want because it doesn’t run like an import model. It’s not the same thing.” And again, I’m going to lean back onto that whole text interchange between two different islands inside the model. That’s that’s the thing. Yeah. So if you’re if you’re a spoke and you’re doing this, I would let you do it. Now, is this part of the core BI team design system? No, it is not. So like right

46:45 system? No, it is not. So like right like so I when I look at this question, I’m looking at going like maybe from two angles, Tommy. One angle is I look at this question and I go, would I ever encourage the use of these composite models using this pattern? No, I would not. I would not encourage this. Central central teams should not use it. Composite models are not something for you, don’t use it. If you let the the independent spokes or the departments build something or they figure out how to build a composite model, I’m okay with that. What I would do as a leadership or my guidance here would be is I would want

47:15 guidance here would be is I would want to be able to monitor all that, have ample monitoring so I could I could every day pull all models from the entire workspaces that people are using. go through and see if there’s any composite pieces and have a flag or report or something that tells me where our composite models are showing up, which is fine. Just monitor it. I’m okay that hap letting that happen. But if any of those models from those spokes need to get pushed up or I get any change requests because I need to have the central model changing,

47:45 central model changing, I would then flat out deny. We’re not,, one of the main rules of certified content is not composite models. Mike, I got something. thing I think you’re going to like. So, I think you’re going to like this story because you mentioned parts about the monitoring and that’s always I think been the hardest part with composite models or the hub and spoke like what the heck’s going on. Usually, it was either manual effort or the audit, but you had to be very proactive doing that. Well, let’s take something Austin said here and I think you’re gonna maybe the light bulb will

48:16 you’re gonna maybe the light bulb will go off on your head like it did mine. Well, Well, sure. We use PBIR pip format and use the git features and feature branches with local development. Huh. Mike, let’s say I had a something with a repository and I’ve always wanted to monitor any changes that occurred. What would be the best way to do that? Huh? Huh? Wait a second. We can do that. So Mike, I’m gonna introduce to you here that if I were if I were to have composite models, I’m going to mandatory

48:49 composite models, I’m going to mandatory again blanket statement here. I’m not saying this is for all use cases, but ideally I’m going to blanket I’m going to mandatory, what am I trying to say here? I’m require the word that’s the magic word. I’m going to require those teams use get for commit and pushes of any reports in that workspace. we can easily scan that stuff and know exactly what’s going on and it’s it’s all code easy to audit. So I’m going to install I’m going to have agents and skills in each of those

49:20 have agents and skills in each of those repos and going to say you must have these agent in the GitHub and or docloud however you want to do it that will scan it. it’s a GitHub action however you want to do it that will actually pull back provide the documentation on what’s there combine that with the usage that’s something that for me from a centralized team or from a data governance point of view I don’t care if you’re enterprise at this point we can push this all to a central place so we can monitor hey are people actually following the rules are people going off the rails combine that with

49:52 going off the rails combine that with the usage so for me I I think git is now allowing us to do data governance It’s better.

50:00 better. I I Yeah, I would agree with that 100% Tommy like and that would be another I really do like that any So in the story the hard requirement is you want to build stuff from our main semantic models. I’ll give you a workspace and it’s automatically going to be attached to get 100%. get 100%. And we teach you the bare minimum of at least you can build in the service you can build like whatever you want to do build what’sever easy. You can even build a desktop, not a problem. But but that way at least I can observe when things are checked in and when things

50:31 things are checked in and when things are not are not and at least I get a copy of everything you’re building. So yes, I agree wholeheartedly. I think this is a great win for for monitoring this one. Going back to maybe Austin’s question specifically though is I would as the central BI team I would not build a monolithic large many table model because that’s just not easy to use as a as an end user. And what I would do instead is I would rather have a domainbased modeling. So you may start

51:03 domainbased modeling. So you may start with a really big model and as that model grows you may need to realize you need to peel off portions of that model to make smaller mini models. And so I would either work with tabular editor I would work with Michael Kowalsski’s master and mini model where you have a single model that has everything in it and then you carve off. So I think this is actually a governance question for central teams. Yeah. Yeah. The governance question is do I make one monolithic model for everything and then from that monolithic model do I build copies of that model and how do I

51:34 build copies of that model and how do I keep them in sync? And I I think that’s the problem I’m trying to solve with lineage view. I’m trying to solve that problem. I’m trying to say pick your mega master model that has all the things in it. select portions of that one large model and then publish a mini model. It’s literally an actual data source. So like what we want is we want a graphical interface that sees the entire holistic model tables and measures and then peeling off portions of that that you

52:04 peeling off portions of that that you then publish to many models. All this can be used through timol. So that that’s the story here. It’s funny, Mike, because I hear I you sees all the buzzwords and products people create, not yours particularly the companies, and they say data store, and I’m getting more, maybe it’s just me getting older and grumpier, but you should not allowed words in your name if it does not exist. Like, I cannot tell you how many companies tell me they have a data store and I take a look and it has nothing. No, not a data store. You’re like, “Yeah, you can look at your model.”

