11# June Dershewitz on AI Investment, Reinvention & Skills That Survive Disruption

0:11 of a reflection on the course of my career where at certain points I realized that um even though I was um hired and valued for the technical skills that I had and had accumulated, the thing that kept me moving forward was my understanding of the business and that and I and it came as a surprise to me. I didn't even realize that I'd have to get a full understanding of the business when I came in. I was like, "Yay, I I know SQL and uh and other things like and I thought that that could sustain me." But that wasn't enough. It actually just makes things easier for data analysts and then that frees up our bandwidth to do um higher order things and that's like there's a lot of steps between that and actual dollars or sat customer satisfaction or whatever, which is the thing everybody should be striving for. AI for BI. I it's the it's the hardest thing.

1:01 Everyone just wants to chat with their data and um and it's it's it's hard. It's a struggle right now. Um and I think that you can make it work in a proof of concept setting with like um data sets that are already well curated for one reason or another, but it doesn't take very long for someone to come along and ask a question that's out of scope for the area of coverage that you have in your proof of concept.

1:24 They're like, "What do you mean I can't ask about this thing way over here?" Um, is it efficiency? Well, like we were never able to um scale uh humans across uh that much unstructured data to begin with. So, it's not really efficiency. >> Welcome to Knowledge Distillation where we explore the rise of the AI analyst. I'm your host, Katherine Ryan, CEO and founder of Ask Y. June Dave is one of the OGs of our industry. June, I think you started as a web analyst at a San Francisco startup in may I say the year?

1:56 You may >> 1999. >> It's true. >> Back when there was literally no template for the job. I mean, literally, there was nothing. Not even a user manual. Most of the tools didn't work. >> The I mean, if there were any any tools tools, we got reports um once a month >> and it was uphill in the snow both ways. >> Although one might argue most people are haven't evolved a lot from from that time. Yeah, we still have a lot >> reports once a month.

2:25 >> I know. >> Well, since then you've led uh analytics teams in various highly successful companies. You co-ounded uh web analytics Wednesdays. So that if you've never been to one to one, you have June to thank for it. Like it's like I recommend if you can go to one of those really you should. Um you served on the board of the digital analytics association and you confounded uh invest in data which is an angel syndicate investing in early stage startups. You also write the measurecraft newsletter on Substack which I also highly recommend. So June welcome. Thank you.

3:03 We are actually both together here in the same room and that's kind of like a little bit of a first for us. Um I'm lying. We're not exactly in the same room because we couldn't make the audio audio work, but we're like almost in the same room and we can have a drink together. >> Cheers. >> Which is amazing. >> So, before we dive in, um, what was that like being a web analyst in a in a startup in 1999 of all years? Like the hype at that time, I can't even imagine >> in San Francisco, right?

3:39 >> It was wild. It was the first um.com boom that came to San Francisco and I had just finished working as a research assistant for a mathematician on the east coast and I knew I wanted to move to San Francisco and I I thought I wanted to be a software engineer but when I showed people my resume they were like oh she knows SQL she seems curious about the way businesses work. We think we she'd make a good data analyst. And so I got this data analyst job offer and I was like oh whatever I can quit if I don't like it. And I loved it. And I think someone saw in me something I didn't even see in myself at that point.

4:13 Um that I would uh make a good data person. And we had a 200 person startup that was entirely devoted to giving away away free postcards online, greeting cards online. It was called e readings. Yeah. And I was I was on a fivep person data team um supporting this 200 person startup. So it it was it was ridiculous. It's like we we have this army of people who are doing something that, you know, one person could vibe code now. They wouldn't wouldn't even need their own analyst. Um >> give away something for free.

4:46 >> Exactly. Eventually got acquired by American Greetings. Um and so there was um there was something to it there. Uh but uh but yeah, we like at one point we decided we were going to sell stuff online. Um, and we were inventing e-commerce, you know, in 1999. And we had the for the launch of our e-greetings store where we sold um, uh, chocolates and, uh, and roses for Valentine's Day to go along with your Valentine's Day card, we we printed commemorative t-shirts that said, "Hey, let's sell something.

5:18 It's so quaint, you know. >> That is hilarious." Yeah. So, so like having gone through that hype cycle which is you know the hype cycles of all hype cycles until today I suppose right and you know everything in between >> how do you look at this one >> this hype cycle I've been through so many of them at this point um it uh and the booms the bust the booms the bust I'm like a cockroach I keep reinventing myself and finding a way to still remain useful uh which is a good thing um and so I'm used to seeing people around me get uh really uh excited about about certain ideas, whatever those are. Um you know, when it was uh mo when mobile became a big deal, for instance, that was was really transformative, I think.

6:04 Um seeing um uh uh robot cars in San Francisco now, it's like, wow, we're already in the future or we're on our way there. So, um it's uh this this kind of new thing has u become something that I've just gotten used to through my time here. Maybe it's a little bit different this time. This seems like really permeating everything and everyone in a way that some of the earlier rounds have not. >> Maybe the internet, right?

