Microsoft Fabric: Early Case Studies and Business Impact

Rob Collie: Justin, I can't help but notice that we're both wearing the same P3Adaptive polo today, which is a rarity.

Justin Mannhardt: Fun fact about Justin. I usually keep a P3 polo in my backpack.

Rob Collie: What?

Justin Mannhardt: In my role, you never know when you're going to have to jump on a call with a prospect or with a partner.

Rob Collie: I just have this vision of you walking around town in your civilian clothes. Then there's like a data emergency, out in the street.

Justin Mannhardt: He couldn't find a phone booth.

Rob Collie: There are no phone booths anymore. You have to jump into your own backpack and transform into P3 Man. All right, so in terms of real things we're going to talk about today, you and I recently had a call with a prospective customer to talk about Fabric specifically. It was a really great call, and as always, listening to you talking about something with a customer or a prospective customer, I learned things and I earmarked a question for myself that I didn't want to ask on the call, but now I'm going to ask it on the show.

Justin Mannhardt: Lay it on me.

Rob Collie: We do have half a dozen or more accounts where we're doing something very explicitly Fabric related right now.

Justin Mannhardt: I think that's about right.

Rob Collie: And you mentioned that they follow the two buckets, and listening to you talk about the those two different buckets, very keenly interested in the contrast that you described between those two. But also I wanted to know more about for the identified business case, what those identified business needs were. And I figured this would be something I could ask you on the podcast because other people would be interested in the answer to that as well. So it's kind of a two-parter.

Justin Mannhardt: I do think the context of the two buckets, and you could make debates that there's even sub-buckets within the buckets, is really informative in terms of how we and our customers together are approaching Fabric and trying to get value out of it. And then the specific use cases are interesting too.

This is not an original thought of mine. I just want to give credit all the way back to the Microsoft product team because this is a question that got posed on a call that I was on during the early private preview stage. If you've been keeping up with Fabric, you do know that a lot of the workloads that are available in Fabric are workloads that were maybe available somewhere else, like Azure Synapse or Azure AutoML, for example. And so someone posed a question like, "Hey, if we've got a customer that is either starting or very far along or even very happy with their implementations and some of these other services, should we be concerned about that? And should we be shifting gears over to Fabric for these well-in-progress projects or even completed projects?"

And the advice from the product team was, "No, we think you should be thinking about your roadmap and the projects you haven't lit up yet as the means to get started." And that really clicked for me. And so when I put on the business leader hat, I would break the first bucket down here for a second because I think that's something we didn't get into on the call you're referencing. If you've got workloads, data models, dashboards, reports, things that are built, they're serving your business well, of course there may be some things that could be better, they could be more efficient, but if it's not hair on fire mode on those things, your benefit of moving over to Fabric to me feels a lot lower than the benefit you're going to get by introducing new capabilities to your company.

Maybe you'll save some money because something's going to be less expensive in Fabric than it is over in Azure, maybe some things will run faster, but you really need to assess is what's happening broken in some way before you go down that road. And the reason I say that is Fabric's still a relatively new platform, and so there are still feature parity things. And so you got to ask yourself, "Do I want my teams wrestling with all these feature parity gaps to maintain something that I've already got versus bringing some new capability to life for the company?" So I think that's important context on the first bucket. And so where I've been steering a lot of our customers is, "Let's go look at your roadmap." One, because I do believe in let's add more value to the business unless there's a real compelling reason to go migrate some of these workloads that already exist, let's add more value to the business.

The other thing is Fabric starts to open the door to lots of possibilities that didn't exist yet. The Lakehouse idea and the way data can go into data models, it can go into AI, it can go play with other third party things. We just have more opportunity. And it also presents maybe different ways of thinking about how you go about projects. And when you're trying to migrate something exists, it's very hard to ignore how it's already built. And it's easy to get caught up in that. But when you're doing something new, you just more naturally have some more freedom, more willingness to experiment. Even the best of us, we can get really caught up in, "It was set up this way and we need to make it work this way again." If you do eventually want to be all in on Fabric, you do want to retire other technologies, figuring out how Fabric's going to work the best. I think you just stand a better chance if you're starting with something the business isn't getting value from yet and figuring how to set up your environments, figuring out all that stuff on something new? To me feels like a smoother path.

