If You Want AI to Work, Fix This Boring Thing First with Veronika Durgin, VP of Data at Saks

Richie Cotton: Oh, yeah. It is interesting that in the last few months, AI has gone from being, okay particularly generative AI has gone from being it's... can do some bits of the data role, like exploratory data analysis, but it's not necessarily trustworthy to being, oh, it's actually very competent in a lot of areas.

But it's still not at the point where it is replacing data science. It's not replacing data analysts. And I don't see that happening in the near future anyway. But maybe we need to figure out then w- what do data analysts, what do data scientists, what do data engineers need to do?

Like, how do you, first of all, how do you make more use of AI in your job? And then where do you need the humans still left? 

Veronika Durgin: So I believe the future is actually analytics engineering. 

Richie Cotton: Ooh. 

Veronika Durgin: And I love this title, right? I just maybe 'cause... And again, it has nothing with the, i- in my mind, it has nothing to do with dbt, who originally kinda came up with it.

I think it has everything to do with engineering for analytics. And these are the people that are actually talking to your business users, understanding the business, understanding business processes, diving into nuances, unders- aligning on languages. They create conceptual data models. That part, that conversation- That is actually not oh, every, I don't know, healthcare business is the same across the board.

It's not. It's those nuances that matter that makes each business unique, and that's where people still need to exist. I think everything else will be automated, some parts of it more than others. So this is kinda again how I think about it. Oh, by the way, the other, just want to say disclaimer, my opinions are my own and don't represent, so before I offend anyone, my, my current company or former em- employers.

I think the action of create- creating ABI reports is gonna be also gone. I think everyone will be capable, will have enough tools that are easy enough to use where you don't need a specialist for that, right? It's I think one of the areas that will go away. Data science is another interesting one.

I think the barrier keeps getting lower and lower for everyone to kinda have access to do machine learning or use gen AI. Data engineering, moving data from place A to place B is becoming less and less of a specialty, right? It's kinda becoming easier and easier with certain exceptions. So to me, the human part kinda that's left is that actually talking to each other, like identifying nuances, aligning on definitions, and I don't think AI can help us as much as we wish it would in this area.

So that's why I think actually analytics engineering will be the kinda catch-all data role in the future. 

Richie Cotton: That's really interesting that you're bullish on analytics engineers 'cause it's half data engineer, half data analyst, and I think the skeptical- Has 

Veronika Durgin: BI right?

Like it's- yeah. 

Richie Cotton: No, the skeptical side of me was like, this role only exists 'cause you really need to employ two people, but you've only got the budget for one, so you make one person do two, two jobs. But it is having that translation between the technical side of data and the business side of data, it's always been a big problem, so maybe there is a use for that role there, just being translation between the two sides of things.

Okay. So you mentioned analytics is getting a lot easier and like particularly routine reporting where you're doing the same report every week, every month that's becoming something that anyone can do. Data science, like bits of it to a certain extent are going the same way. So maybe there's a positive aspect to this where it's good if these roles disappear because now everyone can do this, and I guess that's been like the dream of self-service analytics for a long time.

How close are we to getting there? 

Veronika Durgin: So I think self-service analytics is a bit of a what's the nice phrase? I don't have a nice phrase. Pipe dream. I think self-service is really up to the person who is willing to self-serve themselves, and in some cases it's very hard and it's complicated and it takes a long time so self-serve was very interesting. You'll get people that are willing and excited to self-serve, and there are others that are like, "I have no time. Just give me something that is very easy to access and does exactly what I need it to do so I can do my job." So it's, I don't think it's self-serve necessarily.

I think we're just making a lot of maybe complicated things a little bit easier. So for example, pick your favorite or least favorite BI tool not naming name, but let's say I have a BI tool and just can't figure out how to use it. It's just... So having ability to just kinda in plain language describe what you want, and for it to pop up a report for you, that's a massive unlock because it's easier to make changes to something that already exists versus creating from scratch, right?

I think we're lowering, we're making all the sophisticated tools that we have a lot easier to use. But again, self-service analytics also fails when your underlying data model is a mess. And I think, again, back to my analytics engineering theory, analytics engineers are the ones who are creating models, and they need to create models that are easy to understand and easy to use.

And the only way to do that is to understand who is using them and why and how, right? And what the business actually is. 

Richie Cotton: Even at the years I've been at DataCamp, we've changed BI tools several times, and it's happened in every company I've worked at. You swap the tool set and then, "Oh how do I create a dashboard in this particular tool?"