6:32 >> The just the internet world too. >> No, no. I mean I mean it permeates the layers of society in the same way the internet did. >> Oh yeah. Yeah. on the scale it's like on the same magnitude as the adoption of the internet by by mass society. >> And so you said something recently that really stuck with me. You said 2025 is not the year of uh uh data science and analytics professionals see the major shift >> uh not in the way software engineers are seeing it which I agree but what do you think about 2026?

7:10 >> Is this the year? I still think I mean really in in 2025 I think uh all the data people were watching the software engineering people and seeing how uh AI was transforming the work of software engineers and it's it's really far along now it software engineer work is is uh quite a bit different than it was just a couple of years ago. We can't yet >> I can tell you for you know obviously I have a team of software engineers >> and we're AI native uh and we have been from the onset AI native both in the platform and in the u you know all the processes around the company >> it's nothing like what I've done I mean it is it is still a lot like what I've done before in some ways but in terms of um the actual you know coding and the way you envision features and and development etc. The bottleneck has moved. It's not that there's no bottleneck. There is there are bottlenecks, right?

8:11 >> Um but it is really really different. And uh the way the software engineers work is very very very different. >> Yeah. >> And it's also not necessarily the same type of people that are good like like that are doing well at this. There's really a certain type of people that >> that just don't do well, right? >> If uh if you are I think there is something about and I hate to say this because I'm a craft person myself, but if you're really >> bothered about the elegance of your code >> and your solution, etc., it's problematic.

8:48 >> Yeah. Yeah. I I see a similar kind of attitude among data practitioners and I certainly know how I feel about that that like we we want excellence, we want perfection and there's something about the new way of working that's really disconcerting because we don't have that control and some of the stuff that's getting produced in as data artifacts. It's not high quality. It's really low low quality. Yeah. But it's it's um empowering for everyone who now can dabble in it for the first time.

9:22 >> And I think there's a real question about is it good enough >> because all well often at this point often not for us obviously right in data not really like the scaffolding that you have to put around it to make it good enough is really significant. >> It is. I actually think that some of them is go that is going to stay true for a very long time because this is data analytics not software engineering.

9:46 It is different >> without a solid semantic layer and a solid business understanding of what your data means. >> Your analytics isn't going to do anything much that's useful >> because the paradigm of data analytics is fundamentally different from the paradigm of software engineering. Right? M >> we're not building features or building platforms. We are creating a model of reality translated into data literally >> called a data model. Uh and then we uh we we manipulate it, get to some information from it and then retransate it into reality. And if we don't retransate into reality in a way that it that something can be done with it, then everything else we did is just an intellectual exercise, right? It's it's it may be very fun to do, but it's not going to affect any change in the world. That's a completely different process than from software engineering.

10:45 >> The way that you just describe that workflow is is part of the reason why analytics is lagging behind software engineering. It's it's it's difficult. >> It is. And I, you know, I can go on the record on recording to say I don't even know if 2026 is going to be the year when analytics sees the same kind of transformation that software engineering has seen. I would like to think that we're progressing in that direction. And I know I'm I'm like reading the news every single day and it seems like there's some new development or new startup or something that's promising, but so far we just haven't had that that silver bullet that has um changed things in a remarkable way as we have with software engineers who've had been through several waves at this point of uh of transformation and every time it makes it easier. Now you have um AI that's writing software you know that like um it uh the AI is designing the systems that design the systems for software engineers.

11:46 >> Well let's meet in a year have another drink I know >> in these same rooms and talk about this again and see see what happens and you know hopefully as going to be part of the reason why we can celebrate that it actually does. I would love to toast to that. >> Me too. Uh so on that note, you're currently running a poll on LinkedIn asking analysts about upskilling. Um which is a subject I'm really really very interested in because um my thesis around that is people have finite time and ability to focus. Uh we all do, right? And there is a lot out there.

12:29 It's changing very quickly. And part of what I'm trying to understand is from point from different point of views from different people different angles what should people focus on in terms of upskilling for the perenniality of um of you know their skill set and employment because >> it's wonderful to have bots helping you. We all still want to have a job. Yeah. >> Um so so why did you decide to do that and do you have any early learnings like you can share or something?

13:01 >> I do. So um this poll that as we record this is still running but by the time this podcast is out will be closed. The the question is compared to two years ago which part of your work increased the most and the options are sensemaking and judgment getting more done faster coordination and orchestration or no real change because I wanted to know I see things from my perspective I see the you know my uh individual work changing I see the work of the team I'm on and the company I work for changing but I didn't know if it's what other people see. So um and by the way this wasn't a poll that was run just to analysts. it was my entire um LinkedIn network. So, it's pretty broad. Um but what I found was that only 5% of the of the 44 people so far who have responded said there was no change. Everyone is feeling the change, but I think we're feeling it in different ways. And my I my hunch is that we're feeling it in ways that are complimentary to the strengths that we already have, right? Um, and so maybe um, uh, uh, we're a data analyst and, um, and we're used to flexing, uh, sensemaking and judgment. We might be finding ways to use that more now. Uh, maybe like me, you're the kind of person who like goes broad across entire organization. I can get more done faster. I don't need to expand on my ability to uh, coordinate and orchestrate because I'm already I've already got that. But for me um a geni in particular is like the key to unlocking getting a lot more done faster. Um what and what I found is that um sense making getting more done faster and coordination were kind of um equally um subscribed as responses here. Uh I find it a little bit dangerous. I was kind of hoping that sensemaking would come out on top because I think even if we're not doing it, even if we're not doing it more, it's the most critical thing. And if we're just like churning through it and getting it done really fast and like running all our our armies of agents and all that, but we're not focusing on sensemaking and judgment, we're creating chaos. So, >> how could we all all hope that sense making comes on top of >> I wanted I'm roo I'm rooting for it, but right right now it's actually in third place. Um, but the poll's not closed yet, so there's hope for the future.

