Using AI for Data Viz – Ep.561

On Explicit Measures, Tommy Puglia and Mike Carlo pick up Rituparna Das’s Nightingale article on three ways to use AI in data visualization with the tools most analysts already have. The same episode covers a Fabric Data Warehouse medallion best-practices post and a private-network path for copying data between Fabric and Snowflake.

News & Announcements

• Using AI Harnesses to Harness Microsoft Fabric — Thursday, September 24, from 3:00 to 5:00 PM Central, Tommy Puglia hosts the Chicago Fabric and Power BI user group at the Microsoft Technology Center, 200 East Randolph Drive. With two weeks left, the session is an end-to-end demo of a two-harness loop: a second brain and MCP servers, taking a client meeting and the project context through instructions that can build a pipeline or a Power BI Desktop bridge. Register on Meetup; he said people have already started.

• Fabric data warehouse best practices for medallion architectures — Tommy reads part three of the Fabric Data Warehouse medallion series — part one, already covered on the show, was choosing the pattern, and part two was the bronze, silver, and gold layers — and he treats this post as a case for operational discipline that keeps the workflow predictable, reproducible, and reliable: set-based work, bulk copy through pipelines, files between 100 MB and 1 GB, repeatable insert, select, and merge steps, quality checks for required fields, formats, duplicates, and reject reasons, and a business-facing gold layer, whether that is a star schema, a wide table, or a pre-aggregated summary, with the warehouse left on its defaults as a delta-managed SQL engine in OneLake. The failure mode he cites is the one where bronze starts cleaning, silver starts serving dashboards, and gold spends its time repairing the layers above it. Mike hears a medallion a lakehouse can follow just as well, he cannot find a V-Order control inside the warehouse and suspects that line was written for Spark, and the August billing change he describes — a 20-second query rounded up to a full minute — is why bursts of short warehouse queries worry him, so he keeps bronze and silver in a lakehouse and treats the warehouse as the fit for a team that already knows T-SQL and will pay more to stay there.

• Connect securely to Snowflake in Fabric pipelines and copy jobs — Fabric pipelines and copy jobs can move data to and from Snowflake when public network access is blocked, using trusted workspace access, a private link on the workspace, an Azure staging account, and Snowflake’s private-link storage integration so the Fabric workspace, the staging account, and the Snowflake endpoint can all stay on a private path. Tommy’s setup list is the virtual network and private link service, the staging storage account, Snowflake with public access closed, and a copy activity or copy job, with the limits he flags from the post: the region has to support the pattern, private connectivity requires Business Critical or higher, and the private endpoint has to be approved on the storage account. Mike calls it a real option for teams that keep the public network shut, notes that Snowflake has been getting easier to use beside Power BI while Databricks still feels stuck, and he would build the application on Fabric.

Main Discussion

Topic: What AI is for in data visualization once the output is a Power BI report

Rituparna Das’s Nightingale article, Three Ways I Actually Use AI in Data Viz, came out of a data visualization conference where the hallway answer was MCP, headless analytics, and conversational analytics, and almost nobody could say what those words changed in the work. The tooling is fragmented, organizations are cautious about what data employees can hand an agent, and the tools getting the stage time are often the ones most practitioners will never open. She wrote the practical version for a Gemini subscription and Tableau Desktop. Tommy Puglia and Mike Carlo take those three workflows and ask what they become when the result has to live in Power BI, on a semantic model, as a page of related visuals.

• The conference words still need a workflow behind them. Mike can demonstrate MCP. He opens VS Code, saves the PBIR, and asks the agent to add pages or visuals, and he is also wiring an agent into Penpot so it can work inside a design system on backgrounds, images, and shapes. Headless analytics he cannot pin down — an agent talking to a semantic model with no report on screen is as close as he gets, and the phrase still sounds repeated more than defined. Conversational analytics he can use: the system does the calculation, and the person stays in a clear interface, talking to an agent about the data. Set the product names aside, he says, and none of the three tells you how a report gets made.

• Requirements are written so the agent can use them. Tommy treats the meeting itself as part of the build. He asks questions the skills and the conversational tools will need later — what the semantic model already shows, where the bottlenecks are, which measures exist, how the visual should read — so the notes become context, and the agent receives something clearer than “they want a bar chart.” Mike hears that as the second pass of discovery. The first pass stays wide, because early on there is no bad idea, and the second pass checks the model: a request for contact-center data goes nowhere when the model has none, until someone names the source and how it joins what is already there. The conference habit he wants to drop is starting at the visual. Discovery comes first, and only then do you ask whether the model can return what you found.

• The useful ask is specific, and the useful output is a page. “Give me a sales report,” or “why are sales down,” is loose enough to come back with real numbers that do not change a decision. Mike wants the two or three factors in a quarter that moved sales, grounded in particular information, and then an agent that helps choose the visual for that finding. Tommy agrees the time that matters is the time between the idea and the result. Drawing the bar is incidental. The work is taking the exploration, the measures, and the client’s pressure points and landing the related visuals on the page together. Das’s three examples each finish on a single visual. That is a smaller unit than the reports they ship.

• Sheets and Gemini are the discovery pass. In the first workflow, Das has sales-style data in Google Sheets, talks to Gemini, and gets a table, a pivot, and a chart, with follow-up for annotations, legends, and formatting. Tommy maps that straight to exploration: you have the data, you look until a pattern shows up. He points at the Power BI data visualization world championship, where the story has to be found in a provided dataset — many tables, measures tried in combination, trends over time. Mike’s caution is the assumption you carried in. They have watched the data contradict the story they expected, which is why he would rather let the data speak before the narrative locks. He also separates a one-off visual, the campaign and the dates you call out once, from a sales report that has to keep framing the story as the numbers change. Storytelling in that second sense is a poor fit for Power BI. Story framing is the job, measures in the semantic model can stay relevant, and the particular chart may not, which is where self-service earns its place. For the spreadsheet step, he would hand the workbook to Claude or Grok. Copilot in Excel is the one he keeps setting down.

• A notebook is for the question in front of you, after you have seen the rows. Das’s second workflow is the one she stress-tested for about six weeks: the Gemini add-on in Colab, iterating a line chart in conversation, with the coding optional once you know what the visual should show. Mike does not enjoy Copilot in notebooks. He has Claude, and especially Grok, open a Python notebook, write the code, and hand it back. He will name the library — Plotly, or something else — because that specification is still a lot of code, and he wants the picture more than he wants to memorize the options. The speed shows up in Excel, where he dictates the change and the model finds the control, and in the notebook, where the agent does shaping he used to write by hand. He still checks it. He builds cells, prints the first hundred rows, and trusts a chart only after a narrow table makes sense, then he widens the dimensions so one view carries more of the story. That chart is for him, today. A report for a wide audience is the next step up the pyramid Matthew Roche describes, from a personal view to a team, an organization, and a company: fewer reports, a larger audience, still built with AI, and only after the discovery.

