New Linguistic Schema – Ep. 256

Power BI’s Q&A feature has been around for years, but it’s always had the same problem: the demo looks incredible… until real users ask real questions using their business vocabulary.

In Ep. 256, Mike, Tommy, and Seth dig into Microsoft’s latest updates to the linguistic schema tooling—new ways to connect concepts (beyond simple synonyms), plus hints at organization-wide sharing. The big question: is this finally the groundwork that makes natural-language BI useful at scale, or just more setup work that nobody has time to maintain?

News & Announcements

• Power BI September 2023 feature summary — What shipped in September 2023, with notes you can use to decide what’s worth testing (and what changes user experience immediately).
• Suggest a podcast topic — Submit a question, scenario, or hot take for a future episode.
• PowerBI.tips Podcast — Browse the archive and subscribe on your favorite platform.
• Power BI Theme Generator (Tips+) — Build consistent report styling with generated theme JSON instead of manual formatting-by-clicking.

Main Discussion

Power BI Q&A is chasing a simple promise: type a question in plain English and get the right visual back.

The linguistic schema is the layer that makes that possible—teaching Power BI how your organization talks about the model (synonyms), plus how concepts relate to each other (so questions don’t require perfectly-named fields).

Here are the key takeaways from the conversation:

• The semantic model still does the heavy lifting: Q&A doesn’t fix messy definitions, weak relationships, or inconsistent naming—it amplifies those problems.
• Richer relationships matter more than more synonyms: binding concepts like “stores sell products” helps interpret questions users naturally ask (verbs/adjectives), not just keyword matches.
• The most interesting shift is shareability: synonyms that can be shared start to look like a real organization glossary—something that can outlive one dataset.
• Governance is the prerequisite: without certified datasets and agreed definitions (what is a customer? which sales?), Q&A answers will be inconsistent or misleading.
• Users need guardrails and options: modern LLM experiences set expectations—Q&A should offer alternate interpretations (sum vs. count vs. average) when intent is unclear.
• Q&A can reveal missing report content: the questions people ask are a requirements signal; use them to add measures/pages rather than treating Q&A as the final UI.
• If you invest, invest where it pays off: start with high-usage certified datasets and a narrow domain with low ambiguity, then expand once results are trustworthy.

Looking Forward

If Microsoft pairs these schema improvements with AI-assisted suggestions and practical governance patterns, Q&A could become a useful front door for curated, certified datasets—without turning BI into a guessing game.

Episode Transcript

0:30 hello hello and welcome back to the explicit measures podcast with Tommy Seth and Mike good morning everyone good morning good morning oh that’s way of saying it yeah and on a high note what do start on a high out I gota I got sometimes I got to build up to the excitement Mike I guess yeah well are you on your first cup of coffee during this call or is there any pre pre oh yeah there’s a first this is first hang

1:00 yeah there’s a first this is first hang on hang this is my first sip whoa whoa no wonder wonder the excitement is just exuding from this podcast oh my goodness guys it is so good to be here if I told you this is my fourth cup of espresso doubt it okay okay is it really your fourth cup yes it is okay do you do you count them as each espresso shot is it different cup or just four shots of espresso or is it a double shot you’re double shot

1:30 double shot you’re really tripping me up here no it’s a it’s when I say a shot of espresso I’m not saying a whole cup of coffee I by that point I’d be doing this upside down how much how much are you through the espressos have single shots or double shots do you have like a single shots it’s four single shots okay four singles still a lot of coffee so do you have like an espresso I do and that’s probably why I can I’m drinking it’s dangerous just like right pop it in run the machine pop it in run the machine literally right you guys can correct me if I’m wrong but it’s my

2:01 correct me if I’m wrong but it’s my understanding that like a cup of lighter roast coffee has much more caffeine than espresso oh I’d have to ask chat to know if they know that I don’t know the answer of that one off the top of my head I think I just found that out too maybe that’s why I like all the light blond roast coffees because there’s more coffee in there maybe that’s why I’m doing doing zip sense oh yeah I’ve had like four cups espresso and it’s like well if

2:31 cups espresso and it’s like well if there’s anything that SS more counterintuitive when theye they call it bold roast you’re like yeah that’s what I watch something bold it’s like that’s actually more close to caffeine free well apparently Google thinks it’s it’s a thing because I I typed it in halfway into Google and it’s like does blonde Ro does lighter roast have more caffeine and it literally filled out my sentence for me so apparently people are Googling this already as well I’ll see if I can do a quick research thing here while we’re we’re diving into immensely important topic for all of our

3:02 immensely important topic for all of our Cofe drinkers out there Tommy want to give us a little bit of a run up on our topic actually before I do that I I forgot to mention any announcements that we should we should discuss I do have one that I should at Le Point on I don’t know if you guys had any other topics you wanted to touch on no just the only most important one I the only elephant in the air okay go ahead in the room you guys take it away

3:36 Seth earlier this week the power bi. tips team oh my gosh no the exciting announcement we’ve been posting about a little bit is the new edition of the gallery to the theme generator that’s been a major Improvement oh my God it’s it’s a major major so it’s the fun part is it’s just starting so essentially what we’ve done is take all of the previous things that we’ve been developing over the years layout scrims theme generator obviously and with pbip packages and all

4:06 obviously and with pbip packages and all that we can combine all this together so what the gallery is are the combination of all these artifacts that are open to the community to download and use so it’s super cool and exciting because it’s it’s very easy for business users to make all of their reporting look great which is we have a video out there so check it out it’s a it’s a lot of fun our subscription allows you to actually

4:36 subscription allows you to actually customize and build on top of or change all of these project files which is I think the real where the real cats pajamas is but save them up in your subscription alter the theme change visuals add more whatever you want but the the basis of all of that work to generate background themes and like good-looking reports is pretty much done for you so like you have such a huge Head Start from just using our tool and

5:08 Head Start from just using our tool and building from scratch and it also opens up the door to the community and contributors so on the blog post and maybe we should post that in the the chat here Mike There’s an opportunity for if you want to contribute because you are also a theme generator oh that’s that’s actually kind generator oh that’s that’s actually funny of funny a producer of themes and your theme to be in the gallery for the larger Community to download and use fill out a form on that blog post and and we’ll