It changes, and having that being able to be at least partially automated is a good thing 'cause you don't have to get into depth learning how to create dashboards. You say, "Okay I want the end thing to look like this." And yeah, that does seem to be co- becoming easier. 

Veronika Durgin: At the same time, I also wanna say I don't believe BI tools are gonna go away.

I'm like, why would you wanna type the same question every day? You just wanna click and... there, there's value into predefined reports that are showing up at some cadence. So I al- also struggle of oh, we don't need to create anything anymore. You can just ask questions.

What if I ask the same question every day? It is annoying. I'll have to have a template and copy-paste, or I have to set up an agent that will be running it. Or we already have a solution to that. It's called schedule report. So I kinda in some ways I feel like we're trying to solve a solved problem in a more complicated way.

We shouldn't. 

Richie Cotton: Yeah, certainly once you factor in the cost of running agents and things like that, then actually- Yeah ... suddenly dashboards look cheap to build. Yeah. 

Veronika Durgin: It's and it works perfectly great and there's a ton of value to that too, so I kinda I don't wanna dismiss the fact.

I don't Think BI is going away. I think it's maybe changing a little bit. Maybe it will be, we'll have a different set of users in how they interact with it, but I don't think it's going away. 

Richie Cotton: Okay. Yeah, so I think there's a big distinction here between doing routine analysis where it's like it is the same thing every week, every month, and I'm asking a new question that is more ad hoc.

You mentioned analytics engineering as being a hot new skill, data modeling as being very important. Which skills do you think in general are becoming more important for data people? 

Veronika Durgin: I think asking questions, listening and understanding, and then translating words into kinda more of a model, right?

It's the bridge skill of having a conversation with a non-engineer and then translated it into data structure that this non-engineer can then consume, right? And it's and again it's open-ended questions. It's truly listening. It's figuring out nuances. It's working through politics. It's, that's, I think, is becoming more and more important.

Richie Cotton: Okay. Yeah, so I, I think the idea of going from a business problem to a data problem and then back again at the end it has always been, like, a very important skill for data analysts. And maybe if the middle part, the actual number crunching, becomes easier, then that becomes relatively more important as well.

You mentioned politics, office politics as being an important skill. Talk me through that. 

Veronika Durgin: It just... office politics is, it's different teams talking about something they feel is the same, but it's not. Aligning them on kinda common definitions or agreeing that they're not the same thing, actually.

Or the other... I was I just did a presentation at Open Data Science, and one of the things company politics and somebody's "You can't say that." No, but it's true. I don't mean it in a bad way. It's just, there are different priorities. Everybody has their own roadmap.

Everybody kinda focuses on their subject area. And we talked about specializing in their thing and when they step outside of the, of their world, people are saying the same thing and it means different thing, right? So kinda being that bridge across all of these things and then just trying to align.

You're becoming a little bit of Switzer- you need to be Switzerland. You need to really focus on kinda holistic understanding of your company and of the nuances of the company. So that's like in that sense, so not fun. I wish, I'm sure many of us wish AI would do that for us, but this is very much so human.

Richie Cotton: Absolutely. Yeah, even if you're not doing Succession-style backstabbing of all your colleagues, then yeah, you, you need to understand what their goals are. N- 

Veronika Durgin: Not a bloodbath. There is, and again, like every time many humans are together, there's, difference of opinions and personalities.

But I think like office politics in the sense of, different priorities, different incentives, different leadership styles, we still have to work through all of that, right? To arrive to something that supports the business and helps it move forward. 

Richie Cotton: Okay. Yeah. Definitely there's a whole set of like career skills, which I think sometimes when you you're a technical person, you just wanna be like head down, and I'm like, "I'm crunching my numbers," or whatever.

But actually, yeah these career skills are very helpful for your career, maybe obviously now I say that out loud. Okay, so the flip side is are there any skills which are becoming less important now? 

Veronika Durgin: I don't know. That's a tough kind of question. I think every skill that helps you get better, like learning mastery, mastering your craft is a good skill.

So I, I don't want to say that there is a skill... I'm actually thinking back to like my early career. I used to create formatting files for printing four-by-six shipping labels. That is probably the most useless skill I have. So maybe that would not be a useful skill. But everything else, as you go through your career, through your journey, anything you learn helps you become better at your craft.