• Build the report in Power BI, where the model already is. Das is explicit that Tableau Desktop has no built-in AI without another license, so her third workflow uses any chat tool as a patient Tableau expert: describe the fields and the story, ask for dummy data that matches the schema, ask which chart, and ask for the click path. Tommy’s move is to do that work in Power BI. An MCP server and the Fabric skills can format PBIR and assemble a full report. His exploration layout is one large page of tables. He can ask for six charts from a named table — bar, line, and column among them — placed on that page, or ask for another page when the story needs one, and speak the result into place. Mike’s sharper use of the same tools is the data the visual needs. A separate table, or display measures in the semantic model that exist for that page, is the piece he keeps seeing agents build well, and Power BI Desktop bridge with the right context is the way he would do it. He marks that as its own episode.

• Stay the architect, and let AI help you build. Das closes on two profiles the tools tend to produce. The conversational analyst gets fluent at directing a high-capability agent, hands off the build, and spends her leverage on translating the result and tying it to the business question. The architect analyst does the harder thinking first, pushes back when the output is off, and stays behind every design choice, with AI as the working partner. Mike puts the value on the architect. The tools he has been using, Copilot included, are not yet the conversational analyst he would trust with the answers, though a discovery pass — what this table means, what these dimensions do, how this relationship works — is a fair use of them. Once that pass is done, AI belongs in the construction. A chat box in front of a sales team, with an invitation to just ask, is a pattern he would leave on the shelf. Tommy’s close is the same line: use AI to help with the construction, and the results get better.

Looking Forward

Use AI where a Power BI report is actually assembled: requirements the agent can read, a sheet or a notebook you can check row by row, and a model that can add the page once the story is known. The analyst Das calls the architect is the one Tommy and Mike would staff for that work. The open thread they parked is Mike’s, and he already called it a future episode: using MCP and Power BI Desktop bridge to build the table or the display measures a visual needs, with the context in place before the agent starts.

Episode Transcript

0:01 Dance to the tunes of the day to laugh in the mix. Fabric and A. I understand your wish. Explicit standards. Unleash the rhythm now. The pumpkin is making the audience feel it. Explicit standards. Explicit standards. Hello everyone, and welcome back to the Straightforward Metrics podcast with Tommy and Mike. Tommy, good morning. Good morning, Mike. How are you? I’m you? I’m fine. Things are going well. I spent a long weekend in the United States

0:31 long weekend in the United States. Here . Here we have Memorial Day. So, we have a we have a three-day weekend. . labor day. We are coming back. Sorry. Labor Day weekend. . I know. I mix them up all the time. I mix them up all the time. And so, here we are back at work, resuming things. Let’s start again with the podcast. How was your weekend? Did you do anything fun? It fun? It was good. Ah, I was a little sick on Saturday, but it was one of those cases where I think the weather is changing now, but we’re feeling better now. Ah, and then yesterday was a busy day of travel. I had to buy a bookshelf for my wife from the Facebook marketplace to complete the

1:01 from the Facebook marketplace to complete the set. We have lots of books, birthday parties, and,, there’s no such thing as a long three-day weekend. Yes, . Yes, everything else is busy. We took the family on a family camping trip, so we got away a little from the home routine. There wasn’t much work, so we relaxed and let others organize the activities and events, and that was fun. We enjoyed it. Okay, after this, let’s talk about our main topic, and then move on to some

1:31 then move on to some news. Tommy, this is an article from the developers of Nightingale. I think we’ve discussed this topic several times. We will discuss it again, or maybe not, but this is another great conversation about how to use artificial intelligence to build to build data visualizations. How does this actually work? What concrete experiences can we build to help build this project? Well, I don’t know

2:01 project? Well, I don’t know how to pronounce their name, but there’s an article by someone on this blog. We will discuss it shortly. We will talk about how they approach things using visual elements and artificial intelligence to help shape what they build within their visual space. Okay, Tommy, what’s your news for us? Well, first of all, there are only two weeks left until the next Microsoft Fabric user group meeting in Chicago. And if you missed it, we’ll review a comprehensive solution for using

2:31 for using augmented reality tools to take advantage of Microsoft Fabric. We will talk about constructing the context. This will actually be based on some on some articles I’ve written about your tool, your agents, your MCP, your instructions, and how to link them together. How to write instructions correctly to build a pipeline or even a Power BI Desktop bridge. How is all of this done, from holding meetings, through obtaining resources from the client or project,

3:01 from the client or project, to employing them effectively so that your task is not limited to simply suggesting new ideas, but that you can provide the appropriate context? What is the best method we have seen succeed based on experience? What tools should be used? Are we really relying on a ” second mind” approach using MCP servers? You should register early. Make sure to use the Meetup page to register. Tommy will meet you in the city center. This is at Microsoft’s office in downtown Chicago. Yes, people have started registering

3:32 people have started registering and joining the user group. Anyway, it looks like it’s going to be great. It’s at the Microsoft Microsoft Technology Center in downtown Chicago. excellent. Good. Now, we have some of our recent favorite articles, Mike. We talked about this topic about a week or two ago, and there is a new series on the Microsoft Fabric blog about Fabric Data Warehouse Medal best practices, and building a medal structure using the using the data warehouse. This is the

4:03 data warehouse. This is the third part of the series. The first part was choosing the right style, and we have already talked about that. The second part, which we have not yet discussed, was about the layers of bronze, silver and gold. Part three deals with best practices for running the Medallia architecture in a data warehouse. It focuses particularly on success that depends less on less on ratings and more on what is called operational discipline, which maintains the predictability, , reproducibility, and reliability of workflow. Mike, we’ve

4:33 reliability of workflow. Mike, we’ve already shared our views on why to use a data warehouse or to use Lakehouse. Does the Medallia architecture work best with a data warehouse? I have a few points to make here, Mike, that I’ll mention. I want you to tell me what caught your attention. One of these points, and this is from Sydney, oh, Shakira, who talks about why Medallia paths usually break down with the data warehouse, is when the layers stop working as planned. The bronze layer begins to clean, the silver layer begins

5:03 silver layer begins to provide information panels, and the gold layer addresses the problems of the previous layers. The best practices they talk about are for the data warehouse to improve group-based operations, not row-based operations. Use bulk assimilation, copy to copy to pipelines, and avoid small files. Target specific file sizes, recommended to be between 100MB and 1GB. Make the data repeatable. It should be repeatable and testable, and use specific operations such as insertion, selection, and merging. Add

5:33 selection, and merging. Add quality criteria, such as mandatory fields, formatting, duplicates, and reasons for rejecting records. As for data for data intended for consumption, it is a business-oriented layer model that focuses on how users ask questions. It can be a star chart, a data warehouse, a wide table, or a pre-compiled summary. Finally, use the use the default settings. Do not attempt to change the data management system. Fabric Data Warehouse provides a delta-managed SQL warehouse in a single lake. Don’t use outdated practices like micromanagement of distribution, and focus on