15:21 >> It's in third place, so that means just above >> no change at all. >> Yeah. >> Okay. >> Yeah. Come on, guys. >> And also like my my guess judging from like uh the coordination and orchestration piece um being so popular is that perhaps it's data engineers or software engineers in my network who are now uh cross-pollinating with other people in ways they didn't used to. um and um also um dispatching agents to do things and then having to act as a manager um even if they're an individual contributor. So it's it's like flexing new skills for people.

15:59 >> Yeah. >> I think that I think that's a really like broadly speaking a very um strong reality across the board. Everybody is flexing your skills with this. >> Um yeah, we really are. So you frame the AI opportunity for data teams like as two distinct lanes right building AI into products for customers and using AIs for uh internal internal process improvement where like are most teams you see right now and um whether you know it's in your company or uh in your network >> and where do you think they should be? M so I made this observation that you're referencing two years ago as I was involved in uh an initiative for my own company um to do the the first case building AI into products for customers and that was one of the first kind of proof of concepts I got involved in with Genai. Um and and I you know it seemed kind of equally balanced at the time. Um in internal process improvement and genai for um employee efficiency wasn't really as developed at the time but um now I think for data teams especially that balance has shifted um and um there's we're not I think data teams are not owning the building of AI into products for customers. we're contributing and the owners are um the chief product officer or um interface designers, right? Because they're like reimagining how humans um engage with stuff online. Um and we're and we can contribute to that in the classic product analytic sense like we always have. But um using uh Gen AI for process efficiency and organizational efficiency absolutely I think we we have um we're beginning to recognize the potential of doing that. Um getting this I hate to say this getting the same amount done with fewer people or um getting um the same size group to actually produce more of value. Um and so I think that the balance is shifting for data teams to the latter. First like as individuals, how can we make um our own how can we improve our own personal processes as data teams? How can we improve the processes of our teams using new tools available to us and then um working together with our stakeholders how we can we make the entire company more efficient and productive?

18:42 But you to your first one which is like making more done with fewer people. It sounds bad obviously, right? Except >> in reality I've never met a data team that didn't have too much work. >> Never. I know. >> Never. Yeah. >> Never. Right. Exactly. And I mean and I mean way way too much work. >> So getting more done with fewer people doesn't actually sound that bad to me given that there generally is way too much to do anyway. Yeah.

19:12 >> And that the part that the AI would do for you or with you >> is not the part most people like the most. >> It's like all these brainheavy sort of coding and and making sure the the syntaxes are cor syntax is correct and you've got the right variable names and everything crisscrosses correctly uh you know etc etc. It's very time consuming. It's not necessarily the most creative and most interesting. And I understand that not everybody's interested in creative and interesting work, right?

19:47 >> Um but it's also really not where the value for the company is. >> Whichever whichever company you work for, that's not where the value is. And ultimately um if you want to continue having a job you want to you want to produce value >> right for your the end customers of your company not just creating capacity for your team and I think that's a problem that data teams have in general um that I've seen plenty of initiatives that have been proposed that like you say well what's the benefit to the whole company and they're like well um it actually just makes things easier for data analysts and then that frees up our bandwidth to do um higher order things and that's really there's a lot of steps between that and actual dollars or sat customer satisfaction or whatever which is the thing everybody should be striving for >> and I would say if that you know if this shift pushes uh data and analytics people and teams in general to worry more about the value brought to the company >> that's a that's really positive it is >> because because you really shouldn't, my opinion, be content to um sort of be a a administrative cost center doing you know data and analytics processes. Um that's kind of not where you generally want to be. You want to be in the value production value generation part of the business. And that's I mean that's been a a very very long theme is like how do we switch from being a cost center to being a profit center >> and and I I think with with >> often limited success >> and profit center I don't know if that's necessarily possible in all organizations >> but what is possible is to have a profit oriented mentality >> so that when you are working and you are analyzing data You do actually think of who your stakeholder is and what they do and what they could possibly action where the value is and preferably left side of the decimal not right side of the decimal.

22:00 >> Yeah. >> Which is you know something that us digital analytics people have been guilty of a lot especially in our younger years. We knew everything about CTRs and uh you know and we would optimize the 20 people on the website and think that everything was was you know was was running around around that. Yeah. Um obviously things has things things have changed from that point of view right but but it is really something that is >> sort of >> an easy uh rabbit hole is focusing on things that seem important in your domain in your remit but actually the impact this the the the magnitude of the impact is just simply too small.