5:39 form on that blog post and and we’ll take a look at you your your stuff and get you access and let me just tell you as an Avan theme creator and someone who uses this tool weekly not because sponsored by powerbi tips or not because we do a podcast Tomy you must we for Tommy too if he doesn’t we can’t be friends unless he uses our tool not be farther from the truth but and who just likes to use the best tooling out there I who finally saw that this week this is huge because I’ve always loved seeing what other people have done and

6:09 seeing what other people have done and trying to get there obviously usually starting from scratch yeah but now to have that Gallery I know there’s already too much to ask but will you also put other people’s themes there if they can publish them and just choose that’s the whole point of the contributor program so the blog post has a form for people to show us some of the works that they’ve done right and contribute your access Andor start publishing stuff out there I think I think the hardest thing people have is not just like obviously the backgrounds are incredible going be a

6:40 backgrounds are incredible going be a great start for people but just having not just a quick color palette but having the whole style there too quickly go into man guys I Kudos and and to be perfectly clear everything on the gallery is free you don’t need a subscription you don’t need to show up you can literally go to the page drop on the page pick something that you like by color and there’s even a great little little preview button so if you click on one of these if you click on the icons you you click on it it’ll open up a preview and show you here’s a sample like not sample of report but sample of

7:10 like not sample of report but sample of the pages where the visuals go here’s all the background images you get so it’s basically off-the-shelf ready to go pre-built reports for you and then all you have to do is show up download it and then you can start connecting your data to a live connected data set and I think if we think about the design process someone should be focusing on this side of things the design the images the background the layouts and someone else on your team should be focusing on the modeling side of things and what visuals

7:40 modeling side of things and what visuals make sense to to put on your report Pages these two work streams can now live side by side so if you have a larger team you can focus a style and design uiu side of the world and then you can have another part of their team so this just takes away care takes care of all that so it gives you at least a good place to start yeah yeah but it’s it it and I I think Mike you and I talk about this a little bit in in the video itself but there’s this real like I think unique correlation for this is for business users too right like there’s a large gap between business intelligence

8:12 large gap between business intelligence people who are already developing reports for wider sharing yes but the the the story here is if we think about PowerPoint right it’s not if you’re a business person you’re presenting slides to your manager or somebody else you’re not just throwing text on the slide you’re trying to make it look and you do that and every slide you you’re making changes and making it pretty and like this is the same concept there yes right and that’s what you get in our free version of this is like you just get these slides that look really

8:42 just get these slides that look really good Y and you can tweak them and you can change them in the powerbi files where that get really cool is once you understand that hey yeah every time I present something if I’m presenting a powerbi report and I’m building all my own stuff wouldn’t it be great to have the master slide file part of of PowerPoint MH and that’s where having the additional subscription where you can download tweak modify make that original work your own and tailor it to all your needs really comes in strong so

9:12 all your needs really comes in strong so it we’re already doing this presentation type thing I think this just opens the doors to a much wider audience and we’re really excited about it who would have loved this the most chuck chuck Sterling would have died I actually I I did a I did a full Academy Day with him and I I still use a lot of that training I have to redo our demo on The Wire framing now and use this and I know you guys are not going to say this but I need to say this because I’ve seen people online what the subscription cost me rather

9:43 what the subscription cost me rather than four cups of an espresso I have three in the morning for one day a month that’s it so you guys don’t have to say anything but man alive best non cup of expresso ever at there you go and and it accelerates your excitement in the morning when you get in you can just start building reports amazing already there’s some cool there’s already some cool stuff coming out in the gallery one of them I’ll point out here is Seth and I’ve had this idea of being able to pick create backgrounds that have a transparency to them so you change the background page color and then you add shading on top of

10:13 color and then you add shading on top of it that changes the color there’s a really goodlooking visual out there right now called translucent and there is no color on it it is literally gradient shades of color across the page

10:24 gradient shades of color across the page and if you change the page color you can change the report page color to match anything that you want it’s it’s literally 100% flexible and you can just start from a a quick working spot and just build a t it’s amazing so I’m very excited where we’re coming with this many more things to come the designs are getting better we’re doing more things with it so so stay tuned there’s a lot more stuff to come from the gallery we hope this helps you build awesome looking reports on top of your data that’s really the point there awesome the link is in the chat window and I’ll also add it to

10:54 window and I’ll also add it to description as well so it’ll be in the description of this video additionally I know we’re we’re going to get to our topic today but I do have one more thing I’m very excited about and this is another area I’d like to push push into if you follow there’s a there’s a a specific LinkedIn I guess company I guess it would be but it’s it’s basically from micro Microsoft and it’s from Miguel Meers it’s called PBI core visuals and I don’t know if you saw this update I believe it literally came out either this morning or something I saw come across my LinkedIn feed and I’ll

11:25 come across my LinkedIn feed and I’ll copy the link here and you have to be a LinkedIn I guess into LinkedIn to at least see it so I’ll drop the LinkedIn link here and then this is a preview a video of the new slicer visual that’s coming there are images there are different ways to format a card there were more buttons you have to watch this video this is going to fundamentally change the game when it comes to slicers

11:55 change the game when it comes to slicers and slicing inside your reports this is going to be amazing amazing you’re going to need some time to really sit down and digest all the things that this slicer can do because Miguel Myers who’s a wizard with amazing things designing in parbi he was doing this stuff like really early on before parbi had all these new fancy features Miguel was designing things that I didn’t understand how it worked I just literally flabbergasted and so what he’s done is Miguel’s now hired by Microsoft he’s now on a team and he’s leading a

12:26 he’s now on a team and he’s leading a team of BU of of revamping and building new vision uals called PBI core visuals he’s branded it of course with his branding and everything he’s he’s always got branding things on top of everything so it’s awesome I’ve been helping out with that team of people to test and develop and design and and think about what needs to be included in these visuals because there’s a lot of things you can build but Miguel’s doing a really good job of picking some of the best most challenging features and he’s giving you a demo of this new slash that’s coming out it’s going to be amazing so anyways

12:58 out it’s going to be amazing so anyways I just wanted to point that out there you have to go check out the LinkedIn post you got to go watch the little snippet or video it’s a it’s a short little video of things that you can do with it but wow it’s going to really change and enhance our reporting capabilities I was throwing away slicers because I just didn’t like the experience yeah now that we have this slicer with all these extra features I’m thinking I probably need to revisit a slicer now I need think about yes it’s coming up soon for his