6:03 of distribution, and focus on sound data planning. There’s a lot of stuff here too, Mike, but based on what I’ve heard, is there anything special for you that’s different from a from a regular Lake Data Warehouse? , this bothers me a little based on what’s happening in some of the pricing models related to data warehouses. I don’t know if you’ve heard of this, Tommy. Have you heard about the controversy within the data warehouse community? Well, it’s interesting that these two articles praise the advantages of a data warehouse, and how you can

6:33 data warehouse, and how you can use it just like you use a house on the banks of a lake, and do other things. They talk about best practices, the first of which is using payment rights, then using “silver” that can be reused reused again and again, then ” forming” (, best practices, the third of which is forming “gold” for consumption). These patterns are not exclusive to to data warehouses, but are simply a “medal” structure, yes, , integrated into the data warehouse. Again, what bothers me a little about

7:09 what bothers me a little about this article is their attempt to promote great data warehouse experiences, while I don’t see any fundamental difference between it and ” Lake House”. I don’t see any major change in either direction. Now, the new billing model for data warehouses appears to be more expensive. So, there are a few things. First, let me put this article here in the chat window. If you are using a data warehouse, this article discusses

7:40 data warehouse best practices in the Medallion architecture. Here it is. I’ll put it first. These are best practices. The other article I will share with you here is about the change in the data warehouse billing model in August. Previously, if it took 20 seconds to run the data warehouse, only the time required to create the query would be counted. Now, Now, time is approximated. If the query takes 20 seconds,

8:10 the query takes 20 seconds, the time is rounded up to a full minute and the cost is calculated. Therefore, CPU usage accumulates. . Well, this also applies to long-term tasks. If you have a long, time-consuming operational task, you are unlikely to be able to measure any significant impact. But if you’re making making custom or one-time queries, and I’m thinking about that too, Tommy, what happens if you have a direct lake-to- repository connection, or, oops, a direct-to- repository query? Well, I’m just executing

8:40 Well, I’m just executing the query the query very quickly, and this query is working extremely fast, and it’s just loading the initial report. Therefore, it executes a series of series of queries. They are all very fast. I finish. Then it remains pending and holds you accountable. Then someone clicks on something else on the page. I finish. One more minute, one more minute. Although these queries occur in less than a minute, or should occur in less than a specified billing period. So, I see this as if they are discouraging quick, short-term queries

9:10 quick, short-term queries on the repository for whatever reason. That’s what they do. Anyway, here’s a consumption model. This is the second article I will send out that deals with data warehouse consumption. Good. What bothers me about this is that the model for billions has changed. Now, running a running a data warehouse will cost you more money. So why don’t I use Spark then? Why don’t I use the data lake? I can store data in text files.

9:41 text files. This article discusses payment processing rights. That’s what I would have done at Spark anyway. With the advent of artificial intelligence, I no longer need these things. So, for me, this feature has to be there because it is essential. Some companies may want to use it, and that’s fine for project and job sharing, but if you don’t need something that works continuously, you don’t

10:03 works continuously, you don’t need a need a data warehouse. I don’t see a good use for it. No, I just don’t see the benefit of a data warehouse if you’re dealing with large amounts of big data or large tables, so I don’t use it in almost any of my projects. projects. I rely primarily on Lakehouse. There might be another observation, Tommy, but I’m not sure about it. I cannot determine it precisely right now. I haven’t conducted any tests

10:33 conducted any tests on it, so don’t rely on my word. Is the endpoint of SQL analytics a data warehouse or something else? else? No, it is not a data warehouse. It functions as a database, but it is not a data warehouse. Okay, are they built in the same way? So, what I’m trying to say is that the data warehouse changes its billing method, but but is the is the SQL analytics endpoint billed when a Lakehouse is created

11:03 and acquired, or if it ‘s enabled by default, in the same way as the data warehouse billing? correct? You will not bill for the time it takes to execute queries, but rather for the entire minute. Therefore, your costs will increase. I wonder: Why this change? Why do we need to change it? It seems to me like an attempt to increase profits. In any case, I understand the concepts presented in this article, but I’m not sure why sure why they focus so heavily on the data warehouse. Yes,

11:34 data warehouse. Yes, you can do that, but I will use my Bronze and Silver cards in the Data Lake. If it’s at most, most, I’ll use the gold drop within the data warehouse. Perhaps even this. this. Yes, I completely agree with this, Mike, because I don’t want to talk about a conspiracy theory here in terms of these articles appearing when costs rise. Let’s not play this game. However, Mike,, all of this is intentional, Tommy. All of this is

12:05 intentional, Tommy. All of this is intentional. It’s like raising prices here, then we need someone to explain why we want to use these things, let’s get people excited about them. Let’s make sure they know, because if you have some good news alongside some bad news, the good news probably outweighs the bad. But I look at this and say: I don’t know what’s going on here. here. With the advent of football, it’s as if you and I are on a radio show asking: Do why he got injured? I know the reason; he went up the stairs injured. That’s why he won’t play on Sunday. But I do

12:36 play on Sunday. But I do n’t want to speculate, even though these these speculations seem certain, don’t they? However, it seems to be a coincidence. I think the most important thing for for me is that I haven’t seen in these articles a better reason to use a data warehouse for the entire data path (bronze, silver, gold) compared to the current path. The reason, as you stated exactly: the data warehouse usage share. I have yet to see any

13:06 real reason or benefits to use only a data warehouse. Now, I would like to point out that many organizations prefer, and are somewhat obligated to use, a data warehouse architecture. Not . Not necessarily from a data point of view or from a or from a measurable point of view. Absolutely, that’s what they feel comfortable with, and that’s what they have. So, there’s one side to that. Perhaps this is the main reason I see it: if you are comfortable with it and willing to pay more for it because this is what your team knows, then a data warehouse is the best option., I don’t

13:36 best option., I don’t think they’re trying to convince anyone not to use a use a lake house and to use a data warehouse. exactly. It’s like, “We’ve just decided to use a use a data warehouse. We’ll be using these things from now on, and our team already knows SQL and T-SQL, so we’ll continue.” That’s how I see it; that’s the option available here. It’s not about evaluation. For example, I had a client years ago who wanted a data warehouse, but he didn’t have anything in SQL at the time; they were

14:06 SQL at the time; they were all SharePoint lists. They said, “We need a data warehouse.” So I said, ” No, no, no, you need things that aren’t in SharePoint, and then we can talk about the data warehouse.” There is clearly an element of appeal, but yes, I recommend you check out that article if you are thinking about creating a data warehouse. The series is useful in this regard. However, my position and Mike’s. Speaking of moving large amounts of data to Fabric, a