52:34 “Yeah, you can look at your model.” Okay. Okay. That’s like me. That’s like me going to Tiffany’s to buy my wife a ring and not buying like just looking in the window basically. You’re like, “Okay, yeah, there are products here that I’m not purchasing.” Anyways, so that’s that’s a little off the point. No, but I agree with the lineage thing. I think there’s a combination here. So, I’m going to wrap up some of the things answer Austin’s question. Is the structure right? Yes, but if you are intentionally building in a managed self-service environment, I would not recommend composite models in this format if for just the hub and if you do

53:06 format if for just the hub and if you do not have people building it. So if you are not doing managed self-service now Austin, Austin, you really need to make sure that’s becomes a priority for the data governance initiatives for the teams. The other side of this is the Git integration lean on hopefully your company has enabled GitHub copilot or something like that in some capacity because if you want to be able to monitor this you need to have that integrated with usage to tell grove stories and really the last part here too we

53:36 and really the last part here too we didn’t really discuss as much today but with direct lake and with the ability to connect to different lakehouses I think there’s another story there as well with having multiple tables from multiple lakehouses. But for the sake of at least the argument here, I would say you manage self-service has to be a must. A growup story in a data contract data handshake and you have to use git in order to monitor this. So that’s my wrap. But Mike, Mike, where do you land on Austin’s question here on the

54:06 land on Austin’s question here on the pattern and how would you yeah go about it? What’s your take? Yeah, I’m I’m going to say the the wrap-up here is don’t build a monolithic model. that’s not that’s not going to be useful for anyone., you going to be useful for anyone.,, use composite models in a sparing know, use composite models in a sparing way. Don’t go overboard with them.,, use them in very specific use cases. Go watch Kevin Arnold’s article or video as he teaches about where is the best case to use composite models, when you want to maintain them, how you need to maintain them. The core function of that is what data are you

54:36 of that is what data are you transferring between island islands, right? If it’s a high cardality column that you’re trying to go between different islands, that a composite model will not be a great solution. If it’s a low cardality column, probably not a problem. So that’s that’s something just to be aware of when you build these composite models., I really like this idea of the master model and mini models. If you have questions, hit me up. Let me let me do a demo for you. Let me show you how Lineage View helps you maintain and manage these these models. data

55:07 and manage these these models. data governance is is a bearer and you need to either again with with copilot you can invent your own tooling if you want to invent your own tooling hey this is a master model I’m going to select these kinds of items from it and I want you to build a timal file that’s a mini model dude that should be straightforward and easy you can already do that in notebooks with Michael Kovalski’s tools right so right so stuff already start is starting to exist to have a centrally managed definition of all the tables and relationships and I think that’s going to be useful to later on as you build mini models from

55:39 later on as you build mini models from standard stuff. So anyways, all that said, I think that’s a my approach. I would leverage master and mini models more than I would anything else. Build the the centrally designed model and then carve off domain based things for those specific users. That way you can manage one thing centrally and you can carve off or repeatedly deploy those items over and over again. Right? So,, and if you don’t understand or you want more tooling to have,, support for this, hit me up.

56:10 have,, support for this, hit me up. Let me know because I’m building I’m building actual tools around this with our lineage view. We’re trying to make this a reality. We’re trying to figure out how to make this become like an actual, feature of fabric where you can easily control the master and mini models. models. And I really like I really like what’s happening here, which is, your models aren’t large enough. You’re putting all your data engineering in fabric, making a lakehouse with real tables, and then jumping from tables, importing them into

56:40 jumping from tables, importing them into models that are in the pro workspace. That’s amazing, right? right? And if your company’s if your company is very small, I’d highly recommend you substantially look at if you have like 5, 10, 15, 20 people that are interested in using PowerBI, you may want to seriously look at using premium per user because your model size goes from 1 gig to 100 gigs in size. massive models and the mini model issue goes away. Like you have zero con I have zero concern

57:11 have zero con I have zero concern anything you can [ __ ] can can generate at the master model can be easily deployed as a mini model somewhere else for other users and other teams. So I I really I think this is an underrated story and not many organizations are looking at this and I I think it could add a lot of value. Perfect. All right dude. Well this is awesome. I think hey talk to us talk to Mike about the lineage view and what is the workload free to the workload’s free you can go test it out you can go in you can go play with it right now if you want so the only

57:41 it right now if you want so the only caveat we have right now is you can go play with the lineage view and go use it right now if you want to save your lineage view if you do some work in it that requires a tips plus membership which is $3 per user for for less than a price of coffee you can go ahead and use lineage view save these things these items or artifacts that you’re building can totally be saved So,, fully aware, fully aware, I’m trying to make it a very a cost effective tool to use., but as soon as you want to save these Lineage View items,, we we have you sign up for the subscription. Anyways, regardless, I think it’ll be

58:12 Anyways, regardless, I think it’ll be super useful., and help out users here to help you manage and distribute big large models into mini models for domains or spokes. Dude, this has been fun. This all professional by us, by the way, for once. for once. I know. I was about to say only took us 550 episodes to get to professionalism, but, we got there eventually. I I don’t don’t again don’t push. I don’t think we’re going to this going to allow this episode. I’m not saying we met the threshold. I said it’s like you just went over the speed limit once and

58:43 just went over the speed limit once and then it just goes back down. So, So, we we’ve peaked a little bit, but then we’re coming back down. Well,, this is the best we’re going to get. It’s all downhill from here. Amazing. Awesome. Also, thank you, Austin. Thank you for the question. We really appreciate you giving us the question around this one. A good topic. I think composite models are useful. They have their place. but I think more often what we’re trying to build is master many models. And I think that’s that’s the story we should really be honing in on figuring out how that works. That being said, Tommy, thank you

59:13 works. That being said, Tommy, thank you so much for the conversation. Where else can we find the podcast? You can find us on Apple, Spotify, wherever get your podcast. Make sure to subscribe and leave a rating. It helps us out a ton. If you have a question, idea, or topic that you want us to talk about in a future episode, well, head over to powerbi. tips. tipsodcast. Leave your name and a great question. And finally, join us live every Tuesday and Thursday, a. m. Central, and join the conversation on all power. Tip social media channels. Thank you all so much, and we’ll see you next time.

59:50 Dance to the laughs in the mix. Fabric and A. I get your feels. Explicit measures. Drop the beat now. Pumpkins can’t steal the crowd. Explicit measures.

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.