So I don't like this kind of "Oh, f- forget about it." No, I still... Learning to program is very important. Understanding fundamentals, learning operating system, all of that is very actually helpful and useful in the way that you might not appreciate in the moment, 'cause it sucks when you have to do it maybe in college or school, whatever, but then like years later you're like, "Oh, that actually...

Oh, okay, that makes certain things make sense." So I kinda te- tell my child who's "Oh, I don't know. Is that why I'm learning this?" I'm like, "Yeah, you will probably not appreciate it until about years from 

Richie Cotton: now." Now I saw some wag on X recently who was saying like, "Oh, in, in school they teach things that like children don't need to know.

In gym class, they have to run round for an hour. A, an Uber would take them the same distance in five minutes. What's the point in that?" 

Veronika Durgin: And like it blows my mind. Okay, maybe we'll talk about this. So now they're like, "Oh everybody has access to everything." True, right? You can easily ask a question, get an answer.

But have you really looked at what types of questions you ask? I have my interests, right? And I just keep asking questions that are related to that. So I am not actually expl- exploring the world unless somebody tells me to. So here in college, my, my undergrad is in biology. I actually grew, in Petri dish.

I understand how vaccines work. That's a great thing to know. Would I have ever asked that question myself? Probably not. It is not something that I ne- So I'm like, the fact that somebody, philosophy and psychology and like... you train your brain how to adopt and learn new things that are, you're not naturally inclined to be curious about.

So I'm like it... Learning and education makes us more balanced individuals. Like, how can you possibly say that you don't need to do that anymore because, oh, you can ask an LLM a question? Would you know to ask a question? Which questions are you asking? 

Richie Cotton: I do have the idea of having other experts curate information for you.

It's a useful way of becoming better about understanding the world rather than just trying to have to figure everything out yourself, like what questions to ask. That's, it's a hard problem, is like knowing what you need to know when you don't know it up front. ... It gives me hope for the future of education.

Teachers still need to exist to tell you things because you didn't know what to ask. 

Veronika Durgin: Very much exactly. And also if you, you oftentimes, and I'm like going off the topic of data when I mentor and somebody's I want..." I'm like, "Get a job, any job, because you probably don't even know that a job like that exists."

And then you don't you can't possibly even... you will discover something you literally never knew existed, and you might fall in love with it. It's this is how my career started. Did I know there was a DB? No. Absolute- I just happened to fall into it, and I fell in love with it, and here I am.

Here's my entire career. So you, again, you don't know what you don't know. So how would you learn that? By being taught. 

Richie Cotton: Oh, yeah. I have to say, I would never have guessed that podcaster was like a real career until I started this. 

Veronika Durgin: Exactly. So you have to just go for it. Yeah. 

Richie Cotton: I love that. Very motivational try things, see what happens.

If it doesn't, try something else. Going back, you were saying how a lot of the stuff you were talking about was about how you need to have business understanding, you need to have technical understanding. Does this change the position of the data team within the organization? Does that mean you want like a business teams closer to, sorry, data teams closer to business teams? Should they be structured differently? How does this change what the data team does? 

Veronika Durgin: So I always Truly believed that data teams need to be as close to business as possible. Like org layout doesn't matter. Like it doesn't matter where data team... What we need to do is pivot from being help desk to being more of I wanna say customer success, like more proactive reaching out, not just taking requests, right?

We kinda all operate in like you come to us, you tell us what you need, you tell us what your problem is, which kind of worked, but not so much anymore for multiple reasons. One, many can now do more stuff, but also on the other side, many don't know everything that's available, so they don't even know what to ask for.

So we need to really establish, again, the relationship and truly understand what our business partners, what their day-to-day is, right? What they're struggling with, what they wanna do, and kinda start, again, building solutions to that. And again, like I think, again, it doesn't matter where data team sits.

I think always we had to focus, to truly succeed, we had to focus on actually delivering things to those who need it in the way that it's the easiest for them to consume. This is kinda again, like back to self-serve or not self-serve. For some it needs to be a dashboard. For others, they're just like, "Just give me data, I'll figure it out myself."

So but that also has kinda like... This isn't necessarily new for me because of AI, right? But I think now it's just a lot more obvious that's what we need to do. We can't hide behind business analysts. We can't hide behind tickets. We need to really truly engage with those who are consuming what we produce.