13:28 about yes it’s coming up soon for his well his says coming in November so it’s coming soon crazy so it’s not out yet so stay tuned it’s coming but’s he’s giving you some really big teasers here they’re I guess in the final stages of testing since it’s coming out in November release anyways very excited about that super excited to hear that the visuals are getting some some much needed love here as they’re getting some updates there’s a lot of missing features comparatively to from powerbi to other tools that are out there right now so anyways I just want to point that out as well I think it’s a really good link and

13:58 think it’s a really good link and really appreciate the the hard work that he’s putting into that 100% I am I like nothing in almost not in all of powerbi but one of the best things that Microsoft has done is hire him a long time ago yes and put him in charge of this because everything we’ve seen thus far but like I’m so like looking at this I’m like yep now I now I remember all of all of his influence on making making things look fantastic and

14:30 making making things look fantastic and we’re we’re going to get to play with it and it’s going to be part of the tool by default that’s amazing super cool anyways I thought that was really good to put point that out there as well there’s some very exciting things coming in the next couple releases it seems like a lot of the energy has been moved over towards fabric it’s been a lot of you towards fabric it’s been a lot of it feels have you have you ever know it feels have you have you ever watched little kids play soccer right you watch little kids play soccer again soccer the ball moves and then there’s

15:00 soccer the ball moves and then there’s like a little swarm of bees that like and they Buzz over to the ball and they all swarm the ball in the different so as the ball moves you can literally tell where the ball is without even seeing the ball because there’s a little horde of children that are all running around the ball no one’s playing positions everyone’s just trying to get to the ball and just randomly kicking it sometimes it felt like it has felt like recently Microsoft has not been playing positions and we’ve been like oh fabric let’s all run over there and so everyone runs towards Fabric and then oh powerbi let’s run back to the other side like so I’m feeling like we’re seeing some really substantial

15:31 we’re seeing some really substantial improvements on powerbi visuals getting some substantial love back to the core part of powerbi because we I think we really need it I think we need some more enhancements there so it’s been encouraging to see the the team spreading out a bit a little bit more it feels a bit more even as we develop things along those lines have a last opener here before I get going I’m getting all excited about stuff in fabric now I was doing some testing the other day other day have have either of you had a chance to

16:01 have have either of you had a chance to test shortcuts inside fabric it’s not part of the workflow so I I’m sure you sounds like a really good example right now well I don’t have a good example I was just exper okay one the documentation on Microsoft’s website sucks I don’t know where to go to get the links it doesn’t make me it’s not very clear I had to try for like 30 minutes to figure out what’s the right combination of so let me say it this way there will be a video incoming of how you can connect an external Delta table

16:31 you can connect an external Delta table into your fabric environment so what I have done is I have lots of projects that I work on where I do a lot of data Engineering in data bricks it’s all in spark it makes Delta tables so I have a ton of tables just living around some of those tables are tables that I’m using for powerbi monitoring I have a whole bunch of apis coming in I’m ging the data down I was like I wonder how how hard it would be to make a shortcut to those tables well once you get the right URLs and get it all synced up it it’s amazing because I’m doing all of this data engineering work outside of

17:02 this data engineering work outside of powerbi and I just provide the folder you path to my Delta table boom it lights up cool it’s automatically in the warehouse wonderful I can write squel against it sweet and then it’s in a default data set already ready to go so I had from connecting to the blob the the the Delta table externally sweet to like within less than 30 seconds I’m in report trying to rip out columns and say what’s this column doing what’s that

17:32 what’s this column doing what’s that column doing I was flabbergasted I was like okay this is what I want to see I want to see other tools that can do data engineering things and lift that stuff and bring it directly to my data sets that’s what I want I want my data set to directly dude I was floored this felt like this felt like next level stuff that I had just stumbled into I was like oh my gosh this opens up a huge world for me me where I don’t need to engineer everything inside a Microsoft inside a Microsoft tool I

18:03 a Microsoft inside a Microsoft tool I can do it however I want as long as it gets to Delta format I was I was very impressed well and but this is what’s really cool with that that whole path and I hav’t I like I’m glad you vetted it and tested it because I need awesome there’s one other piece in here I need I need to talk to you on the side about which will be interesting once they actually light up fabric in Central and North Central where I have all my fun stuff yes but the the point I was

18:33 stuff yes but the the point I was going to make is it you don’t have to rebuild all of your engineering like if you’re already in doing data breaks and building and all that stuff like it’s fantastic that you’d be able to leverage that in other areas for different things without data movement right it’s another without another step to copy data somewhere and that is one of the really unique things of of them introducing that feature in fabric which I I love glad glad you’re betting it for me and making it easier for me down the

19:03 me and making it easier for me down the road when I any company who’s using data bricks to do some data engineering things whatever they’re doing any company they will find immediate value from showing up and and using their existing engine you could have been doing this for five years six years and the fact that I can just now hook into the table no no import the table just appears a little obsolete but we talked about the migration too they actually just

19:33 the migration too they actually just announced too you can now push data verse as a shortcut to fabric okay another one I was like oh my gosh another another really good like it’s literally a one button click thing well have to read the article a bit more I haven’t done it myself but I have a customer who’s using Dynamics and I’m like dude you going need to check this out like just just do your data work inside Dynamics just keep it there just make what you need to do tables views whatever you want do it in Dynamics and just boom now it’s a table inside powerbi that’s huge so there’s a lot of really interesting things that they’re doing and I feel like they it’s

20:05 they’re doing and I feel like they it’s the the vision for what they’re producing now I think is the right vision and it’s giving me a lot more options to be able to design what I need to with the tools I want to work with which anyways didn’t they didn’t they also just release colum and Ro level security yeah colum and Ro level security was just announced as well too there’s this is what I’m saying like I feel like the ball has been running around the court here the field and everyone’s huddling around the ball and like finally I feel like we’re starting to spread out a little bit and like some features are coming across across the platform now as

20:35 across across the platform now as opposed to everything just being fabric Centric yeah anyways super excited about those things I think that’s all I have for introductions of