Rob Collie: Yeah, I just had a pretty intense visual in my head in the story you were telling. When you're implementing a project for the first time, whether you realize it or not, you're kind of like navigating a maze. You're sending out a number of feeler tracks through the maze.

Justin Mannhardt: Yeah.

Rob Collie: Eventually one of those tracks does connect with the endpoint on the other side of the maze. And then you just sort of forget about all of those feeler tracks. You just bake in this one path and that's what you want. Once you find the way to solve the problem, you're going to commit to it. But then when you want to move it to a new platform, the tendency is to want to take that path from this other maze and lift that path and move it over here and drop it on a completely different maze. There might even be a much, much shorter path through this second maze than your original path, but you're so committed to moving that other path.

And it even came up a bit in our podcast with Christina when she was talking about the navigation from Tableau to Power BI at one of her previous companies and how the Tableau expert got really upset about there were little features, specific and awesome to him features of his Tableau reports, his Tableau dashboards, that couldn't be perfectly re-implemented in exactly that same way in Power BI. The real goal is the same, but the implementation doesn't have to be. If you allow yourself to explore that platform and sort of do the optimal with it, you're going to find a different and sometimes more optimal, hopefully more optimal solution. And plus, just in general, you zoom back, technology in search of a problem? How many times has that been the way? So instead of technology in search of a problem or even an opportunity, let's talk about problems and opportunities. Let's talk about that second bucket. What have you been seeing in that bucket?

Justin Mannhardt: It's really like the things we haven't done yet. So an example, one of our customers, we did a lot of work for them many years ago. So they've got data models for lots of functional areas in their business. They've got AR and AP reporting, they've got operational reporting, they've got a lot of things. When they were interested in Fabric, "What do you not have?" "We don't have a great solution for our supply chain team. We don't have a great solution for this other compliance team." We were really focused on where is the pain still the greatest? So that's a way for you to start thinking like, "Okay, if I've got things in Power BI, who am I not yet serving? And can I use that as a catalyst to figure out a couple things that I think are true, like can we deploy faster with Fabric compared to what we would've done with Power BI plus Synapse plus other things? Can we be more iterative? Can we set up the data in such a way that we give people more autonomy and more capability?"

So for example, yes, this team needs reports and dashboards to help them understand the movement of material and what they need and just to be smarter about doing their jobs. But they also need to be less dependent on the Central BI team. I'm sure our listeners, you can imagine the bottleneck that happens when every time you need something new, you got to go to somebody, you got to wait. And so Fabric opened up a possibility that wasn't really there before, having my semantic model sitting on top of a Lakehouse. There's more like draggy, droppy, self-service capabilities that exist in the total Fabric ecosystem than just in a Power BI reporting environment alone. They also have more tools to dig deeper without always having to go back to some sort of central BI team to get new stuff and new insights. And now they're also more self-sufficient. Now you sort of see, "Oh, this is how we could do this over in the compliance team and this is how we could do this over in this team."

And then I think if we do that enough now we're creating fresh momentum that if we carry back all the way around to, "Oh yeah, the AR and AP team that is not as painful as supply chain because they already have some set of Power BI solutions." We can come back and say, "We've got fresh momentum on a new way to serve you. Now let's come and approach that with fresh eyes." So instead of recycling things that already exist, create some momentum and then come back around. I don't want to advocate for leaving those things for dead.

Rob Collie: Yeah.

Justin Mannhardt: But I think you run the risk of not creating fresh momentum for yourself and your org.