22:43 >> Yep. That actually reminds me of a conversation I was having with a new data scientist on my team earlier today and she was asking for my advice on the kinds of things that she should propose uh to include in her work. And I I said look at the company goals first. Read through the goals and understand those goals and then think about how you can connect the work that you do with the skills that you have to further and support those goals. And I think it wasn't something that she had necessarily considered as as someone who was younger in her career.

23:18 >> And I think that that is a choice that everybody can make, right? Everybody can make individually the choice of dedicating brain power >> to understanding their environment and >> the environment of the community they work for and make the bridge between what they do as best they can and those goals. Yeah. >> And you will fail in that over and over again as you fail in everything you learn for the you know for the first times. It's not like it's going to be super successful generally from the first from the first go uh get go but you are the person who needs to bridge that gap. That gap will not be bridged for you. Mhm.

24:01 >> And that I think is an attitude uh when it comes to you know thinking of oh AI will steal my job or automate my job. Well no not really definitively not if you have that point of view to begin with. >> Right. Right. That's a right kind of mindset to have right now. >> Um so AI is still in the exploration phase right for analytics team. I think we we agree on that. Um and changes will in my opinion happen from the ground up because that's how they always happen in analytics.

24:38 Ultimately if you don't get adoption if it doesn't get adapted by the people who are actually doing the work dayto-day nothing really sticks. Um I think that one of the trends that will force that or like accelerate that is um agent agent eiccommerce. Um because I think that the rise of agent e-commerce will kick off really a sort of an enormous work stream across organizations because they will be forced to adapt to um talking to two different audiences and I mean by by that organizations who market right obviously. So um and sell online. So all of a sudden you you have these like humans and bots type of audiences and they have very very different needs and when I say different needs I mean technically different needs in the way you communicate to them which means your website has to be structured in a certain way which means now your data layer changes >> which means you also have to think about what are you going to ask us questions that your data layer needs to answer that you weren't ask asking before because there actually weren't um AI agents that did e-commerce before. So like you know questions like how do you get into the context window and how do you get selected but also questions like how do you make sure that the agent can read your content correctly and then how do you measure that traffic on your website >> and then ultimately because you are going to lose um some of the relationship management capabilities you have with those customers because they will not actually touch your brand physically necessarily. Yeah. So how does that then change what happens in your loyalty programs etc.

26:28 And so what I'm thinking is this ultimately we're at the very beginning of this but I think that in terms of magnitude it sort of like spans some aspects of the mobile replatforming because it's a very different customer experience >> right >> some aspects of uh the rise of Amazon because there are aspects of um the loss of the relationship with the customer right that are kind of similar >> and then some aspects of uh the G4 for migration because it really is a physical change to the website and to the data layer that needs to happen right including the metrics layer uh on top of that. Um >> I'm gonna I'm gonna add one thing to that I think which is um multi- channelannel marketing when people realized that like um people were um uh you know uh going into stores and shopping digitally both and it was the understanding of like both of those behavior types combined that was valuable uh to commerce organizations like that was a big shift too and so yeah I I >> that's true.

27:33 >> Yeah that's totally true. Yes. Anyway, >> so do you think that does that that that that could force this sort of like analytical maturity shift? >> Um or does it just create panic? >> Oh my gosh. Um I don't think it creates panic. I think it's uh as someone who's been through like all all the waves of this, I think it's exciting. Um the the common thread that that I see and this is from like the beginning of my own career until now is um being someone who um uh strives to understand uh customer behavior and um and and and for me I I work prim primarily in the digital realm but that's why I brought up multi channel um at first it was you know looking at how customers were engaging with websites humans were engaging with websites and then humans engaging with mobile apps. apps. And then for me personally, personally, I went to go work for a company that did live streaming video. And so I was like, "Oh, it's not just static pages anymore. It's live streaming video. How can we understand how customers are engaging with that?" And then I went from there um to work um for a company that focused on audio. And so how are companies enga how are customers engaging with audio?

28:42 How are they just engaging with voice? even if you don't have a physical um a visual interface and and the common thread still through all of that is like understanding how customers engage with your business. And so whatever it is that happens next, it either is the same sort of thing like we're humans, we need a thing, we we're waving money around, you know, how does the business have a transaction with us? Or it becomes different where it's like the humans need a thing and the robots need a thing and those are different things. Um, but I've always also been fascinated by um, robotics and I've never worked in robotics, but I think it of it as an interesting parallel to um, understanding customer behavior is understanding robot behavior. So maybe this is the convergence, you know, >> interesting. I never thought about it that way. Um, I I I'm with you on the, you know, it's really all about the customer behavior and understanding the customer behavior. Ultimately, I think that that's really one of the things that's so interesting about marketing analytics is it's about what people do and what people want.