14:36 great article was recently published that covers Snowflake, secure connection to Snowflake, Fabric pipelines, and copying functions. We know that Microsoft Fabric now supports the secure transfer of private data to and from Snowflake using pipelines and copying functions. Again, if you are blocking access to the public network and require that every step remains within a private network, this is the right tool for you. It’s a really great mechanism of action. Okay, let me stop here. Well, I don’t want to talk about this subject. I’m still thinking

15:06 subject. I’m still thinking about the article about the warehouse. Okay, let’s begin. Good. I . Good. I might need to check some facts. I probably don’t understand anything here. I may be confused. correct. In best practices, there are best practices number 1, 2 and 3, which relate to bronze, silver and gold medals. In point four, it is stated that the Fabric data warehouse uses uses default settings. Do not attempt to change the

15:36 change the engine settings. Fabric Data Warehouse provides you with an SQL data warehouse above a delta lake in a single lake, which means getting getting transactional behavior. It indicates how to do this, and advises keeping the VAU arrangement and platform improvements unless unless you have a reason to change them, and in my opinion, a good reason to change them. Well, I do n’t know how to change the VAU order in an SQL data warehouse. Everything I’ve read in

16:07 I’ve read in Microsoft articles, when searching for delta tables optimized using VAU sorting, points to Spark. Do you have… so, if anyone is listening, does anyone use the data warehouses and have they changed them? Is it possible to change the sort order, such as VAU, ZAU, or VAU sort order, within the data warehouse? I don’t think so. So, I look at this and say to myself: This article appears to have been written by someone who uses Spark and is talking

16:38 someone who uses Spark and is talking about VORE, which is something I don’t think you can even decide. I don’t know. I don’t want to know. to know. Yes, I am Yes, I am also hesitant about this because I cannot find where to use VORE within the data warehouse, and I can’t find it, but I may be wrong. Again, I do not use this technology. That’s why I’m hesitant. Well, anyway, sorry. I finished. Go back to your new article, Tommy. Good. Let’s make sure there’s nothing else that might have upset you. Okay, we’re done. I finished.

17:09 done. I finished. Good. Good. Let’s move on to Snowflake. The situation is more interesting here. Again, Snowflake allows Fabric to copy data for those who have blocked access to the public network, and requires that all communication remain within a private network. You can do this, and it’s all explained using trusted access to the workspace via a private link at the workspace level, where the private link comes from the trusted Fabric network, the workspace as access to Azure storage, and Snowflake’s private link storage integration, which involves

17:40 which involves the use of a private endpoint. This simply allows you to solve the three-sided security problem: whether you have an Azure Blob Staging, a Fabric workspace, or your own Snowflake endpoint, you can do it now. Frankly, I see many organizations being very cautious and hesitant about many of many of their network problems, and this is true in many cases where public access is disabled or denied. This simply shows that you can connect directly to Snowflake,

18:10 directly to Snowflake, copy data to it, and push it from there, which is really cool. But cool. But again, there are a few things you need to do: build the network foundation, the virtual network (V-Net), the private link service, create a staging storage account, configure Snowflake, close the public password, and create a copy activity or copy task. So, these are some things that require some setup, but they are entirely possible. Now, here are some restrictions, Mike, before we move on to the next step. The range range and area of ​​your region must be supported in your

18:40 and area of ​​your region must be supported in your own connectivity system. What does that mean? You might be wondering about the scope and size of your area. Well, some regions may not support everything yet. So, be careful about that. The Azure Azure Blob hierarchical namespace is not required in the testing environment. Private binding requires version B Business Critical or higher, and the private endpoint must be approved in the storage account. Therefore, you do n’t necessarily have to activate this option simply. However, if your tasks are critical on Snowflake, this is definitely something worth checking out

19:11 this is definitely something worth checking out. . I think it’s an excellent option as well. Again, this is a very specialized area in Snowflake. I feel that Snowflake has become more user-friendly within Power BI. Some other platforms, such as Data Bricks, were not easy to use, and there seems to be ongoing friction between the two platforms, and they cannot cannot fully integrate and find an effective way to exchange data between them. But Snowflake seems to be on the right track. Okay,

19:41 right track. Okay, once again, I think you’re splitting your team into two sections, right? There is a data engineering team, which is responsible for the systems used, how used, how data is loaded into the data warehouse, and how this data is processed at the warehouse level. This is a specialized team designed to do this work. Then there is the business intelligence team, which is responsible for creating semantic models, building a security architecture

20:05 building a security architecture based on based on workspaces and distribution models, and using elements in the workspace to help manage and access data. This is a completely new aspect of the work, independent of the data engineering team. So, it seems to me that this is a pattern of two teams evolving. The same applies to data units. The data units follow the same pattern. The data architecture and data modules are implemented, and then the business intelligence control surface management layer is moved to

20:35 to Power BI, because users— regardless of the company— feel that all the companies they have spoken to prefer easy access to, trading, and exporting of data within the system. Moreover, Power BI is the ideal choice for these organizations, either because they already own E5 and everyone gets Power BI by default, or because they have already chosen to manage

21:06 already chosen to manage this domain. When you move from Power BI to Data Modules and Snowflake, the user interface is completely different, as is the creation of visual elements and the level of control, and you now have to constantly update both systems with regard to data access and management. Both are good in their own right, and you could simply do without Power BI and rely on traditional databases for everything, but I think there is a closer correlation

21:36 there is a closer correlation between business and office integration, and that is what is driving organizations toward using Power BI. What is What is happening now is very interesting. I’m Mike, and I’m talking to some customers right now, and they’re saying, ” Well, I think we’re forced to use Fabric because Power BI is now part of it, and Power BI works better with it.” Technically speaking, Power BI did not lose any of its limitations when Fabric was released. If you still want to rely solely on Power BI and connect to

22:06 and connect to SQL databases or other endpoints, all of these features will still be available to you. Nothing has changed. I think the message that Microsoft has presented, as , as you mentioned, is very clear, which is to integrate everything into the same path between Power BI and data architecture on the same platform, to the point that Power BI seems to be part of the solution, even if not at its full potential. But technically speaking, you can do everything you did five years ago and

22:36 five years ago and still be just as successful, right? , right? Because he didn’t do that, right? Power BI didn’t lose anything. Yes, I agree. And I now feel that a big part of the Fabric story for me is that I have become more committed to my opinion about sharing my data with other external tools. Whatever this this external tool may be, the obstacle that prevents me from bringing programs, solutions, and applications into my business that I can develop, maintain, and

23:06 develop, maintain, and build myself the way I want is I want is very strong. I became less enthusiastic about trying to use Salesforce, or experimenting with these other applications. I believe the key now lies in using artificial intelligence to build customized software applications, tailored specifically to your organization, that fit the way your company operates., ultimately, the intellectual property that we own as a company is our own, and it is represented in the