Richie Cotton: Okay. If you're moving the data closer to the business end, would you have... If, I know some organizations have individual analysts who are embedded within commercial teams or product teams rather than being within a data team themselves. Do you think that's gonna be an increasing trend?

Veronika Durgin: I think it's kinda I don't know if it's gonna increase or decrease. I don't think it's gonna decrease. It's always been the case. I think there is a mix and match depending actually probably on the size and maturity of a company, at least kinda what I've seen. I think this will continue trending that way.

My philosophy, what I believe in, doesn't necessarily mean I succeeded at it that data needs to be centralized, but analytics needs to be decentralized. 

Richie Cotton: Oh. 

Veronika Durgin: So I struggle to see how decentralizing Data will work because you have to have tight alignment, again, on definitions and how everybody talks about things.

And we all know how hard it is because it involves actually humans talking to each other. So when you have kinda like pockets of ownership and nobody owns the whole thing, it's very hard to create something that is truly easy to consume, it's governed, it's aligned. So that's why I feel like data needs to be centralized.

But analytics, again whether it's a BI analyst or a data scientist, those that consume data I don't have really a strong opinion whether they should sit within the business or in some sort of centralized data team. I think, again, both models work as long as there is a relationship between, the data person and the person who's consuming data.

Richie Cotton: Okay. All right. The data engineering, the governance side of things is better centralized 'cause I guess that makes sense. You don't want anybody- I just 

Veronika Durgin: feel like it just works better. It just... And again that only works when you have ownership. You can't split ownership, right? You can't be like, "Oh my partner owns a kitchen and I own a refrigerator," and then it's how do you make dinner?

I don't 

Richie Cotton: know. It is get- gets a little tricky when you're dividing things up like that. 

Veronika Durgin: So that's why I was like, no, somebody has to h- like holistically be in charge of the kitchen and what's in the kitchen and what's for dinner, 'cause then there's kinda "Oh, okay, I bought chicken, therefore I'm gonna make chicken."

Then it's not "Oh, I'm gonna make steak," but there is no steak in the f- fridge. 

Richie Cotton: Absolutely. I love that analogy. If the data team's changing that way has it changed the profile of people you hire for data roles? 

Veronika Durgin: Again, philosophically, no. I always looked for... Anytime I interview, right?

So the questions I ask, like I, I don't actually believe in like standardized tests, though sometimes you kinda if you're looking for specific... If you're looking for an expert in Python, you need to kinda get a sense of how dep- how deep they are. But in general, my questions were always more of like I wanna see how people think.

I literally just want to see the logic, the questions that come to your mind, how you try to get answers to those questions. So same pro- like critical thinking, asking questions, not making assumptions, like a pretty standard pattern. So always kinda looked for those types of people. And the other thing that I always ask, and now if anybody, ever interviews with me, they know what to prepare for, how do you stay current?

How do you learn? Always a very- It's a tricky one. It is. I got some very interesting answers. For example, one person said, "Oh, I just wait until I have a task assigned at work, and then I maybe explore." And I'm like, "How do you know the best way to solve this task if you don't learn outside of the immediate need?"

So that was, like, a red flag. Essentially, this person isn't actually trying to become better at their job. They're just kinda working on things as they pop up. And I'm like, especially now in the world we live in, if you don't stay current with what's happening, h- how will you possibly kinda, I don't know, solve new problems in new ways, right?

New, new better ways. How would you propose to modernize, improve, our customer experience? Kinda like our end user experience. So that's a tricky question, which it shouldn't be tricky, right? 

Richie Cotton: No, actually I've got a good answer for this one, 'cause any time I want to know stuff, I just find a world expert, and then I bring them on the show and chat to them for five minutes.

So- 

Veronika Durgin: Exa- this is what I do. I follow people, and then they recommend people, so I follow those people, and that's how I consume content, and then you build relationships. And now, like Richie, like your podcast I'll be listening to that, and yeah. 

Richie Cotton: Yeah. Podcasting is it's a life hack for learning about stuff.

But I do love that's the approach you take when you're hiring, and I guess for anyone who wants to apply to a job at Slack, then a little tip for you if you wanna go in the data team. So yeah of the... It's like having some curiosity and figuring out like- ... okay, how do I learn about stuff?

That's a good way of finding new employees. Okay. Since you mentioned, like, how do you keep up with learning, it's a tricky problem, Ram, because there is so much going on. What tips do you have for learning that work? 'Cause I think there's often competing deadlines between I've got a deadline for my project and I also need to keep my skills up to date.