29:51 >> Yeah. >> Um, which I find interesting. Um, obviously, you know, one person's excitement uh is another person's panic. Uh but so you know we've talked about this like bottom bottoms up motion >> a lot of people and I think that's something that you know we probably see a lot of that in your poll or at least I kind of like feel it you know between the lines in your poll there is also a lot of top down pressure right >> u pressure especially around efficiency um >> and uh and and sort of you know transformation of workflows So, do you think that top down is ultimately how it actually plays out or do you think that bottom up actually has a chance?

30:43 >> It's it's it's got to converge at some point, but I think it's pretty starkly different right now. Um, a friend of mine shared uh an article she would thought I would like. that came out in the Wall Street Journal about a week ago on um AI efficiency and it it pulled um workers and the seauite and asked both of these groups how much time do you think you're saving each week by using AI and for the workers the people on the ground the grassroots people they said 40% of them said they were not saving any time at all with AI in their job is the current poll um and but executives said um that they believed um they were saving 19 that 19% of them said they were saving more than 12 hours a week by using AI. And so there's this stark difference, right? If the sea >> I wonder what these people are doing.

31:39 >> I don't know. I mean maybe it's just like completely different workflows for the people who are on the ground versus the people in the seauite. Um but still I think like we're we're all experiencing very different realities right now. Um and so I've seen and I've talked to other people who are getting a lot of cheerleading from the seauite to to be like we're all in on AI. Yay. You know, go learn, do um prototype, build.

32:05 And um workers on the ground um data teams especially are like, "Oh my god, I have more work than I ever did. I was my data team was swamped to begin with and now we're swamped by like the service desk work we've always been expected to do versus plus this like can you please automate all of this stuff and make me an AI chatbot that does everything that you do and um and it's it's been a struggle.

32:32 >> Have you seen any concrete wins? I mean not conference demos but like real improvement things that you've seen and touch and go went like oh yeah that's actually useful. >> Yeah. Well, I think there's a difference between efficiency and innovation that comes out of Genaii. >> And I think that uh many of us are brought into hackathons and perfect contests and whatever with the idea of efficiency in mind, but I think really what we're getting out of it is innovative solutions to problems. Um, so for example, you might be able to use Genai to um, uh, classify and structure a bunch of unstructured data and produce a new data set out of it that's useful to the business. Um, and so like the the easy production of new data sets that mine unstructured data, awesome. I would say that's innovation. Um, is it efficiency? Well, like we were never able to um scale uh humans across uh that much unstructured data to begin with. So, it's not really efficiency. Um or you know uh uh looking across um lots of say structured data to pick out patterns and um write narrative that describes um to business people what's happening. Um that's great. I mean, humans can do that too. Maybe it's a little bit about efficiency, but I think it is the case that we've never really been fully staffed enough to be able to write that kind of narrative to begin with. So, it's not efficiency, it's um it's it's doing more um doing more than we were able to do with just humans alone, >> if that makes sense.

34:20 >> It does. So if you're an analyst listening to this five years into your career um you know >> somebody not not somebody like us like you know 30 plus somebody more >> somebody who's more of like like beginning midcareer who has their entire career in front in front of us of them and their entire career is going to obviously you know happen in this world of AI um and you're obviously feeling the pressure to become an AI analyst is um >> what would you say they should be investing in in terms of upskilling?

34:57 >> Yeah. Um so I wrote this article um about a year ago called to thrive in data know the business and it was about it was kind of a reflection on the course of my career where at certain points I realized that um even though I was um hired and valued for the technical skills that I had and had accumulated the thing that kept me moving forward was my understanding of the business and that and I and it came as a surprise to me. I didn't even realize that I'd have to get a full understanding of the business when I came in. I was like, "Yay, I I know SQL and uh and other things like and I thought that that could sustain me, but that wasn't enough." And I think that now as um some of the technical competencies that data people have are getting easier to do with AI, the thing that we can really continue to lean into is knowing the business. And and I'm not saying like, oh, you don't have to be technically competent anymore. I think that that's um that it goes handinhand with business sense, but I think the combination of of business sense and technical knowledge is incredibly powerful. And that so I would encourage people who are early on in their career, if they're like, "What tech skills should I learn?" I'll say, you know, get your the grounding of tech skills that um that like excite you and give you energy. Um, but also uh you better be starting now if you haven't already in developing business sense.

36:32 >> I I agree. I think that you know when you say techn technical skills are still going to be important. I actually think technical skills are going to be >> way more important. >> Really? How's that? >> Yes. Well, because um so when I'm thinking about the main things that are important to be a good AI analyst, one is you have to really understand how others work. >> Mhm. >> What are these things? What is your tool? It's your tool, right? What is your tool in reality? How did it get trained? um really understand how the inner sort of mechanisms. What is context window? What is an attention mechanism? How does it work? What is delusion? Attention delusion. Like you really need to get that because if you don't get how it works, you're going to talk to it. It's going to do things you don't want.

37:28 You're not going to understand what is happening in the background. >> It's like working with a database and not understand the concept of projection and joints. >> Yeah. Yeah. Um so so that's one thing. The other thing is uh you need to learn to talk to it. So you need to become really good at prompt and con context engineering which does not work >> which does not work if you don't understand the inner mechanis mechanisms of the LM.