20:44 that’s all I have for introductions of things let’s get into our main topic today we’re going to talk about the powerbi Linguistics schema and who’s using it what’s the impact of it how’s it going to go from there so Tommy give us some more details around this topic and where do you want to see this go yeah so I this came up because in the September update to powerbi they actually released a really really great new feature to the QA tooling and the QA tooling is something if you haven’t seen it before in powerbi desktop actually allows you to look at the model and

21:15 allows you to look at the model and have related words so if you want to say well we have councils but we really want to call them products or we our sales team calls it something else units whatever it may be we can associate those words together so if you wanted to say for example like hey what’s our you say for example like hey what’s our what’s my best customer like know what’s my best customer like what customer may not be in that model so we can form that so about it’s been great it’s a it’s a really nice set of tool and you can actually even see what other people in the service have asked questions say oh everyone’s asking this way but there

21:45 everyone’s asking this way but there were there were still some limitations where you had no association because if you want to say like well who’s my best customer well or you like we can’t really associate certain words in terms of binding together so that’s what they just came out with now where we can actually Define the adjectives and the verbs where we want to say like hey what’s the who sold the our what’s our best books or who sold them now we can say well we have stores or we have these managers that’s they sell books

22:16 these managers that’s they sell books and that idea has never been there where you always have to Define each field in your question so now we can actually say who sold the most books stores sell books and you can teach power that so binding Fields together to say this one is this one prits the other is really neat and there’s actually now new tab in QA tooling that I’ll actually choose where the adjectives fit in What fields are binded which one’s in a sense apparent to the other and this is I

22:48 apparent to the other and this is I think one of the very needed things we’ve talked about this before but I think this opens up the question again or opens up the door on okay is this enough for us what does this actually mean for us are we going to start building right away with this or what are still some of the limitations because we’re getting closer and closer to associating our models with really that literacy but that doesn’t mean the organization is yeah this is going to be interesting to go into this topic a bit more so just just for reference here

23:18 so just just for reference here Q& A or the Q& A portion actually refers to questions and answers so Q& A is is the feature and I believe I think they’ve actually removed honestly if you have the new visual editing experience turned on I think wasn’t it previously you could double click the canvas and the Q& A would just pop up I don’t know if that’s still on by default or not but I believe you can turn that off now and so in your settings for the reports you can just

23:48 settings for the reports you can just have that not double click turn on the visuals there so you can’t get cute so in some respects I feel like the language or the the linguistic schema they’ve I I don’t know moved away from it I don’t let me say it this way I don’t have a lot of customers coming to me saying hey can you help me improve the linguistic schema of my data model so I can get people to ask questions of it better we’re struggling with other things that are like way more simple than tweaking the linguistic schema of a

24:19 schema of a model well what what are your thoughts I I always remember this one because when man this is this is like a four or fiveyear old feature yes originally yes right so it’s been in the tool you we built a to for we have a yeah lingo right we have lingo which is there was there was no code editor no code you had to use VSS code so like you got really excited about this mic and built a tool around it so if you folks want to go like look on the site

24:49 folks want to go like look on the site yep there there’s a Code editor the the the thing I remember about this announcement the demo was super cool right like it was almost four or five years ago what the experience for presented in such a way in that like you would have with AI where you’re like I’m goingon to like ask this question and I get this answer and the reports were amazing and it it was like the most meteoric rise and crash I’ve ever seen because that is it super valuable

25:20 because that is it super valuable absolutely the amount of work required means proper model like great proper modeling everything in there you’re spending an inordinate amount of time building on all of the business speak and phrases and relationships and all of the Linguistics on top of your model so that when you’re using that interface for business users they don’t ask questions that result in nothing or that

25:50 questions that result in nothing or that don’t mean anything to the business and that was that’s where it fell down right if you’re not building the linguistic schema behind the model or on top of it then there’s there’s the model and and that interaction between the end user is not a great one so it’s interesting to me that this is part of the September update and that they built Tooling in here but it makes sense if I read the last sentence or the the last part of their their post where they

26:22 their their post where they say in the other direction defining synonyms and relationships can be a lengthy process just asking to be streamlined with the power of AI generated suggestions keep an eye out in the future for the ways we’re Bridging the two to bring out the Best of Both Worlds now that’s compelling yeah I don’t want to do it honestly no nobody does which is why I don’t think I don’t think any one of us between between our combined like years

26:52 between between our combined like years of experience have ever run into a client that was like built a solution around this maybe you never so I think two things there one I don’t think they know exist or they’ve maybe seen the QA but I how much of people are researching this this is not one of the major features so yeah no one’s saying oh this is one of our big core initiatives the other side of the coin though too that data literacy side where a lot of people are we know organizations are struggling just to get the names right in their organization

27:23 the names right in their organization that right and so much less someone says sales you get four of sales at this is one you can’t touch and if customer tell me what tell me what’s a customer I don’t know like trying to make sure you get the metric right the calculations like no no no we count to based on this and the other side of the coin so I think we’re always running into those problems I I want to play Devil’s Advocate real quick though because I think about this all right the devil let’s hear it yeah yeah so appreciate that but

27:53 that but if you were to have a proper model with all this built out and again this is one of those imagine all the people but like in that world where you had a proper model with this fully built out okay theoretically what type of value in time and in efficiency and an insights could this actually save even if it was just a team that have this fully built out let me let me I’m going to take your question I’m going to maybe extrapolate

28:24 question I’m going to maybe extrapolate what you’re looking for here on this one right so there is immense am of value so let me let me give you just some general Trends I see going on right in a general Trend typically whenever you wanted to write a program or do something you had to write it in code you had to write a specific language that the computers could understand with the Advent of chat GPT and now large language models I no no longer need to write code as if it was like JavaScript or C or whatever the language is for the computers I can now

28:54 language is for the computers I can now literally type a phrase or a sentence and I can prompt it imagine I am a and then you give it a scenario and then the large language model is able to interpret that pick out keywords that it thinks are relevant to your question and then either scrape the web or go find information or go pull out data and come back to you with an answer that says okay based on what you said here’s some websites I found here’s some details here’s here’s a written language back to you about what’s going on so I think Tommy to your point right