How do you approach this with your teams? 

Veronika Durgin: So I... Another sort of thing that I... By the way, this is personality. I c- I can't help it. I'm just like, naturally there something piques my curiosity. Back years and years ago in my career, I always had to have a happy place project. So I was like, I can't, I cannot have hours of just mundane grinding of doing work.

I need to have a project that's just creative. So I called it happy place project. And I would go to my boss and I'd be like, "Hey, I was thinking I want to try blah, blah. Can I spend just a little bit of my time on that?" They're like, "Sure." Nobody will ever say no to that. If you're like, "Can I spend four hours on a Friday?"

Just exploring this new thing, I wanna try it, here's the reason for that. And if it works, we'll solve a problem, right? It wasn't random, "Oh, I'm a DBA, I wanna, do some carpentry on the Friday afternoon." It wasn't like that. It was more of related to the work I was doing, but it's something different that we aren't doing.

Nobody ever said no. So right now what I'm trying to kinda encourage my teams to do is to have a little bit of their capacity dedicated to that, to creativity, to learning, to whatever they want. I will tell you that not everyone, almost actually no one does it. So people seem to just kinda they go in, they do their work, and they leave.

So maybe they, their creativity is kinda left for outside of work. But I always encourage my team if you want, if you're interested to do something like that, just let me know. We will make sure that you have time and space to do it. 

Richie Cotton: I do a lot of that, and it reminds me in the early days of Google, they popularized this idea of % time, where you spend % of your i- Ooh

Your week just doing something new and, Yeah ... I guess solving moonshot ideas. 

Veronika Durgin: I- it's hard. I think as hu- I think it's humanity we give in into pressure, right? So if there's something urgent and someone says it's urgent, you kinda literally just hyper-focus on delivering that as fast as possible.

You're actually making a lot of bad decision along the way, taking shortcuts, creating tech debt, right? So it's hard to push against urgency, so it takes a little bit of almost self-control to be like, "No. No. This will take a little longer, and that's okay because I actually need to... my Friday afternoons are mine," right?

For... Because ultimately it's a long-term investment, right? As you're learning, you're actually bringing this back to your team. So for me it's it's an investment in what we're building, right? Because whatever you learn, whatever you figure out how to solve some problem in a new, better way, you'll bring it back.

So it's, everybody's gonna benefit from it, so I'm like, I'm, like, trying to invest long term into the team. 

Richie Cotton: Yeah. I love it. It is always tricky to think long term when you got short-term- Yeah ... deadlines and- Yeah ... yeah. The sort of pressures don't go away. So I guess more generally particularly in, in larger organizations, like, how do you keep that agile nature?

'Cause, I think we talk about small companies, startups can often move a bit faster than enterprises. Is there any way to keep things- Moving quickly at an enterprise level 

Veronika Durgin: I think so. I actually just recently wr- wrote a blog about agile ruffled some feathers. As whenever you talk about agility, somehow somebody gets upset.

I... To me, agility... Okay I apologize to the agile police ahead of time. It's a mindset. I think agility to me is doing work in small increments and iterating on it. So kinda try small piece, roll it out, understand whether it's working or not, come back and iterate. So it just a mindset.

And I know there is a manifesto and there's all these other things and Scrum and Kanban and whatever. But agility is a mindset. I think, and again, like to me it's just really focusing on small chunks, delivering often. What I, again, personally love about that, because if you make a small change, it's easy to roll it back if it breaks.

The blast radius is small. I love that. Can you imagine making a bunch of changes and pushing it all at once and something breaks and you don't know what broke? It's a what, Amazon calls it like two-way door. To me, like these little things are two-way doors, and I'm so comfortable.

I'm like, "Oh I'll see if it breaks. Okay." It's very quick to roll back or fix it or whatever. So I- I'm totally leaning into that for very kinda selfish reasons, and I'm also kinda trying to push teams. And I know as data people, we're great because we love to spend time in analysis and discoveries.

So agility is actually hard bec- because of just kinda like the profile of a data person, right? Data scientists, BI analysts, like we spend days on data and we find all these nuances and little use cases and all the tiny exceptions. This is the world we love. So it's very... it's actually very hard to kinda like how do I do it in like a small way?

How can I ignore the outlier right now? I don't need to... I do not need to resolve .% of exceptions right now. I just I don't. It's very hard. So I, I understand that and I appreciate that, but I still feel like we can absolutely continuously move forward, and moving in agile fashion also le- helps us pivot very fast if, business decide we need to change or the world around us changes.