37:52 >> Yeah. >> Um and then you have to become really really good at reading code >> and I think that reading code is actually more difficult than writing code. Reading code that you have not written. Yeah, >> it's not easy. And you have to now read this code, understand what it does, decide whether it is sound or not sound, put it together with the rest of the code. >> Uhhuh. And for that you need do you need to like this is something that I've always said to everybody who asks me is what what do you think is the the sing if I had to say one single technical skill that is the most important in data analytics in my opinion is data modeling.

38:37 If you do not understand data modeling, you do not understand how to translate reality into data that is going to actually give you answers or insights or whatn not that you can retransate into reality. If you do cannot do this translation >> journey, you're just amusing yourself with data which is very fun >> but not contributing to the value uh that your company is creating. Right. >> Yeah. actually. And >> so that to me is the basis of business sense. In fact, >> yeah, I I you're right. I I think um connecting that to work that I see kind of on the rise for data teams, uh one of one of the air the categories that I see people doing more of is focusing on um uh developing and maintaining really uh well-curated data sets um that serve as input for AI >> and input for humans too. Um but um we're rec we're all kind of seeing the light and recognizing that that's an important thing to do and um and that requires technical competency like you said data modeling and also business sense all of it together >> and actually this is a wonderful segue into what I wanted to talk about next which is data quality. So in your super week uh talks to your measure craft writing um basically all throughout what you write you keep coming back to trust in data >> right trust in data data quality making sure as you just said well curated data sets >> um so there's been this promise for years that's started with NLP uh um I remember at Dorama we did something with NLP like that that you know, ask questions, get answers.

40:26 >> I think it was 12 years, 10 years, 12 years ago. Um, there's been this promise for years that a business user will just ask a natural language question and get a beautiful chart or an insight or an answer that makes sense. >> Yeah. >> Um, what do you think about that? >> I'm laughing. I like um AI for BI. I It's the It's the hardest thing. everyone just wants to chat with their data and um and it's it's it's hard.

40:53 It's a struggle right now. Um and I think that you can make it work in a proof of concept setting with like um data sets that are already well curated for one reason or another, but it doesn't take very long for someone to come along and ask a question that's out of scope for the area of coverage that you have in your proof of concept. And they're like, "What do you mean I can't ask about this thing way over here?" and they're like, "No, you can only ask about certain topics over there." And they're like, "But I don't care about that." Um, and so it's a hard problem.

41:23 Well, you know, I hear you you um phrasing this potentially as a data quality concern, but I think part of that is um like the lack of kind of a like a shared definition and understanding of what data quality means. Um, and I I battle with my own co-workers on this. mean a lack of a se of a common semantic layer on top of the significance of data quality. >> Exactly. touche that um but like um I I have this running debate with um with a colleague of mine who thinks that she thinks that um when data is missing like um from a data set like um we actually didn't bother to instrument a system to collect the kind of data we need to answer a business question that that is a data quality problem and as and I think it it's like a requirements gathering problem. Yeah. Um but like uh from the outside someone might say, "Oh, uh, we can't use that data set. It has data quality problems." The quality problem is the fact that we there's the missing data. Um, >> it's a language shortcut. It's not technically exact. Uh, >> yeah. Yeah. But I and it's it's tough because um I know and and many other people know like how hard it is to devote enough time and expertise and resourcing to develop well-curated data sets. I've done it. But like I think that the the thing that drives people to invest in that is a pattern of failure when they haven't done it and they see the result and the wreckage and they're like, "Oh, we can never let that happen again. and then they'll invest in it.

43:07 It's not something that's just like the the way we operate around here like you know we're just not staff >> nowhere. Yeah. >> Nowhere. I agree. But like your example touches on something that I think is um I would say >> these days I would consider that as being the main sort of things thing to solve >> um in data processes which is context continuity >> because it used to be that It used to be that we didn't have the tools to handle the type of data we needed. We had the data was too big and the tools weren't good enough and it was horrible and we need better tools. Uh then came the modern data stack and the tools are quite frankly good enough. Can tools be a little better? Sure, maybe.

43:54 Yes, >> incremental. Yeah, >> they're good enough. Um the problem is the context continuity uh between ultimately the steps of the analics process from connection to data to producing outputs. But those steps tend to happen in in different tools obviously with different people and with a context continuity problem within that whole process because it is not possible to document like it is not just not possible to document and so >> uh keeping the context continuity of what happened in the in the g in the uh requirements gathering which ends up with us having this data in this state and then what transformations have we made choices have we made?

44:38 >> Yeah, >> we made some decisions because we didn't have everything. So, we we made some choices and then when when the the result gets in front of the stakeholder, the first question they ask is, oh, revenue, where does that come from? >> I have to retrace the whole thing. >> Yeah, >> but half the people have left the company uh etc. Right. This is the reality of working with data. This is this this is really the reality, right?