23:36 our own, and it is represented in the way we operate, the knowledge that our employees have about how to run our business, and the processes that we develop to satisfy our customers. This is our intellectual property. Therefore, it is no longer about the software itself. If I were to buy massive software programs, would I use them completely? Probably not. I will use part of it. So why don’t I build the part part of the program that I actually want? Frankly, the obstacle to building and maintaining this program is greatly reduced

24:07 program is greatly reduced. It’s not nonexistent, there’s still some difficulty. It’s true that you’re putting in extra effort, but the days of hiring a team of 15 people to build a custom program for your internal organization are over. It has become much easier, closer to individual work, and these individuals have less technical experience. Honestly, Tommy, I’ve had a lot of conversations with people on all of our social media platforms. Many of the short videos we post on YouTube, and those that are currently being posted, are spreading across various

24:37 across various social media platforms, and I see people on TikTok, Twitter, and Blue Sky commenting on them. There are many comments about how difficult it is to build and maintain applications. There is a great deal of hesitation among developers about: ” If you build an application, are you really ready to publish it? And are you really ready to maintain it?” And I tell you: Yes, we have reached this stage. But I want to

25:08 stage. But I want to strongly warn you as a developer of this application, i. e., the person who programs or talks to the artificial intelligence system. You cannot do without the engineer’s role in the system. You will get better apps, better testing, and better security, to achieve better results, by having an expert app builder working with the agent to build the app, so what to do and what not to do, because agents often get distracted. You need to guide them. Firstly, I agree with you.

25:38 Firstly, I agree with you. Secondly, I am trying to understand the relationship between Snowflake and app building here. Okay, again, that’s another area, isn’t it? This relates to delegating roles and where to place data, right? Ultimately, I must transfer Snowflake’s information to Fabric no matter what. . Yes. Yes. Are you building your applications on Fabric or on Snowflake?

26:09 My current opinion is that the idea of building applications on Power BI and Fabric is more convincing to me than starting to build applications on Snowflake at this stage. I don’t see the same appeal that I find in Fabric. Okay, let’s take as an example the Power BI program in 2015 and 2017, where you had a central site, you logged in using your usual Microsoft account, and you went to the central site to find all your data. It’s one place, and now you can expand this store. It’s similar to using Snowflake and data storage units.

26:40 data storage units. However, you can get this consolidated data in the same place using using enterprise-level tools, but you do n’t have to be an enterprise-level company to use it, and that, in my opinion, is the most important thing. When you think of Snowflake, you think of data storage units. Most organizations think in terms of organizations, don’t they? This is not what medium-sized companies do. Fabric changes this reality. But Mike, you’ve made some points. I think this is a good introduction to our main article today. Here it is. Let’s . Let’s move on to the main topic. Here we are talking about another great article, but I have some comments

27:10 I have some comments on it. This article is from Nightingale. dvas. com [We] have done it; She is a writer specializing in the field of visual data representation. I believe she is a university professor, working in several different universities and programs that deal with the art and science of visually representing data. She recently published an article discussing how she uses uses artificial intelligence in this field. This article is based on her participation in several conferences, where she

27:40 conferences, where she confidently uses specialized terminology, which I like about her, such as her question: “What artificial intelligence tools do you use?” you use?” She asks the participants: “We use AMCPS, we use automated or decentralized analytics, and we use interactive analytics.” But the problem is that most people cannot explain how these tools affect their actual work, which is something I have noticed. I found that the environment for artificial intelligence tools,

28:10 environment for artificial intelligence tools, especially in the field of business intelligence, specifically data visualization, is fragmented, and that organizations are cautious about granting their employees access to these systems or even allowing them to be provided with sensitive data. There are many experimental tools that most practitioners will not use. Therefore, she presents some different approaches you should start with from a realistic point of view, and talks about three working methods she has done. But Mike, let’s take a step back and think about the words that you and I use,

28:40 use, because I think this is a very important topic, let’s try to clarify it from the beginning. It talks about all these these technical terms such as ” interactive conversation analytics” and “MCPS”, and there is no popular term without substance. I will listen to you here and ask you: Why do you think people use these terms, terms, but cannot relate them to their work? I feel that we are not facing this problem. problem. Well, I think that’s also the difference between researching and listening to

29:10 researching and listening to others do others do it, and it, and actually sitting down in front of a computer, turning it on and starting to work, right? So, there is some time to explore. It seems we all want to use the use the popular term. Everyone wants to join that join that elite group that uses that uses agent experiences to develop products. So, going back to this point, the three terms you were hearing repeatedly were: MCP, , decentralized analytics, and interactive analytics. These are all

29:43 . These are all common terms in this field. But when you ask people about the about the workflow you recently did to create a report using MCP, what does it look like? I will start by opening VS Code, saving the PBIR file, then I will talk to my agent and ask him to create additional pages or other visual elements on a given page. That’s what I use in MCP, isn’t it? Building things

30:09 it? Building things using MCP. I also use pen. dev. Therefore, I will buy or purchase a new free open-source program, where I can connect my agent to a design system and work with him to develop that system. What does this look like? How can I develop backgrounds, images, shapes, and other elements within my experience here? That’s another question. The other part is the invisible analysis. This is a very common term for me. When I hear ”

30:40 When I hear ” invisible browser” or “browser without a graphical interface,” customer service agents sometimes fetch a website or run a web application in which no graphics are displayed, but which runs in the background. I don’t know exactly what invisible analyses are. Is it just a conversational experiment with your data? Is it not a report, but rather an existing semantic model that uses an agent to communicate with? I don’t really know what invisible analyses look like. Well, interactive analytics is where is where all

31:10 all logical and computational operations are performed. The back-end system does the hard work, but you stay in an easy- to-use and clear interface, and you don’t have to worry about skills or anything else. So, for me, it’s interactive analytics. Okay. I do . I do n’t fully understand the decentralized analysis aspect of this topic, and what people mean when they say the term sounds made up, and that they are just repeating it. As for As for interactive analytics, they are understandable to me, aren’t they? I can understand this particular term. It’s simply a conversation with an agent about data that helps you

31:41 about data that helps you develop things. In fact, these three terms— content management platform, decentralized analytics, and interactive analytics—do not describe the process of creating creating graphs. I think if we keep it simple, let’s remove the tools and artificial intelligence from the solution. Tommy, explain to me how a report is created? Again, I’m not talking about data discovery, nor about communicating with the team. I am not like that.