Rob Collie: If the lift and shift starts to run into obstacles and it's not a lot of fun was. Was Direct Lake involved? If Direct Lake's involved, there's all kinds of changes I can make to data sources that doesn't require some refresh chain to go pick them up.

Justin Mannhardt: It opens up other attach points with far less consequence and confusion. Quick architectural background, so all of the data is coming into a Lakehouse, and then maybe there's some more transformation that's happening on its way up to a semantic model. The semantic model is running our core reports and dashboards, but if I'm a supply chain analyst and I've got a cool idea or I'm insatiably curious about a question, if the model isn't set up to answer that, maybe there's something missing from the model? I, as the supply chain analyst, I can go downstream, play around without jacking up the semantic model and I can do that without being in Power BI Desktop. There's all sorts of things in Fabric where I can run ad hoc queries, make simple reports, play around and find out my idea, that doesn't require me to know every little thing about building models, writing DAX, being in Power BI Desktop, more like Excel user-friendly things at my disposal that I didn't have in a pre-Fabric world.

Let's say if I'm in supply chain, I can just go to the table in the Lakehouse that has all the material numbers in it and I can very quickly visualize that data on my own, without building a model, without writing more DAX, without doing other things, and I can explore that without needing to understand how all the intricacies of a semantic model are set up. So I can explore these little things like people would dump out into Excel, Rob, all of this can get saved. I dumped all of our opportunity pipeline snapshots into a Lakehouse. I didn't build a model. I just said, "I just want to get an understanding of this." And so I was able to build a Power BI report on top of all of that without doing anything special in Power BI Desktop whatsoever.

Rob Collie: There's a Power BI semantic model out there, and I've got a question that isn't addressed by existing reports. Maybe the existing model doesn't support the creation of those reports. If I don't have permission to modify the model, then I'm kind of out of luck. In certain organizations, I have the ability to export certain visuals to Excel. This is all a pre-Fabric world, right?

Justin Mannhardt: Right.

Rob Collie: But if there's something in the model that I need in order to run my one-off custom analysis, and it's not represented cleanly in one of the published visuals, I'm kind of at a dead end.

Justin Mannhardt: Yes.

Rob Collie: So now I'm that same person except that this time, this set of reports I'm looking at and the semantic model behind it, is now with more Fabric. So what is now available to me in this situation that wasn't available to me before?

Justin Mannhardt: Pre-Fabric, you likely didn't have access to the original data.

Rob Collie: Right.

Justin Mannhardt: You likely had access just to what happened to it and then it's up in the model. Now I have access to both the model, the original data, and things that have happened in between.

Rob Collie: Okay.

Justin Mannhardt: So sometimes when we build semantic models, we make intentional choices to aggregate the granularity, remove certain things, filter certain things out. And so now, this is very common where people will be like, "Hey, this group of materials is not showing up in the report where I thought it would." And they'd go, "Hey, Rob and BI team, can you explain this to me?" And go in your desk and you'd get around to it. Now this person can just, "Well, I know it's here because I can see it in PeopleSoft or whatever the system is, and I can be more self-sufficient in figuring out what's going on there. Either I'm just answering an ad hoc question or I'm going to end up proposing new business logic back to someone that is going to change the model for me. That's that idea of decreasing the dependence on a more highly skilled team that's doing your models and your high-intensity DAX and things like that.

Rob Collie: So what is it that I get to look at? One of the tables in the semantic model? Do I get to look at the data sources that were ultimately used to construct that table in the model? Those are also sort of more transparently traceable and potentially visible, but they have the opportunity to be, I'm assuming read-only?

Justin Mannhardt: Yep.

Rob Collie: But I can kind of see the pipeline. I can see the different steps in the pipeline that result in the semantic model and connect to those steps to spin up my own little one-offs to prove/disprove that something's possible, not possible, or run an analysis that wasn't supported by the semantic model. Am I getting it now?