Richie Cotton: No, that's really interesting, the idea that data teams are not by nature agile. I get, yes I've d- I think may- I've maybe fallen into this trap myself before, is like you wanna get all the numbers right before you start publishing your findings and things like that. And- 

Veronika Durgin: i'm like that too.

Somebody's "Oh, can you give me like a quick few?" And then I get like a list of things of oh, I need to validate this, I need to validate that, and there's this exception, that's... And it's no. I just need like directionally. Again, that's why we're good at our jobs. So it's not a knock on anyone, it just kinda acknowledging that it's a little bit contradicting.

Richie Cotton: Yeah, and it depends what you're doing. I- if you're doing some science, then maybe it's gotta be exactly right. If you're just trying to figure out like a, I don't know, I guess a simple business problem, then s- sometimes a bit more leeway for error. All you mentioned data modeling before, and I know this is your passion.

Talk me through what does everyone need to know about data modeling? 

Veronika Durgin: I think I will distill and simplify it. And again, I am sorry to all of you diehard data modelers out there. I think the simple conversation with anyone to understand the things, the business concepts, the things about the business and the relationships, and just creating that kinda like a business flow sort of diagram, is something we don't do.

If we nail that down, if we kinda I, I'm like, I strongly believe, again, conceptual data model, like that I... Boxes and lines, literally, and I know there are tools, and I know I'm oversimplifying it, but if you literally sit down with someone and draw a bunch of boxes which are things you're talking about and lines between them that describe relationships, you will discover so much about the business, and you will challenge the person on the other side of the table to actually question their assumptions.

Just that. Like to me, this is a part we barely ever do, and it's so simple. Because once you have that, then you can start kinda figuring out your metadata, you can start looking at data that you have and actually see how it maps into your business model. Oftentimes you'll see that it doesn't. I have... I had this like almost epiphany a couple of companies ago.

I worked for an agriculture technology company, and we were writing a software, I think it had something to do with transportation and grain elevators and all that, and we kinda look at the data that we were receiving, and we created a model, and then and it was kinda like weird. Then one of the data scientists, like he literally just called the farmer and spoke to him.

And what we realized, there was a missing link that our software didn't accommodate. Because we reverse engineered what was known instead of actually talking with the people who are in that. And I was like, I had such a moment, I was like, "Oh my God." And that data scientist, I was like, to this day I'm like, "Good for you for picking up that phone and chatting with the person who is actually living in that," right?

This is your business stakeholder. Yeah, we identify the gap in software basically. Have you guys talked to your, end consumers of, so to me this is kinda like data modeling, conceptual data modeling, where it fits. I think everything else can and will be automated at some point, but this is important part.

Richie Cotton: Yeah, especially picking up the phone and speaking to your users it's yeah. You can get robocalls, but I don't think it's quite the same as as performing an interview with a s- No ... with a farmer. Okay. Some important things there. So the idea of the data model like what you talked about was like basically sketching a graph of your business processes- Yeah

On paper. And this of course is behind one of the hot ideas at the moment of graph models to provide context for AI. So yeah, do you wanna talk me through like the relationship then between data models and what AI agents are doing? 

Veronika Durgin: So I, it's kinda so I believe, like knowledge graphs is kinda like area where I'm, I have a lot of curiosity about.

I haven't quite made the leap between I kinda conceptually understand what it is and how it physically applies. But I think again, when you supply very much deterministic thing to an LLM, hopefully it doesn't guess, right? When you have a clear definition and you have a clear relationship, it doesn't need to make things up, or it makes things up a little less.

So I think that's kinda where, how I'm looking at it. Does it have to be knowledge graph? No, I think whatever data model you have. Obviously there is, of course, a question of unstructured data, right? So how do you model unstructured data? You kinda still can, right? Like your document is describing some business concept, right?

Or it's describing a relationship about business concepts. So I think you can still do that, but it's a very interesting area. 

Richie Cotton: Okay. I feel like I put you on the spot with that question. 

Veronika Durgin: I was like, "Let me download it." Oh, and by the way, if anybody figured that out, let me know. I would love to talk to you.

Richie Cotton: Nice. Okay. Perhaps do you wanna talk us through what are the business implications of having better data models? 