45:07 >> I actually think that that context continuity problem is um is really one of the probably the main issue in most processes. Would you agree with that? >> Oh. Oh my god. Yeah. I think you've just described some pain that like I have definitely felt. And I think that that you haven't used this word, but the word previously that was used to describe that is well, we just need something that shows lineage. >> And um and I kind of rolled my eyes at that because I'm like lineage what? From what to what? And it's it's never extensive enough. And you can get like a little piece of it here to say there were these tables and these um transformation processes and the summary table that came out at the end and whatever. But then there's the everything that happened before that and everything that happened after that. And you ask about like where's that in the lineage and they're like oh we don't have visibility into that. You know >> the business context about the around the lineage you need to understand why these choices were made.

46:05 >> Yep. >> Yeah. what choices were made, why they were made. Um, >> yeah, >> is it possible to make other choices? >> There's a there's a lot of lore and and I think what you're hitting on right now is how this is exactly why data work is different than software engineering work and with regard to like AI maturity right now. It's hard. >> So, it is hard. Uh, let's talk about goals. Um you've written a framework for data team goals I think seven types um foundational data literacy uh all the way to like quantifying business uh uh impact.

46:47 >> Do you fit AI skills somewhere in that hierarchy throughout that that hierarchy? What what do you think? >> I think um I see it in uh two different ways. In one sense, it is infused throughout all levels of the of the hierarchy. Um, for example, if your uh data team is developing um some kind of geni platform that makes it easy uh for business stakeholders to to do a thing. Um then it could be like shipping the thing. um we're gonna we're gonna like produce a first draft of of this platform or maybe you have already developed this plat platform. You might take a goal that's adoption and usage.

47:35 How can we get more people in the company to use this thing? Um but ultimately it should be about dollars. So like we have we have a thing we built it potentially like with J Genai or using GI how is it uh relating to like the business outcomes that we want to achieve making money saving money keeping customers happy and like that should be eventually what we're striving for I don't think we're quite there yet but like when I when I look at at my this this pyramid of goals for a data team again I think where we are really um realistically right now is at the very b bottom which when I wrote this u a couple of years ago, organizational data literacy is the the the foundation along the bottom. There's um certainly something in there where it's it's also about like upskilling um everyone on a data team to know as much about um AI as they need to to do their job effectively today. So it might be conducting a certain number of trainings or um you know developing uh AI workflows or um uh producing curated data sets or whatever that is like that's that's kind of like on the foundational level of the kinds of goals that data teams are should be taking today related to to Genai. I'm going to do a shameless plug here or watch videos of the flu fees on Askwise website because >> Oh, >> they will teach you everything you need to know about LLMs.

49:06 >> Um, you you've never watched the fluffies. >> I need to watch the fluffies. I I like >> You need to watch the fluffies. It's fun. >> I have an actual fluffy in my pocket right now. If I can find it. I probably >> Are you gonna make some people jealous? >> Oh my god. Here it is. Here it is. So, um, for those in the live audience, this is a fluffy. Um, you can you can watch videos of them, but I have like a threedimensional fluffy that I'm going to take home, >> 3D printed two days ago by me. So, the fluffies, basically, I came up with the fluffies uh um completely by chance. I didn't want to have my face on social media to do promotion of Ask Why. So, I was like, "Okay, I'll create an avatar.

49:46 That's fun. I'll create an AI avatar." >> Um, it was mildly fun. Um, so I was like, well, I'm just going to create like a little monster so that it the monster appears, you know, and makes it a little more fun. >> And then everybody wanted to see the monster and nobody wanted to see the AI avatar. So, uh, I was like, okay, well, I'll just redo it just with the just with the little monster. And it dawned on me uh that that would be a really good way to explain point one of what I think everybody should know about how do LLMs actually work >> because ultimately if you think about it I don't know I thought about it anyway I was like if I think about it the flu fees really can represent um weights and biases right they are the parameters in the LLMs >> and they can be the untrained parameters in lost landscape and they can be the be the train parameters in uh the latent space and if I give them different colors and personalities they can even be the fine-tuned parameters. So like the the metaphor really works. And so I spent an entire weekend in a rabbit hole of m mapping floofy land versus a IML um concepts to make sure that everything worked in the metaphor when I was uh telling the sto the you know making the explanation within fluofy land. Um and so I started producing videos about what is trading and then I shifted to uh attention mechanism because uh I realized that really what most people need to understand is how does attention work in prompting because the most frustrations from prompting come from a lack of understanding of the attention mechanism.

51:36 >> Yeah. Um and then now I move to uh AI security because I feel like especially with you know all the mold bots etc. there is a there is a real need to understand where you actually where your data actually goes between the moment where you're prompt and the moment where the LLMs generate an answer. There's a whole application layer that makes multiple copies of your data >> and those copies persist for uh multiple reasons that are structural that aren't conf that aren't in any way covered by enterprise contracts because enterprise contracts cover um training etc. They don't cover the need the physical structural need of the system to make copies of the data for compliance for speed for calculations etc etc. So like when you use an LLM, you kind of really do need to be a little bit conscious of that.