Justin Mannhardt: You are getting it. Supply chain, there's a reasonably high frequency of ad hoc requests. "What's going on with this? What's going on with that?" Now, hopefully your semantic model can answer the vast majority of those questions, but sometimes you run into a situation where it doesn't.

Rob Collie: It's also a point in time, like it doesn't answer those questions today.

Justin Mannhardt: Right.

Rob Collie: And after we run through this exercise, there's probably a circle back process where we do evolve the semantic model to address this case, but what to do in the meantime and how to explore it, the pass through the maze. If we have to wait to explore it in the context of modifying the semantic model, we might be waiting a while.

Justin Mannhardt: We might be waiting a while. The moment may pass. And so I just think it's cool, the technical steps that I could go connect in a read-only fashion to a table or tables that ultimately my model reads from. Maybe it's transforming some data in Power Query as well, but I could grab that either by writing a query or just drag and dropping that data and still creating Power BI visuals with it. This is very different to me than like, "Oh, I dumped it out to Excel." I can still create interactive visualizations with that data, understand it better, share that with other people, and it might be just a moment in passing, like "Okay, good. We answered that question about what's going on with warehouse B. Done."

Sometimes, to your point, it'll be like, "Yeah, we need to circle back and make sure we're changing how we're handling the materials dimension so that we've got more fidelity in the categories we're using," or something because it blocked us from understanding this issue, for example. So when we build semantic models, at some point, you make intentional design choices that you may realize later like, "Oh, that intentional design choice serves all these things really, really well, but it's not quite right for this particular question." And so the ability to quickly go back downstream, still answer that question without asking for this big overhaul, that's a really powerful capability that teams are benefiting from. I think it's making them faster in collaboration and faster in iteration with their solutions.

Rob Collie: One of the whole themes of the Power BI revolution is getting the creation of BI closer and closer to the stakeholders. Most of the analytics that I consume at our business are either things that I would never want to modify or are things that I myself built. Direct Lake, for all of its pros and cons, the idea of federating out certain portions of the data update process so that we don't have to run a whole big refresh or wait for a refresh or whatever to change some parameters or something like that, especially the human curated data sources.

Justin Mannhardt: Oh, God, yeah.

Rob Collie: Having that just kind of flow through is kind of nice. And again, Fabric is new. We are doing significant work there. But it's still just a fraction of our customer base. And even within that fraction, some of them are doing lift and shift, some of them are doing specific business case things. Are there any business case examples of where we're using Fabric because it provides a better solution than vanilla Power BI that doesn't involve these tiered levels of teams?

Justin Mannhardt: Another one of our customers essentially runs, let's call it a membership organization. But a very small team, probably less than two hands worth of full-time employees at this organization. But they have lots of members that they're serving and what they do. And their data comes from all sorts of fun and interesting places. It's not always easy to get.

And you can imagine most of these people, they're focused on what they do for their members, not technical. Sure, there's the person that got a little good at this or a little good at that. And so when a lot of your data is in these SaaS systems in the cloud, you have usually two options, three, if you're lucky. One, there's some sort of very bad reporting baked into that platform, and if you're lucky, you can export some percentage of that to Excel. Or they'll say like, "Oh, sure, you can get a developer key to our API, then do your thing." Two thumbs down on both options. What they were struggling with is the reality of when you try and export data from these systems, you can only get so much of it. You can't export everything all at once.

Rob Collie: You get a window.

Justin Mannhardt: Yeah.

Rob Collie: Pick your window.

Justin Mannhardt: We've talked about this on the show with regular old Power BI it was also difficult to get all of the data. Fabric, not even with Python code and all that stuff, even just general gooey informed pipelines with the data factory stuff, "Oh, here's finally a way. Here's a way that doesn't feel scary to our company, to incrementally get data out of these systems so that we have it."

Rob Collie: Awesome.