Veronika Durgin: You don't... actually, like couple of things. Let me see. So what you don't do when you have just a solid data model, you stop fixing things that shouldn't have broken, right?

This is s- basic given, right? The other thing you stop doing is reconciling and explaining why is number for concept, I don't know, for some KPI A is different from number for the same KPI A coming from different system. Like the amount of these sort of things reduces drastically if everyone is using the same, underlying data within your model that kind of derives that KPI.

It's just fascinating. The other thing is, the things that you start doing because you don't have to spend time on fixing and explaining, right? It's everything else. I think, for us, having solid data model, again, we're like, we could start, looking at knowledge graphs.

We could put LLMs on top of data. Like you kinda, once your foundation is solid and I'm saying something that's probably pretty obvious. Once you have solid foundation, anything new that comes your way, you can just start using, right? You don't have to keep revisiting and rebuilding. 

Richie Cotton: Okay. So yeah.

It does seem to be a common problem where y- you ask different parts of the business, "What should this number be?" And you're getting different answers, and so there's this problem reconciling different, ... business metrics, and you don't know how well you're doing. So one of the proposed solutions for this has been around having a semantic layer, which is quite closely related to the idea of data modeling.

Do you wanna talk me through what these things are? 

Veronika Durgin: I can tell you what I hope semantics layers will become. Okay. This is my wish list. I'm hope this startup somewhere doing that. So to me, and I think actually Juan Sequeda had a great presentation on that. To me, semantics layer is decoupled.

It's like universal, it's decoupled from tools. It's its own thing, right? And it does kind of translation between how business talks about things to the physical representation of data. And I'm trying to be intentionally ambiguous because physical representation of data shouldn't matter.

Shouldn't matter whether it's in a relational database, in a document, or a knowledge graph. But there's this layer that does the translation of the kind of like the words to the actual physical implementation, and I hope really someday we'll get there. We're not quite there yet. I think every, multiple vendors right now are building their own- semantics layer, but then semantics layers don't talk to each other, so it becomes a little bit complicated.

Richie Cotton: Okay. So it sounds this technology's kind of a work in progress, but the end goal is gonna be a much simpler way for business people to- I hope. I'm trying to manifest 

Veronika Durgin: it. I am trying to manifest it 'cause I talk to anyone who's willing to listen. I'm like, "We need a separate, standalone, independent, universal semantics layer tool."

Richie Cotton: Okay. Yeah, so you don't want it necessarily tied to one part of your data source 'cause if you're an ambitious- Because I want for all 

Veronika Durgin: data sources, right? Yeah. Because it's like why I'm kinda thinking that way because I don't n- want to necessarily move data around all the time either. So it's maybe arguable that's what GraphQL is right?

It's kinda like similar conceptual to say, like similar saying of I'll give you an analogy that I use. So I want for semantics layer to be like you go to a window of a food truck and you're like, "I want a veggie burger with lettuce and tomato." I don't want you to be like, "Oh wait, but you have to go to the," I don't know I don't know what veggie burgers are made out of, but go to whatever factory that makes veggie burgers, then you have to go to a farm to get tomato.

No, you just... I want you to get a veggie burger, but behind the scene it routes to the right places to gather ingredients to put it together. So that's why I'm like, I... But it, it doesn't need to be in the same food truck either, like all those ingredients, because we don't wanna move data around because it takes time, right?

It takes time. We create multiple copies. It's dangerous to have multiple copies. It's high risk, if something's happening. So will we arrive to a world where we have this layer that knows to go to different places to get the right data and also knows how to translate? 

Richie Cotton: All right. It sounds like it's either a work in progress or as soon as this goes live, you're gonna get a lot of calls from account executives at different semantic- I know.

... Web companies. Wishful thinking. I'm like, "

Veronika Durgin: Hey, all of you VCs out there, is anyone working on that?" 

Richie Cotton: Nice. All right before we finish, I think I'd be remiss not to ask you about agentic commerce, which seems to be the next sort of big thing that's being hyped. What's your take on this? 

Veronika Durgin: I don't know.

So as a customer, as a stereotypical woman, my shopping is emotional. So I am not going to... It's not, it's almost never practical. So I honestly I could maybe- lean on a chatbot to, for search. But otherwise I need to be part of that experience of looking for things, right? So I don't know. I don't see myself as a customer of setting up a bunch of agents and giving them go buy me, I don't know, a pair of shoes for X number of dollars.