32:12 . I am not like that. We have already spoken with the company. We have already completed this part. I don’t assume that, I say that it has been done. I don’t . I don’t assume so. No, I suppose so. I say that I don’t want to talk about this part of the process. We will assume that this part of the process has been done, because if I don’t assume that, you will say, ” Well, we should go back and do it.” I don’t want to talk about this part. Let’s assume that this stage has been completed, and we now have the requirements, and we know what the company is generally looking for. Let’s assume

32:43 that this stage has been completed, and I believe we are now entering this step step or this part. When we talk about decentralized analytics or interactive analytics, I think I already have a concept of what I want to extract from analytics, and that is the part I want to clarify here. Is my question clear? decent. I have to do this just to clarify things, because I can’t easily overlook these matters. However, I would like to add that part of the workflow we are talking about

33:13 workflow we are talking about is that if the right requirements are not presented in a way that is effectively used in the workflow, then you miss the miss the main objective. I will make an argument, and I will be completely right. All you’re saying is that you need requirements. You are confirming to me that how these requirements are constructed is a very important part. It’s not just about having requirements. This is like owning a car; I need an engine. I need the need the right wheels according to the needs of my farm. Why do you care about how it is built? built? Good. When I get the requirements now,

33:43 requirements now, Mike, the way I put those put those requirements together is to make sure they’re going to work well with the agents and tools I tools I have. This is part of the way I speak in meetings because I know she will use certain skills. I would ask, “I’m building my agenda based on what ‘s already in the chart from the proxy tool,” to say, “Look at the semantic model, what are the bottlenecks in those conversations I’m already having with the analysts?” I would say, ” Okay, we’ll do some research to see if this is actually possible

34:13 this is actually possible, from a measurement point of view; what visual elements do you want? And how do you want to display them?” I want this text to be part of the context and guidance that I will provide for any interactive tool for the visual aspect. Mike, I ask him questions, not so that I can hear them, but so that my interactive tools can understand them correctly. Therefore, I will ask some questions because these questions are specifically designed for interactive tools.

34:43 interactive tools. It is designed to be understood correctly by the skills, rather than simply telling me that you want a want a bar chart. I believe my interactive tools understand this, but I want to clarify some of the questions or things they said so that I can make it as clear as possible when I get to the conversational side using my tools. I feel like you’re touching on something different than what I expected. expected. That makes sense. Yes, I think we’re talking about similar things, but you’re saying something

35:13 saying something slightly different. Perhaps I would describe what you are describing now as closer to data discovery, or,, the idea of having a chart, wouldn’t you say? When you start a conversation with someone about “Let’s build some analysis into what things are,” the conversation starts very small; It’s too tight. All you have to do is allow the conversation to expand; To become very wide. So, I think that in the initial stage, we allow the scope to be very broad

35:43 initial stage, we allow the scope to be very broad; ; we want to have a wide scope. The reason we have such a wide range is that there is no such thing as a bad idea; Every idea has value. You need to tell me what’s bothering you, so your initial range needs to be very broad. I believe what you are describing now is the second stage of the requirements gathering process, which is: “Okay, now that I have been able to articulate these very general things that are not useful to the model or may not answer, we

36:13 answer, we now need to narrow down our search to: Okay, what does the model contain?” For example, you are asking for contact center data, while , while the form does not contain any data of this type. So, although this may come to your mind, technically the user cannot extract this data unless we identify its source and how to integrate it with our existing data. This is something I don’t know exactly, which is why I’ll keep the MCP certification in mind. I know there are common terms at conferences, because people

36:44 conferences, because people assume they’ll start using these tools once they create a visual presentation. I think they missed the target. You should start early. But let’s get back to the main point: I’m expanding the scope of the work while you’re reducing it. All of this falls under what I call the discovery phase, i. e., gathering requirements. So, all you describe in this process is expanding the scope of work, identifying what already exists, aligning the model, and getting to what users actually want. I would say that, ultimately, we have reached the

37:15 we have reached the end of this phase. We have finished defining the scope. Good. I think there is another step that enters the scope, which is when we start to delve into what I believe this article portrays and talks about, such as decentralized analytics and interactive analytics. Now that I know what I’m looking for, I need to make sure the model supports it. Can I extract it from the data? Is there any information I can obtain? Here,

37:45 obtain? Here, I believe that decentralized or interactive analytics stand out as important. Because when I look at look at artificial intelligence in the field of data representation, I think, Tommy, that we will rely heavily on artificial intelligence to create to create graphs, without neglecting the role of the mind in the process. We do n’t simply say “Give me a sales data report”. You can’t tell tell artificial intelligence that, can you? ? It’s a very loose matter. It will give you interesting answers. These answers may be

38:16 real numbers, but they may not affect anything. I’ve also read some other articles urging the provision of very specific information to artificial intelligence, haven’t I? So, instead of asking why sales are down, many will say general things like this. The most important thing is to identify the two or three main factors in the second quarter that led to a decrease or increase in sales based on specific information. This is a much better answer or question you could ask

38:46 you could ask artificial intelligence to help it search the data or provide it with relevant information. I think the point I’m trying to make with this question, and perhaps the most important part of this article, is that when we start using massive language models in our tools, the biggest part of it is about exploring the data, and then saying, “Okay, agent, help me

39:16 agent, help me determine what visual representation best captures this information.” And .” And I think that the tools in this regard, Mike, I agree with you, if I understand what you mean correctly, the value of these tools does not lie in creating the bar chart for me, since I do not have to do that do that manually, but rather in reducing the time between the idea and the final result. Yes, the tool is still creating the visual representation for me, but me, but Mike, as I mentioned, the value lies in taking all the

39:46 the value lies in taking all the inputs, whether from the exploration process or from my own analysis and available metrics, and helping to come to a logical conclusion based on that. I don’t care if care if artificial intelligence can create a bar chart. It’s true that it’s nice to be able to do so, but that’s not what makes it impractical for me. The value, as you mentioned, Mike, lies in making the necessary adjustments, understanding the key performance indicators,

40:11 key performance indicators, and understanding the key pressure points for the client or for this particular report, in understanding not just the results of a single visual presentation, but the interconnected visual presentations on the report page. Looking at it from a broader perspective, I don’t think we should view view this as a this as a data representation of a single visual display, especially in our field of work. I think Nightingale, in her three examples, is basically talking about a single visual exit, but that doesn’t suit us. This approach has never worked for us

40:42 This approach has never worked for us in our work. Okay, let’s address this. So, I think we are going in a slightly different direction than the article is going. Let me give you more you more clarifications about what was mentioned in the article, so that listeners can better understand our point of view. In view. In this article, the author describes three separate workflow processes, and I think she focuses heavily on Google and Gemini, which is fine because that’s the platform of choice, but I think there’s a skill set that’s applicable to Power PI, Fabric, and Excel.