Justin Mannhardt: I think that's maybe the example where things came way in range. I mean, I'm all about the let's just export some data to CSV files and get going on a solution. But when you're accumulating new data all the time or data is changing and you need to keep getting that? It starts to become a burden. And so I think Fabric's not overly technical. It's not completely scary. We're not creating resources in Azure or figuring out the cloud. We can go do this. That's a really good feeling for me is when people that before just didn't know if there was a way, they're like, "We feel like we can do this." It's really cool.

Rob Collie: I'm excited about stuff like that. Our podcast stats are an example. I'm pretty sure Kellan did that for us, but I could have done it. It's pretty cool.

Justin Mannhardt: Right. Earlier when we were talking about the buckets, do I migrate something I've got out of Fabric or do I go something net new? You gave that analogy of the maze. I thought that was really good. And how important it is just to know what formulas you should be writing versus how to do it. It's almost like, what do I need to accomplish versus how to accomplish it, just to generalize it. And I think the tie in here is it's easier to focus on what when you don't have a bunch of how already staring you back in the face.

And what I mean by that is if I'm looking at something that's already got a bunch of code, there's an existing solution. When you're working on a solution, if you're a Power BI person and you're challenged to come up with a new metric or figure something out, the first thing you're doing is staring at the formula pane? You might be off to a rough start because if you're already thinking about which functions do I need to be using? You haven't cracked the logical puzzle of what do I actually need to get to in the end? I know I stand almost zero chance. Especially if I'm already looking at some code. If a customer called me like, "Justin, can you help me troubleshoot this DAX?" And they pull up the DAX on the screen and it's maybe a longer formula. "Can we get out a whiteboard? Let's draw what we're trying to achieve first before we look at any of that stuff."

Rob Collie: Yeah, I mean, I'm the same way. I have to know what's going on, what's the intent? Even before my brain works, in the end it is a change to the DAX, that in hindsight, I "could" have looked at the DAX and said, "Ah, there's your problem. You've got your variable declared in the wrong place," blah, blah, blah. Right? I'm so fuzzed out beforehand. But I think the other point you're trying to make is that once you get to that level, your blinders are really sharply defined. We're watching the Hulu series, The Handmaid's Tale.

Justin Mannhardt: Oh yeah.

Rob Collie: I really can't recommend it. I mean, it's just misery. But Jocelyn actually really likes it. Anyway, so in the show, the handmaids wear these, they call them their wings. It's like horse blinders. You're always wearing those.

Justin Mannhardt: Right.

Rob Collie: With your business problem. Just a question of how big are the wings, right?

Justin Mannhardt: Yep.

Rob Collie: And if you're looking at code, you're in this tunnel that's two feet in diameter and 30 feet long.

Justin Mannhardt: Yeah.

Rob Collie: Looking at the problem and the other ways to solve the problem, but even better ways to approach the problem in general, regardless of code. What formula should we be writing? The formula that I mentioned in that previous episode that I was writing, the moving average? I've been modifying that formula since then. The moving average has already been clear to me that it's not lining up with reality with what we're actually seeing. So I've been going back to it and modifying the whole definition of it because now I'm thinking about the problem at the problem level and what we're actually trying to do. Even one episode ago, that formula hasn't survived.

Justin Mannhardt: Yeah.

Rob Collie: It's gotten better, and it will undergo two or three more iterations before we get it really dialed in.

Justin Mannhardt: I think the call to leaders out there is, you said this so well, tech in search of a problem. Because that's so strong right now. What are we going to do about Fabric? What are we going to do about AI? What are we going to do about these other things? And without fail, every time I'm stuck, if I get myself into a non-technical media a whiteboard, a notebook, a slide deck, or an empty Excel sheet or whatever, and I just, "Okay, forget about the DAX or the SQL or the Python or whatever it is. Can I just walk myself A to Z with what I'm wanting to calculate and measure and show?"