29:26 on so I think Tommy to your point right let’s talk about what do we what do business users want to see about their models I think we have questions I think we can articulate the questions as a phrase right I think we can and where I see the Gap growing here is I can say what were my sales for this year compared to last year for my most popular products or what are my most popular products this month but were my most popular products the last three months right there’s general

29:56 months right there’s general questions I can ask at the dat now it’ll be up to powerbi and particular teams and I think you’re going to see a lot of movement in the space around this area where teams will try to take that run it through a large language model and somehow I don’t know how this is going to work but somehow the the model definition should be interpreted by that large language model and respond to me with an appropriate

30:26 answer excuse me I was eing I had to mute myself there so I I think the Holy Grail of this thing I think what people really want is I want to be able to talk to something and have it present to me pre-built visuals and things that come out of a report that I can see your most popular product was this here’s a bar chart that explains that here’s the reason why this was the most popular the reason why this was the most popular is because these three regions had above average sales by

30:56 three regions had above average sales by 10% right that’s what people are doing in reporting today they already do this so to have some more of these Munda so I see AI producing for us right

31:09 I see AI producing for us right a lot of the hum drum common things right why can’t I just tell parbi hey parbi I built this visual remove titles clean it up make it work on a dark background and it should just know how to restock iiz the theme properties of that Visual and then make it for me or I should be able to ask questions specifically to the data around things and get it to give me answers whenever I’ve used this feature previously it’s always been a little bit

31:40 previously it’s always been a little bit lackluster to me it it has it hasn’t wowed me in the same way that chat GPT does so I think we’re looking for I think that I think the expectation around Ai and AI based things with the with the introduction of chat GPT and all these other large language models now appearing I think the expectation has gotten a lot higher for what AI could do to make your job easier and so I think Microsoft is going to have a very high bar to jump over now with these new large language models in place do you though because like

32:12 place do you though because like so what what piques my interest is obviously manually putting together a linguistic schema to work in a model to work within QA is if you want to go down that road I’m sure it would be useful for the business to some degree the amount of effort I don’t know if it’s worth it yeah when you can present the data in different ways now when you’re talking about the interaction between J Chachi GPT or large language models right they’re so far I would say my experiences they they help me with the

32:43 experiences they they help me with the mundane work insanely well some of the coding things yeah but if we’re talking about like all of the sudden having something that just magically somebody can ask any question specific to data have you guys seen anything that’s like stress tested that right like and this is this is what’s peing my interest here is why would I why would Microsoft go and

33:13 would I why would Microsoft go and invest more time into tooling that they haven’t done for four to five years unless all of the sudden that’s really important right where where something like the linguistic schema as it’s set up where it’s like this is like this this is the relationship here this is what’s going on blah blah blah not maybe like to tie into the models for sure but at the same time like I I don’t know how far that extends right like are there prompts are there helpers much in the same way that when you’re doing machine

33:43 same way that when you’re doing machine learning right you you ha you’re putting in prompts to divide data in different ways right or like create your hierarchy of decisions or evaluations along the way you’re and you’re you’re more into this than I am Mike there’s words for this that I’m just butchering and hacking through but ultimately it’s it’s a lot of it’s a lot of input to receive the expected output or the evaluation that

34:13 expected output or the evaluation that you want and I don’t like I I don’t think large language models are there from the standpoint that like you see stuff out in the market where it’s just oh yeah well everything’s just going to be business business user and they can they can ask for whatever they want and they’re going to get it and it’s going to be accurate and no no there’s a lot of nuance in business right and how do you take that Nuance that business logic everything that’s up in people’s

34:43 in people’s brains and put it in a place that can even be digested by a larger language model and that’s where I think this gets interesting because I haven’t seen any of that advanced more diving into the data side of things that automatically replaces business intelligence professionals no no and I don’t think anything’s there but I think Mike you said something that makes me it made me chuckle be like where the QA questions that we’re always thinking are who sold the most products this year or like what were my top five products and I always think to myself that’s a Google question and that’s

35:14 that’s a Google question and that’s that’s something you type in Google and no one that’s not totally your report already doesn’t have that where’re that’s not what I think people want to ask or where they would want to invest their time is show me my last product from five years even though I have 10 reports that do that already you have 10 reports that do that already they say give me a breakdown of my know they say give me a breakdown of my sales and just like the same type of question you’re ask in chat chat GPT I I think going with the large language model thing if your company has a glosser glosser you can the you there incredible what you can do with you can

35:45 incredible what you can do with you can just feed those in I want I W I want to point on this glossery thing going though yeah okay so remember we had an episode like I don’t know two 3 weeks ago and we talked about would it be nice if chaty P hit a lot of the lwh hanging fruit fruit right we just have people interacting with data models and data that comes out of our source systems and we just automatically know there’s a whole bunch of things that link together right the the common language of the business is just overlaid into multiple things so I’m in the tool and I’m looking at it right now I was playing around with

36:15 right now I was playing around with adding some synonyms so you have a table here’s a table here’s a columns and it basically rolls through the synonyms on that table so here would be synonyms for what you would see potentially in this table there is this feature now that comes with this and maybe this is what they’re releasing and I didn’t quite get it when I read the blog post about this and maybe this has been there for a while I don’t know but there’s now a new option if you add a synonym to your data model it says do you want to share this synonym with your organization you added a suggestion in the synonym so that way for example if someone else for looks

36:46 for example if someone else for looks like for example I could say the word country the column is country right you could add a synonym to country that meant location cool it works for my one data model but that needs to be into a global glossery for my entire organization and I think that’s what this feature is alluding to this feature is saying share your synonyms with everyone in your organization now that is something I can get behind right because this is what we were talking about before AI should be should be

37:18 about before AI should be should be producing a lot of these low weight value ads across the entire organization so when we say the word customer there might be other ways the business speaks about that phrase or that column or location or region like there needs to be likely other synonyms with that and as I invest my time on this I need my time invested into making those synonyms available to my whole organization everyone can now contribute appropriate synonyms to this organization that that is powerful and

37:50 organization that that is powerful and and now we’re actually to your point Tommy and then you said data glossery right that is what I want to see I want to see people build reports make synonyms do things there and from that I can now start seeing a graph information data about where are my company synonyms what are people trying to relate together language-wise and does someone need to review that holistically not in just one data model across all data models that I think would be really cool and maybe that’s where they’re going