41:14 Let’s start with the first process, which is using Google Sheets and Gemini. I have exported data, which is in a Google spreadsheet, and I need to organize and process it in some way. I speak with an agent who understands how a table works, and then he says, “Okay, create a table, a table, a pivot table, and a bar chart.” That’s what he does, and that’s why I feel like that’s what I’m focusing on in this part. This is the part of the

41:44 part. This is the part of the data discovery that I interact with directly. I have the data, I explore it, and I look closely at the information to see how I can create a visual representation of it. I am impressed with this experience. Honestly, I think data discovery using an agent is extremely useful. And once again, let’s ignore the agents for a moment. How can I discover a pattern, and what insights can I glean from the pattern or existing data? Well, one of the things we do

42:14 things we do a lot in the Microsoft community is the Power BI Data Visualization World Championship. . Yes, and Yes, and world champions should have a carefully crafted story with the data set data set provided to them. So, in that designed story, you need to identify the logical data, and where does the story lie? Therefore, I believe there is significant support for artificial intelligence in story discovery. For me, this involves analyzing many

42:44 many different tables, comparing various calculations with different types of measures or different combinations of them, and studying data trends over time. Have you noticed any interesting or useful patterns or information about a particular trend? When I look at these things, I find that these are the elements I focus on in datasets, or when working with clients, we try to draw conclusions from them. How many times have we had we had false assumptions about data, Tommy, because the data actually showed something different?

43:14 different? This shows a weakness in the analysis. I think we go into these situations with preconceived assumptions about what what the data should do, and what the story will do. And sometimes, you have to dive into the data and let it tell the story. Do you remember the episodes where we discussed the difference between storytelling, which is which is practically impossible in Power BI, and story framing? Because

43:45 story framing? Because data in our world is constantly changing. We rarely get stable data to create custom visualizations, do we? When we think about data visualization, we think about something that will remain over time. We will not, for example, present “How did “How did Campaign (A) perform?” Then we get rid of this perception completely. We will present this and that in a different way; It’s a different way of creating

44:15 different way of creating perceptions compared to creating an organization over a period of time, isn’t it? We will be satisfied with the sales report, because in the first example, the custom visualization, we will highlight certain elements, specific dates, and specific time frames to display the events. But again, this is not useful with constantly changing data. And you’re talking a little, yes, you’re talking a little about… I do n’t know, Tommy. I don’t like that you’re adding a adding a new complication to

44:46 new complication to things. Visions occur at a specific moment, unlike visions that are constantly repeated. I think many reports are real- time reports, aren’t they? This is what the data looks like now. We are updating the data. the data. These diagrams were designed in this way to frame this story at this moment. moment. The data changes within two weeks or two months, and those same graphs become irrelevant. correct? However, the semantic model structure is consistent. The metrics and calculations in the semantic model may remain

45:17 in the semantic model may remain relevant over time, but the but the specific graph you have chosen may not. Therefore, this is where we talk about self- self- service reporting, and this is where many people are putting in the effort to effort to prepare these self-service reports. In short, this example, this article talks about Google Sheets and talking to an agent to help you create spreadsheets and, ultimately, charts, such as columns and lines, within Google Sheets. The equivalent of that is using Copilot in Excel. I can’t stand him.

45:48 in Excel. I can’t stand him. Or use Claude, Cla, Anthorpics version, Grock or similar. Honestly, I prefer using other tools. I don’t like using Copilot inside Excel. It does not appear to perform the function of artificial intelligence as acceptably as current tools. Copilot in Excel is fairly good, but when compared to other tools on the market, Claude is much better. Grock is much better too.

46:18 Grock is much better too. It offers far more functions, and I don’t need to explain much of what I want to accomplish for it. It understands automatically and works seamlessly. Therefore, I think these models are much better within an Excel environment, or you can simply give an Excel sheet to the AI ​​and say to it: ” This is an Excel sheet, work on it with me.” The second model I want to move to, which requires more time, is stress testing using a

46:48 using a Python notebook and Gemini together. Gemini has a Collab add-on that enables you to import Excel or notebook data and work collaboratively with an AI agent, with the notebook being exchanged back back and forth. However, I still argue about Microsoft’s experience in integrating C-pilot with notebooks; I am not entirely satisfied with it. I don’t like her. Actually, I do n’t use n’t use my notebooks with C-pilot at the moment. I prefer to

47:18 the moment. I prefer to use my use my own notebooks with other agents. So, a large part of what I do is work on Python notebooks, but using my own proxies. Well, Claude, and especially Groc, are my current favorites. For models and agents, when I work with notebooks, I ask the agent to open a Python notebook, write the required code, and then I review the notebook or redeploy it to see its contents. It’s not a completely seamless experience, but when engineering

47:48 , but when engineering spreadsheet data, designing designing graphs, or creating projects, I find this method very effective in helping agents with graph design. . What do you think, Mike? Given what you have mentioned so far, do you think you think artificial intelligence is better at designing designing graphs or does it only speed up the design process? Quality is the standard, isn’t it? Whether using Python and Anthropic or

48:19 Python and Anthropic or C-Pilot in your workflow, do you find them more convenient or do they offer better quality than if you designed them yourself? I think that when talking about Python, all those specifications should be written from a visual point of view. But no, I don’t think so. I think you can let can let artificial intelligence give you some of these specifications. That’s what I think if you’re doing it manually, then definitely. Yes. Yes. But it takes a lot of a lot of code to determine what you want to get, does n’t it? Therefore, you can specify the type of library you

48:49 specify the type of library you want. Do you want a library like Plotly or something else? Would you like it to create graphs for you you inside the notebook? You are right. It requires a lot of detail on how to do it, and again, I wouldn’t know all of that. I don’t care to know all that. I just want to get an answer. I wanted to get a graph that I could work with. Do you believe that at this stage of the workflow you find value in the speed of execution rather than its quality? In this particular scenario, I think it’s faster in both

49:20 I think it’s faster in both cases. correct? In Excel, I don’t need to click buttons or search the user interface. I simply dictate what I want, and it knows the knows the available options, adjusts them, and produces the results. The same applies to the Python notebook. If it needs needs data engineering or shaping, that’s code I would have written beforehand, but I’m letting AI AI handle some of that. However, I still need to review and verify it. I will create multiple cells in my notebook

49:51 to collect or generate data, and I will make sure that the program displays the first 100 rows. I wonder: what does this program do? I may be a little naive. I don’t know if anyone else does what I do, Tom. Perhaps you would do the same thing, Tommy. I don’t trust charts at charts at first. I want to see spreadsheets, and if I can see I can see spreadsheets with few dimensions,

50:14 few dimensions, then I can start to visualize them. I can start from this point. Now, let’s see how this looks in a graph. Let’s start by expanding its scope, and let’s add more more dimensional details so that I can see more information on a single page. I believe the main purpose of a graph is to present a lot of information to the user user in a way that helps them understand what is happening within the data. This is . This is also a key part of the process you are talking about: the data

50:44 talking about: the data discovery or exploration phase. Again, . Again, if you want to clarify the anomalies, I do n’t create graphs in a notebook now, and in most cases, I do so to share them with an organization. True, True, often, when using notebooks or charts charts, Mike, , Mike, correct me if I’m wrong, the goal goal is often to create charts for your

51:14 own purposes, for something you need to see at that time, as opposed to creating a report that I’m going to distribute widely. correct. I want to conduct my research. Yes. I think these are two different topics. correct. Because I believe that what you are describing is the presence of artificial intelligence in both cases, right? Yes, I don’t disagree with you on that. I believe I believe there is artificial intelligence to perform exploration work, create tables, and many graphs. I graphs. I need to create many things so that