And if I can get my head wrapped around the logical of progression of what I'm trying to do, then I start answering the questions. "Well, I need to filter this table or I need to build this other table along the way or I need a new fact table in my model that kind of looks this so I can do things in a different way." And that was just a missing piece in my brain thinking about that idea of, "Well, how do you get to what formula to write?" Well, it get yourself away from the place that you write the formula in. You do a lot of things in PowerPoint as a way to sort of brain dump and get your ideas out. I like Post-it Notes and whiteboards. It's just the right thing to do.

Rob Collie: I am trying to constantly clutter our house with all kinds of just plain white pieces of paper, eight and a half by 11, and I'm constantly writing, sketching, drawing all the time. And then there's a collection process. Jocelyn comes along behind me and says, "Do you need this?" And I go, "Oh, no." Or even more appropriately she'll say, "Hey, your desk is really covered with papers. Can you go sort through those and recycle?" And it turns out I go look at them and I can recycle all of them. They served their purpose. And then if good stuff came out of it, then I took that and sort of ported it back into the digital world, did something in Slack or a Power BI report or Word or PowerPoint or something like that. It'll make its way back into digital if it's worthwhile.

Justin Mannhardt: We used to have a rule, it was an informal rule at P3 with, this is a few years ago on the team I was managing at the time, we called it the 24 hour rule. It just meant nobody was allowed to suffer with a technical problem for more than 24 hours without piping up about it. Because sometimes you just need to get pulled out of the fog of war. For leaders out there, especially when you're thinking about Fabric or AI, the amount of time you're spending in the analog, making sure you're clear about what it is you're trying to do, is immensely valuable. And I'm not talking about requirements. I'm talking about figuring it out, figuring out what to calculate, how that calculation needs to happen, what data needs to come in and go out. And getting clear on that? You're just going to set yourself up for success from the jump.

Rob Collie: Wow. Thinking in analog, acting in digital. I love it.

Justin Mannhardt: There's hundreds of DAX functions. I don't know them all. I've not committed them all to memory. But if I get into the research train before I'm clear on what? I'm going to spin a little bit. Is just the right formula. Well, I don't really even know what I'm trying to do clearly yet. I have a vague idea.

Rob Collie: I'll put on my grumpy old man hat for a moment. When I wrote that book in 2012, and again in 2015, we didn't have fancy things like whatever. But all these new things have been introduced in response to a clear customer need.

Justin Mannhardt: Right.

Rob Collie: And it's funny that the functions that are introduced, they don't get named in the long form that you would like. Imagine if every DAX function was named along the lines of, "You know that problem you have when you get into this situation and you wish you could do X,Y,Z, but you can't." They have to come up with this really cool, elegant sounding function name, like TREATAS.

Justin Mannhardt: That's a good idea for a desktop calendar where every day it's like, "You know that function?"

Rob Collie: And you're like, "Yeah, I do know that problem."

Justin Mannhardt: TREATAS.

Rob Collie: When I eventually found out what TREATAS was for I was like, "Oh my God, I had that problem so many times." I like time travel back to 2011, and I'm staring at a formula editor trying to figure out how to do something. I'm like, "Oh my God, I wish I knew how to make the subset of that that was also in that and I can't." That's the TREATAS function. That's all those moments of anguish.

Justin Mannhardt: I think it's a good reminder just a lot of us are finding ourselves staring at new formula editors in a way, like the entries to these AI tools or new places in Fabric where things happen. We're staring at these big open boxes where words or code needs to go and it's easy to get drawn in and be like, "Oh, this is the thing I need to be working with and doing. It's here, it's here." And really it's like, "Well, don't be there yet."

Rob Collie: Grab a pen, grab a whiteboard.

Justin Mannhardt: The power of Post-its. I like Post-its a lot.

Rob Collie: I've seen some of your works of art. You are a Post-it Rembrandt.

Justin Mannhardt: Yeah, good reminder. All kinds of new tech, all kinds of new formula boxes. Get away from it from time to time.

Rob Collie: Go analog.

Justin Mannhardt: Go analog.