No, I actually need to be, like, searching for it. It's the hunt and the kinda like emotional sort of, I feel today like I need retail therapy. So th- there's kinda part to that. So like I don't know. So I don't know what the future's i'm sure there's a subset of customers that will be leaning on agentic commerce, but I don't know who that persona is.

Something-

Richie Cotton: yeah, there's definitely products where I would like, I wouldn't necessarily want to outsource buying that to a bot. I like my shirts. I'm not gonna like- I- I'm not gonna completely outsource that to a bot. It would be easy ... I would like ways to make it easier to find the products I want to buy though.

I think there's probably some use cases for AI there. 

Veronika Durgin: Yeah, I agree. It's it's basically improving search, and maybe like my dream world of oh, I have, I don't know, a pair of beautiful pink pants with purple flowers on them. What's in style now that goes with it? So I actually wanted to go and follow a bunch of fashion people on Instagram, and like that part I want easier.

Like what's in right now, but I still wanna do the shopping. It's give me options, give me images, give me something to like kinda help me be creative, but not necessarily buy for me. 

Richie Cotton: Okay. Actually, that sounds like a really fun build-an-agent kind of project where it's like you find all the Instagram influencers or TikTok influencers or whatever who are showing off fashions that you like, and then you somehow collate this information using the bot and get it to give you recommendations of things to buy.

That sounds fun. 

Veronika Durgin: So I think it's an opportunity like what's in your closet? It's oh, you know what's in now, like maybe. Yeah. 

Richie Cotton: But then it's again 

Veronika Durgin: like how much, how many tokens you'll burn just doing that for fun, and then you'll only really only buy total cost of ownership of the solution, is it worth it?

Richie Cotton: Yeah, if you're buying very high-end fashion, then maybe a few tokens is, here or there. But for everyone else yeah, it's maybe not a money saver. Actually, I'm wondering, w- we talked about the fashion use cases in there. It's you are emotionally invested sometimes in, in what you're buying.

Are there things where you think, you hate shopping for this and you want to outsource it. I'm thinking like no one really enjoys shopping for insurance, that kind of thing. Yeah ... are the, do you think agentic commerce is gonna come for different kinds of shopping? 

Veronika Durgin: I don't know. I think, I hate food shopping, like at the top five hated activities for me.

But I don't see outsourcing that because there's kinda quality and all that. For insurance, interesting, but I think even for insurance you need just a better search and comparison capability. Like I... Insurances are like across the board, like whether it's, car or home or health, they're damn expensive to not have a say in and truly understand what you're buying, right?

Like maybe I also kinda don't trust robots quite yet at that level. Maybe that's part of it.

Richie Cotton: Yeah, definitely I think not trusting robots is a good idea at least for the foreseeable future. But, Yeah ... yeah maybe discovery and finding good prices for things. Yeah. Like search is still an unsolved problem for- for shopping in a lot of cases. Okay. All just to finish I always want more people to learn from who are you most excited about at the moment? 

Veronika Durgin: So I kinda already mentioned my kinda, it's been a while and I'm still, I am more of a explorer area of knowledge graphs and ontologies.

It's a very interesting area. I think for me it translates really well between kinda like data modeling because it is modeling as well, right? Maybe slightly different. Jessica Talisman is, she has a blog. Incredible. Like the depth is just incredible. So she's been kinda like my favorite person to learn from lately.

And there's a couple of other people, Juan Sequeda, he's right now with ServiceNow. He's been talking about knowledge graphs forever. He thinks in knowledge graphs. Like he lives knowledge graphs. So if you kinda wanna get like real depth there truly appreciate kinda like the content he is putting there.

He has a podcast as well, Catalog and Cocktails which is fun. You get to learn and drink at the same time. Guess who doesn't want that? Yeah, I think that's it. I can't think of kinda like- Okay ... yeah. 

Richie Cotton: So yeah some great ideas for learning about data modeling, about ontologies. I do like the idea of cocktails while you're podcasting.

That seems quite fun. I would like to introduce that to Dataframe, but I'm not sure whether I'll get away with that. We shall see. Oh, 

Veronika Durgin: you don't have to officially do it. We can just, 

Nobody knows what's in your cup. 

Richie Cotton: Could be anything in here, 

Veronika Durgin: yes. I know. It's clear, you never know.

Richie Cotton: Nice. All right. Thank you so much for your time, Veronica. 

Veronika Durgin: Thank you so much for having me, Richie.