51:44 many things so that I can choose what I want. Well, this makes sense, and this doesn’t. Let’s continue, let’s move forward or not. On the other hand, I think I now know what the story is designed for. I think I know what the user needs to see each time. Now, we move up the pyramid that Matthew Roach talks about, such as , such as personal reports, to team, organization, and company reports. Therefore, the number of reports should decrease, shouldn’t it? But the

52:15 But the audience for these reports needs to increase. Therefore, we are now trying to build existing tools, solutions, and reports that I can distribute widely. And I still believe that I use use artificial intelligence to help me with that. But this is just the first stage of discovering and understanding things. I am changing the way reports are built based on what AI can help me do and how easy and quick it is to build things using using AI. AI. However, I would like to move on to point three here. The third step is to describe the Tableau

52:45 is to describe the Tableau Desktop program and any artificial intelligence tools. I don’t . I don’t even want to get into this subject, but go ahead. We will . We will not use Tableau. I don’t like it. I think it is very complex, requires a lot of effort to learn, and is also very expensive compared to Power BI. It does not support artificial intelligence integration. . So, you can use it, but it doesn’t support it. This article states that Tableau Desktop does not have built-in AI capabilities except with an additional license. Again, if you want to do anything in Tableau, you will need another license. You want data engineering

53:16 . You want data engineering, and you want another license. You want to use the AI ​​features in Tableau, and you want another license. Well, most practitioners do not have these capabilities, and the author author explicitly pointed this out in the article. However, you can ask him to create a mock chart, or ask him what he can create, or how it will look in Tableau. Therefore, I advise you to stop using Tableau Tableau and use Power BI because it supports AI integration. You can . You can use an MCP server. There are skills available for Fabric.

53:46 skills available for Fabric. You can use it to format PBIR to create graphs in graphs in full reports. So, as explained in this article, you can do without Tableau and use any AI tool, and now you can create reports within Power BI. I think this is the next level, where I say, “This is amazing.” One of the methods methods I use is, when exploring data, I use a large page containing many

54:16 containing many tables that I work with until I understand the data and how it works. Now, I can accomplish this much faster, and I can ask the ask the AI ​​to create six different charts from this table (specify the table name), for example: “Create a bar chart, a line chart, and a column chart, create them all, and put them on the page.” It can produce more data at once, and I can display a variety of selection items in one place. Or, if I am exploring data through storytelling, I may create additional pages

54:47 create additional pages in the report because it is much easier to be asked to create a new page and add these elements to it. Simply put, I speak and the results appear. Anyway, I wanted to say that I admire these three steps. What do you think of workflow number three? As I mentioned, I would completely rule out trying to use Chomp. If you have Power BI Desktop Bridge and Fabric skills, I advise you to look into them and research them first. There is one quick point I would like to make

55:17 because I know we are taking up your time. There is something you talk about in the blog regarding profiles, but before that, you mentioned this three times in three different ways that I think are very important here. One of the best benefits I’ve seen from proxy tools for data visualization is not so much the construction of visual elements as the construction of the data necessary for them. What is that, based on the scenario, this helps me to build the separate table, at least in the semantic model, right? The metrics are intended only

55:48 The metrics are intended only for displaying reports, and are not semantic metrics for displaying the base figure, but rather are based on the required context. It involves building separate report metrics or tables, which I find extremely useful. In some cases, I may need a custom way to display data on a data on a particular page, and I find that MCP and Power BI Desktop Bridge represent a great use case with the

56:18 great use case with the right context to build that. I wanted to write this down. I think that’s a future topic, but let’s end with Nightingale’s talk about two different types of people you think will emerge, and I want to know your opinion on that. We will start from here. The first type is what you call an interactive analyst, who is someone who works with high-powered artificial intelligence tools. This person has become adept at directing directing artificial intelligence, knows what to ask and lets the

56:48 to ask and lets the tool do its work. The construction process is delegated. The analyst’s role is in the early stages, where he translates ideas, clarifies the value of the concept, and links the outputs to the question of the work. Artificial intelligence builds, while the analyst interprets and communicates. The second type is the structural analyzer. This is someone who uses uses artificial intelligence as a business partner. The analyst does deep thinking before artificial intelligence intervenes in anything. It defines the requirements and presents the resulting vision—something AI helped AI helped to build— to build— but the architect is the one

57:18 but the architect is the one behind every decision. So, there are two types of profiles that Nightingale is talking about in terms of what the future holds. I just want to know your opinion on this matter. I don’t think there’s a conversation analyzer. When I look at this type, the structure analyst, as you describe it, is someone who uses uses artificial intelligence as a business partner. I believe this is the area with the

57:48 area with the greatest added value. I think a conversation analyzer is one form of that. Hmm. So, how do I use the tools? Where can I find the value? I do n’t think n’t think artificial intelligence is good enough, or that the AI ​​systems I’ve used recently, or those that work in Co-Pilot, are good enough to be a conversational analyst with you. It is good, but it is not. I agree with you. . I believe the best place to apply to apply artificial intelligence, where I find the

58:18 artificial intelligence, where I find the greatest value, is at the level of the structure analyst I described. A person who uses uses artificial intelligence as a business partner. You’ve already thought about anything related to artificial intelligence, i. e., the specific requirements, which you mentioned earlier, Tom, we need to know what goes into this, right? There may be a data discovery component, as I’ve heard, a conversation analyst is someone who discovers discovers data using artificial intelligence, what does this table mean? What do these dimensions do? How does this

58:49 ? How does this relationship work? It raises general questions about the semantic model for collecting information, right? This is the work of discovery. Once you have that, you build using artificial intelligence. This is where the value comes from. Therefore, I absolutely do not support the idea of ​​putting up a chat bar in front of the Seale team and saying, ” Go ahead, ask your questions.” I don’t think this is good, and I don’t think it’s 100% perfect. Okay, let’s begin. This is . This is how I think I’m framing

59:19 how I think I’m framing this this conversation. So, I think if you want to sum it up properly, use use artificial intelligence to help with the construction process, not to provide you with the answers. I think you will get better results. Okay, after all that, Tommy, where can I find the podcast? ? You can find us on Apple, Spotify and any other platform. Get the podcast, and be sure to subscribe and rate it. This helps us a lot. Do you have a question or topic you’d like us to discuss in a future episode? Go to powerbi. tips/mpodcast. Leave your name and a special question. Finally, join us live every Tuesday and Thursday at

59:49 Tuesday and Thursday at AM Central Time on all of Power Band’s social media channels. We’re done. Thank you all very much, and see you next time. Rhythms that illuminate the day. I and Fabrice are receiving your explicit rhythms. Let’s turn on the beat now. Its kings feel for the public. . Explicit rhythms.

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.