38:20 and maybe that’s where they’re going with this one but this feature looks really promising from a sharing and communication around that thoughts you said like it would go to all data models and that’s an interesting leap because like some of some isn’t some of this data set specific depending on what team you’re talking about and like there’s always going to be conflicts but wouldn’t there be like a column if you have a country column right there’s going to be some things that are going to have common terms with them no matter what you’re going to do with them so

38:51 you’re going to do with them so country brand yeah your product well and it’s not to say it’s not to say that actually well in all actuality your your point is correct because it’s a synonym so just because it was created in one data set doesn’t mean it’s not applicable to another yeah right where yeah somebody may be asking about location yes so yeah exactly so I have another another here so another example right I’m looking at another table has products in it the column is called Product well the thing shows up and says here do you want to add some do you want to add some approved synonyms okay cool

39:21 to add some approved synonyms okay cool they have another one item artifact merchandise Pro produ produce right so they’re actually giving you some there’s an AI suggested thing and do you want to include this column in Ai and Tommy this is the other thing that we had problems with before I never had the ability to toggle on or off hey I don’t want the index column right in AI just don’t do it like yeah the surprise the larger the larger the ID in the index column the higher the whatever whatever like okay like that doesn’t help me at

39:52 like okay like that doesn’t help me at all that’s adding any value to when I start asking questions so we really do you need there’s a lot of things in the model that are not adding value to the analysis that we’re trying to do that’s really something you want to exclude no and when I said glossery too I’m surprised the way you took that I would think about this too in with the large language models where you can actually feed in documentation or you can feed in in a sense your own definition this may if they get to that point which I think this would hopefully from a

40:22 this would hopefully from a technology point of view be a catalyst to bridge the gap from a lit for an organization where if you’re feeding in your definitions that your organizations actually spend time to do and that becomes the supering of when we say customers we’re always you when we say customers we’re always speaking about either this table know speaking about either this table and SQL and here’s the business definition and that can feed into everything else well you’re gonna understand what tables you’re using but it’s also forcing people to speak the right way too because you’re right it’s not just living in a single data set yeah now the question is why is

40:53 yeah now the question is why is Microsoft doing this why are they spending so much time on this well I would be really interested to read the fine print when you’re updating here now because what’s the one product right now that doesn’t have co-pilot at Microsoft powerbi powerbi which is very interesting but it makes a lot of sense because it’s just harder to implement maybe it’s a lot easier to say build me an app and then bu it’s like wow build me a dropdown and compared to actually working with your data model like this is in power automate this is in power

41:23 is in power automate this is in power apps in a lot of places and what does AI need in order to work especially when

41:28 need in order to work especially when it’s the data model it needs to understand how not just us but they they’re going need to know how other people in a sense associate the words together hey stores always sell books books don’t always sell stores ever and we’ve seen that with all these inputs yes I think that’s part of it too but no I this is one of those points where like in order for any of this to work an organization has to be solid on their own definitions in the relationships and this again I’ll I’ll just say this the AI experience like

42:00 just say this the AI experience like when from one who used to edit the in linguistics model previously impossible no way you could do it this new way of showing here’s recommended synonyms that the tool is providing to you like I have again I’m looking through some other columns here other practical examples right sum of cost of goods right that’s that’s my column my measure that I built right it gives me other options total cost of goods summarization of cost of goods aggregate of cost of Goods right it’s giving you recommended phrases that it thinks of this stuff and I would hope

42:32 thinks of this stuff and I would hope right please Microsoft like if if I’m doing this in my company there’s there’s definitely a way of you using this data at the backend level and saying hey look there how many cost of goods columns there are across all of Microsoft’s data models across every customer that they’ve ever made there’s tons of these things there’s tons of sum of sales right so Microsoft should be groom like what is everyone talking about and being able to say well per our recommendation engine we’re feeding in

43:02 recommendation engine we’re feeding in Oodles of data across everyone’s models and here’s other people and how they’re relating it so this whole experience could get just way better just by Microsoft scraping their own data and supplying it back to people like they’re going to train their models on their own data like this it’s interesting to see like especially if you’re getting recommendations for synonyms whether or not as you share those back throughout the organization whether not those are then recommendations that pop up for other people I I would think that more so than Auto getting

43:34 so than Auto getting applied it’s syncing it’s syncing them with my whole organization so as I add synonyms to these columns it’s syncing that with I don’t know something in my or so again I don’t know how it works yet I don’t know like do if I publish a separately totally different model with the same column name cost of goods or cost of sales right does that in fact automatically apply that s for me maybe I don’t know like I I have to I’ll see have to do a little more testing around this but I finally feel like we’re at a place

44:04 but I finally feel like we’re at a place where the the needs of our business are actually being addressed by some good UI good recommendations like to me this is feeling a lot I’m I’m willing to try the feature now because I could see the potential of it now versus before I tried it and I was like I was turned off it wasn’t really great do you turned off it wasn’t really great do where those know where those recommendations really need to come from what we what needs to happen I think they need come from Tommy Tommy no yeah GP governance yeah exactly I Tommy

44:36 GP governance yeah exactly I Tommy someone say toal pasta yeah but but honestly the like if you say you’re using a measure like Master accounts like hey we noticed the certified data sets are using Master accounts is this the same metric yes and having that’s the association that’s needed because obviously everyone’s using like country but is it customer country countes store countryes so the recommendations that Microsoft hopefully will apply and what we need to build is it’s coming from either a certified data set or what you call like that real

45:06 set or what you call like that real master data set of that certain area I have if I’m using certain words like do you mean store countries do you mean customer countries based on these data models and that’s where it binds together that’s where this makes a lot more sense in terms of I think a much higher adoption rather than me having to create the wheel in every data set I would agree with that one and that’s but this this is where I think our conversation was in our prior episode right when we talked about AI previously was where are the things that I spend

45:36 was where are the things that I spend most of my time just chewing and doing things through like that’s the stuff that it should just be automatic and we were talking a lot around less around the data modeling talk about more of the the business rules the collecting requirements how know people if we get people asking questions to their data sets and it actually adds them some real value why not collect all the questions people are asking like they do have so there is inside this new UI there is this area where you can review questions so review