52:29 >> Yeah, >> I think so. That's the fluffies. >> That's the fluffy and it ties back to the goals, I think, because you you believe that everyone on a data team needs to understand these fundamentals and that foundational knowledge for the rest of the stuff. So >> and I'm trying to make it a little amusing to, you know, to kind of like to to learn about it. So data teams in 2026 should be taking a goal to watch all the videos of the Floopies and understand and be able to like pass the the Floopy quiz.

53:03 >> Absolutely. >> Yeah, I agree. >> So let's talk about um tools and investing in tools and creating the new tools. Um you're a founding member of invest in data which is an angel fund um specialized in investing in data. Uh >> what are you seeing that come coming like as deal flows across your desk that you find interesting and h how like >> tell us some good stuff.

53:35 >> What's cool and what's coming? >> Sure. So invest in data is five years old. Um and so uh that that that's as much time as I've been paying like really close attention to um new innovation in the uh data and AI technology space. Um but as I was thinking back over the course of like maybe a decade uh to think about like um how data tech has changed and evolved um where I was 10 years ago I think of there was a lot of excitement around >> the data visualization layer and and and BI tools for that matter and there were like like the tra traditional things were like there were they were being challenged by some upstarts and there were a lot of different solutions in the market and Then um around 2018 through 2020 there were some consolidations.

54:25 Tableau got acquired, Looker got acquired, um Periscope it since merged and so on and then it kind of quiet quieted down. I think these days people don't really talk about data visualization as one of the exciting technologies in the data space anymore. Um there was a period of time when um data quality observability was um was um super hot. um uh the first generation of data cataloges which I mean we could have a debate about that I suppose because I think it's related to context um didn't really work out. We all like believed uh that like oh the in the vision of um data lineage solving so many of our headaches and never really did. These days though I think that there's a lot of excitement around new technology uh related to data infrastructure AI infrastructure um ML ops feature stores inference um those those kinds of things that are like you know selling pix pick access to the miners um everyone is is building all these systems and it's kind of like that that mid layer of uh working with data inside um AI native systems uh where where um people are building and people are investing if that makes sense.

55:44 >> Thank you. >> Um shameless plug. Uh where can people find you? What should they read? What are you what are you excited about right now? >> Um we'll put everything into the show notes so you can just, you know, rattle things out. >> No worries, we'll pass it with AI. >> Okay. Um I have a newsletter. It's called Measurecraft. I've been um publishing articles every two weeks for almost two years now. Um where you get to hear more about my opinion on all things data. Um uh invest in data. You uh if you go to that website, you can find out uh what's in the portfolio of the kinds of um products and technologies that we have chosen to put our money behind as a group so far. Um, one thing I am really excited about and actually I'm going to see Cat there is Measure Camp Los Angeles which is set to happen on May 30th in LA. Um, I'm I'm uh a very light member of the organizing committee there. But um if you don't know about it, Measure Camp as a whole uh event series is the best free unconference uh in the data space and it happens all over the world. And then if you uh want to find me, you can find me on LinkedIn. LinkedIn, I'm Jay Dersh.

57:03 >> Thank you. And so, yes, absolutely. I uh I actually should do way more shout outs to Measure Camp. Um if you've never been to a measure camp or don't know what is a measure camp, go to measureamp.org. It's really really good. Um actually we're going to release on our LinkedIns all of uh all of the New York organization committee. We've got these videos that we've done in the New York measure camp last year about a bunch of people talking about their experience which I think you know hopefully should give a bit of a flavor of what you can get there because it's very difficult to say with words like it's an unconference. Oh my god what's that? Um, well, you know, it's the audience that creates the content. Well, that does not sound great. It actually really is.

57:51 >> It is great. It is great. And, you know, people are like, >> is amazing. >> I have to give up my Saturday to do this. And and the answer is yes, you do. >> Yes. But probably the day after, the year after, you're not going to think, oh, I have to give up my Saturday. You're going to think, oh, I get to go there on a Saturday. >> Exactly. Exactly. So, I'm just going to plug the New York one as well because the New York one is on March 28th and I would like to remind everybody that we do have the best pizza. So, you know, there is something to be said about that. Free pizza, get your ticket. Um, and yes, Los Angeles. So, that's going to be the first iteration in Los Angeles. Uh, May 29, I'll absolutely be there. Uh, thank you for that reminder.

58:39 And uh well, thank you so much for uh doing this with us, June. It was really amazing. >> Thank you for having me. >> I'll see you at Measure Camp. >> Yes, you will. Looking forward to it. >> So, correction, Los Angeles is May 30th, right? >> Yes. >> Not May 29th. Do not go there on May 29th. >> No, that's a Friday. >> Thank you. If today's conversation made you think about how AI is changing data and analytics, visit us at askquy.ai. AI and try Prism, our data platform helping analysts navigate complexity with context. Thanks for listening and remember, bots won't win, AI analysts will.

59:23 Thanks to Tom Fuller for the editing magic on this episode. If you want to work with Tom, head to askqu.ai and check out the show notes for his contact info.