46:07 where you can review questions so review your questions and again this I think this a feature has been out for a while but maybe the UI is slightly updated here but when you look at the the data model you can actually say here are questions people were asking and and in of the model and you can refine those questions and change them in a way that actually makes sense for the user potentially so again this is another feedback so to me this is a great feedback loop of if we do add questions or Q& A into a report I need the feedback from it what what are they asking did my report am I literally this is the

46:38 report am I literally this is the feedback loop of did we do we actually like this report like so so if people aren’t asking questions and the questions are very different from what I think I built a report as I have a disconnect my requirements didn’t meet the needs of the user as they’re asking these questions so I can use those questions and say how would I appropriately answer these questions and try and build new pages or new visuals that do support this so to me this is a great feedback loop from use the report

47:09 great feedback loop from use the report ask questions funnel those questions back into me and refine my report where my head goes in all of this is where where and what do we use this for like I don’t think this is a random data set that Tommy and Michael spin up and and push out to our local team I think when you get to this level of investment of time it’s the same thing that Tomy you talked about the whole time do I spend a whole bunch of

47:39 whole time do I spend a whole bunch of time making a pretty looking report or do I just go to themes. powerbi tips and download one from the gallery and just start from there and be like okay here we go this is a this is an initial preview like here we go start off so in my mind this this is one of these items where if I’m going to invest a lot of time refining Q& A putting it on my reports grooming the questions down figuring out what people are asking you figuring out what people are asking building synonyms all these extra know building synonyms all these extra bits of time we’re now talking about

48:09 bits of time we’re now talking about this is a central data set this is something that’s bigger than just one little thing and this this is a certified data set this this is this is the the central bi team is doing this work or you’re planning it out in maybe a Sprint or two where you’re spending a day half a day just going through and seeing what’s in there but yeah anyways I’ll stop there I’ll make it interesting for you guys I the last 10 minutes I got a scenario because I put on my Consulting and thought if someone asked me hey Tommy we want you to like really build

48:41 Tommy we want you to like really build up for at least one Department right now logistic schema we really want to invest in this like what would be the project plann even if it was just for a subset of the organization like sure what does that structure look like yeah we’re not doing it no one’s asking but okay then how would we in a sense build something up so I want to see what your idea is and off the cusp so to speak or both of your guys in terms of that would be your investment for the next two quarters what what does that structure look like building that oh let me let me give you just you’re

49:13 oh let me let me give you just you’re already pretty on point with me on a few things but that’s why I was like so let me let me first give you a lens into my mind of where do I place the investment where regardless this is a time commitment to something I’m going to have to be doing something above and beyond so my first inclination is one do I have any data sets in my organization that are of a level of where I would want to certify them okay is it certified already if the answer is yes how many certified data

49:43 answer is yes how many certified data sets do I have cool let’s look across all certified data sets in our organization which ones are being used the most so I would invest my time using this feature only only in certified data sets that have the highest or second highest amount of usage in the organization now that’s just around Q& A the second part of this is in my mind I’m thinking this is not adding value unless you actually have Q& A enabled on

50:13 unless you actually have Q& A enabled on the report in in something that you’re actually collecting the data right so if you build a report so if all your reports in your certified data set even though it’s used the most there’s maybe an opportunity for you to redesign some reports but are you willing to put a Q& A box in those reports and let people actually ask questions of the data set so to me the the linguistic schema has like a two-part requirements is it used enough is it Enterprise

50:43 is it used enough is it Enterprise enough oh three parts maybe and am I willing to put a Q& A box on the report that may not fit this like if you’re trying to cater a story in a report you’re not just putting Q& A and there’re and saying figure it out right so there there’s I’m not sure I’m willing to risk my bi’s what do you call it like it’s not credits but Integrity in Integrity credits right it’s it’s like credibility that’s the word I’m looking

51:13 credibility that’s the word I’m looking I’m not sure if I’m willing to risk The credibility of the bi team to have a Q& A feature show up on something that’s that I think should be used but yet get very mediocre results out of it right if someone answered asks four or five questions of it and I haven’t done the investment of time on me to build good questions to ask about the data helping you like this is a learning curve for the user they they barely knew how to make personalized bookmarks and yet we’re telling them ask a question of your data model and get real answers out of it like to me that’s quite a jump

51:43 of it like to me that’s quite a jump that’s fine but I’m saying if the business asking they want to invest in this they want the Q& A in there who’s asking for it I have never had anyone

51:50 asking for it I have never had anyone ask me for this thing it’s a hypothetical scenar scenario I’m not going to I’m calling you after my business wants this because my company me me I’m shouldn’t take too long no but the barrier is so low my my my point here is because no one business is so far behind where I’m asking random qu Rand not random I’m asking questions of the data we’re still just trying to work through how do we get logged in how do we get through personalized bookmarks how do I how do I share something from this report in a way that is is approved by

52:21 report in a way that is is approved by the corporate team our threshold for getting to like Q& A I think we’re so far behind where we need to be that we’re still just covering a lot of fundamentals so yes this is cool that it’s there I really think in my opinion this is probably a Gartner feature so when does Gartner come out what you is that in February is that March when does that thing roll out you gota have the AI you gotta have ai in the thing you got to be able to prove how to do it like that’s to me that’s like a gardener feature you got

52:51 that’s like a gardener feature you got to have it because every other tool is doing some form of this at some level level Seth do you have a different answer if let’s say does like my answer he’s gonna go to Seth now I’m I’m GNA see rephrase rephrase your question let me rephrase it you get an investment they want to do a pilot on this at for one department at your organization they want to invest heavily on getting this working so people can begin to ask questions it’s only a subset of the company softable company what’s your in a sense where do you start what does that project plan

53:21 you start what does that project plan look look like well I’m not going to give you a project plan but I I agree I agree with the main tenant that Mike said the only one I would add on there would probably be choosing an area with the least amount of ambiguity in business speak right so whether that’s just really straightforward areas like sales right there’s opportunities there’s you right there’s opportunities there’s and as I say that I’m like well know and as I say that I’m like well wait a minute there’s a lot to it like

53:51 wait a minute there’s a lot to it like you’d have to think through like the the business logic or how much you’d have to develop and I’d choose an area that has the least amount of variability or friction or like reduce the amount of relationships and synonyms and all this stuff that I would have to build on top of and this is where I think you get the least amount of push into this because it requires fantastic data sets right very well modeled star schema relation relational

54:22 modeled star schema relation relational models that like tie out the way you would want them to all the time and to Mike’s Point like use the most would be fantastic as well right but I don’t know if I necessarily agree Mike with bringing the user along with all these additional features that we have out here right now because this technically would be the easiest feature ever for them to interact with it’s just we’re not confident of the results or yeah or confident that they would get the out I’ll I’ll give you a good I’ll give you an example I’m literally testing it out

54:53 an example I’m literally testing it out right now I’m trying to like add some suggested questions and I thought okay I’ll just do my simple like I have a column called sum of sales measure and I have a country right I’ll do What’s the total sales by country and I added my I added my synonym total sales is is one of my synonyms it and then it just produced me a great visual but it just counted the number of records of sales it just said count of sales like that’s that’s not what I wanted but it’s it’s that stuff like you look at the bar chart you’re going okay it’s a bar chart it’s

55:24 you’re going okay it’s a bar chart it’s literally a St line like every bar is the exact same height that tells me zero about what I want to know so like it’s you need a diversify that’s I don’t know like but why but this that’s that’s the stuff though it’s like it’s to me it’s like those things it’s like if I say something I want I’m going to interpret my lens of the world how I see things I’m going to ask questions that make sense to me and if that visual doesn’t like immediately pop out the right answer it better do a good job of saying hey we heard you ask this here’s three

55:56 hey we heard you ask this here’s three different ways we could represent this data here’s a count of sales here’s a sum of sales here’s an average of sales like here’s three this C this is why code to generate graphs work because it it forces the user to be very specific about the things that they want yes is what what is I I’m I’m not trying to be derogatory but business intelligence people or people who understand data have nothing to worry about like the vast major name me one solution like and

56:27 vast major name me one solution like and you hear this from product people all the time too what we really need is a solution that just lets everybody build their own visuals yes no you don’t because nobody uses it the lay person in the business does not understand how like XY coordinates they don’t understand how to group by different things they don’t know some calculate like they don’t know how to put this stuff together to get the get to the part of which is just analysis which is what we provide there’s so much work

56:57 what we provide there’s so much work that goes underneath the covers to put together a data set and the visualizations which we talk about all the time presented in such a way that they can start to do the work that they need to do they don’t need to concern themselves with all of the other stuff but at the same time like they’re not going to be able like it’s going to be a road to bring everybody up on board I think to start asking large language mod all of the specifics that they would

57:27 all of the specifics that they would need to know to generate the outputs that we we can generate for them in does that does that mean like I don’t think that there’s going to be an amazing efficiency curve for our job no I think absolutely there is it’s going to be faster and better for us to produce more for for the business and I think there’s a lot of value in there but I don’t I don’t agree nor do I think that we’re anywhere close to just opening up data sets to a business and getting accurate results that come out

57:57 getting accurate results that come out of like of like Q& A yes and agree and and to me it’s all these like little things it’s like little tiny things that make a difference right for example like if I said something what is my total sales of my segments right if you change that and say what are my total sales across segments we still understand what that means but the model can’t determine across and from that literally a one-word change produces the

58:28 literally a one-word change produces the wrong visual what’s interesting is as you’re saying that though is like you’re like that’s where I think llm would be great in interpreting interpreting what you’re tried trying to say or what you and maybe that’s where it’s like hey we’ll give you five different options to select from which one looks the the best and yeah and what I would want to produce to me more options yeah give me give me yeah I’m gonna Tye this phrase I’m going to say the word across versus from or of or whatever and it should say oh here’s

59:00 or whatever and it should say oh here’s three different ways you could have said the same thing you can click on each of those phrases and there’s this Google has trained me right as much as Google is an AI based search thing right when you started using Google you didn’t know how to Google for stuff you were just typing random junk and hoping you got what you wanted now that you’ve been using Google for so many years Google has learned you how to type a phrase into Google the right way to get the answers you want yeah you don’t ask questions Google taught you learned you important word at the beginning educated you but but it you have learned from

59:32 you but but it you have learned from Google how to use its tool effectively to get the answers you want and what we’re doing what we’re I think we’re seeing is these large language models are yet another thing that we’re learning we’re going to learn how to program the computer to get the answers out that we want and I think all of this is good AI I think is going to be really interesting I think everyone’s extreme excited about large language models I excited about large language models you’re seeing AI pop up in every mean you’re seeing AI pop up in every tool every if you if the tool doesn’t have ai in it it’s not legit and if you

60:03 have ai in it it’s not legit and if you are a a a brand new developer and you’re looking to do AIS with images holler at me because tips plus is looking at doing AI on images so like give me a holler we’re looking for students in school who may be looking for a senior design project so give me a give me a holler if someone who’s looking for some stuff but as we think about these things like it’s going to make things better it’s just it just needs some more time to get the right mix of how what what information

60:34 right mix of how what what information can we feed it so it does add value to me right now and I think what I’m saying right now is the risk is too high the value is not large enough for me to just like start whipping this stuff out and putting it everywhere at this point it’s getting a lot better and from where it started this has been a major Improvement major Improvement all right with that any final thoughts as we wrap here all right everyone’s just like be quiet Michael go to leave it alone so with this I’ll say thank you all very much for listening to the episode we

61:04 much for listening to the episode we really appreciate your your listenership this has been a good episode good topic hopefully you found some insights here either from some of the announcements or things and thoughts around Ai and linguistic schemas with that we’ll go ahead and wrap Tommy where else can you find the podcast you can find us on Apple Spotify wherever you get your podcast make sure to subscribe leave a rating helps out a ton if you have a question idea or topic that you would want us to talk about in a future episode head over to powerbi tips uhp podcast leave your name and a great question join us every Tuesday and

61:35 great question join us every Tuesday and Thursday a. m. Central and join the conversation live on all powerbi tips social media channels awesome thank you very much we’ll see you next

62:12 [Music] n

Thank You

Thanks for listening to the Explicit Measures Podcast. If you have a question you’d like us to answer in a future episode, drop it in the link above—and we’ll see you next time.