Tim Gasper [00:00:17.314] Hello, everyone. Welcome. It's time once again for cataloging cocktails. Oh my God, it's season eight. It finally is here. We took a million months off for summer vacation. Just kidding. But hey, it's your honest, no BS, non-salesy conversation about enterprise data management with tasty beverages in our hands. I'm Tim Gasper, customer guy, product guy at data.world. I'm joined with Juan Sequeda. Hey, Juan.
Juan Sequeda [00:00:41.754] Hey, Tim, how are you doing? I'm Juan Sequeda, Principal Scientist here at data.world. And as always, we are back. It is a pleasure to come back live, middle of the week, towards the end of your day, to just have a break and go chat about data. This is the start of year five. I am super excited we're back. Just a quick parenthesis, go to data.world/podcasts, and you can actually find now a way how you can chat with all of our guests. We have all the transcripts there. You can do that. So all the chat with your data stuff, you can now chat with all our podcasts. And with that, let's kick it off. I am really, really excited about our kickoff guest today because our guest is somebody who I've been following on LinkedIn for a while. And she is somebody who is just really honest, no BS, sharing those raw learnings of what she's doing in the data product world and using AI, GPT, all that stuff. And I just feel so refreshing to have like that genuine content of just somebody that's completely open. And that's Anna Bergevin, who's a senior product manager at ResMed. Anna, how are you doing?
Anna Bergevin [00:01:43.754]: I'm good. I'm good. I'm here having a drink, talking about data. It's good.
Tim Gasper [00:01:47.274] Welcome to the show. So glad to have you. And thanks for kicking off this season with us.
Anna Bergevin [00:01:51.934] Happy to be here.
Juan Sequeda [00:01:55.834] So let's kick it off our tell and toast. What are we drinking? What are we toasting for?
What are we toasting for? Okay.
Anna Bergevin [00:01:57.860] Oh, I am drinking all-day Rosé Waterloo sparkling water, and I am toasting back to school because my children are back in school. And it's like the summer is always fun, but it is always logistically chaotic as a working parent. So yay for teachers. I'm glad to be back in school.
Juan Sequeda [00:02:16.920] Cheers to that for sure. I think I have two smaller kids that are not yet going to school, but they care about it. I can, yeah. Yeah.
Tim Gasper [00:02:27.040] Yeah. You have back to school to look forward to. I'll cheers to back to school. Also, my kids are just getting back to school. First day was just yesterday, which was tons of fun. And actually, for my cocktail, I'm actually going with a mocktail today. So my wife got me these curious elixirs mocktails, which they can come in the mail. They come in like a fancy box. This one is elixir number seven, which is the curious champagne cocktail. It's got elderflower chardonnay flavor jasmine green tea lemon peel and lavender and you know what I've tried a few of these curious elixirs they taste pretty good they don't pay me this is not sponsored it's just uh it's just tasty, so if you want a mocktail it's pretty solid. and uh I walked into my bar today I'm like gotta try something new something
Juan Sequeda [00:03:15.360] I walked into my bar today I'm like gotta try something new something I haven't had and by pure coincidence i also saw this a bottle of uh saint germain which is elderflower which is what Tim is having. We did not coordinate here, but I'm just keeping it kind of to my core and it's an elderflower old-fashioned.
Tim Gasper [00:03:27.860] There's a part of Juan's in my brain that are connected somehow. We don't know. Today it was elderflower. Cheers, Anna. Good to have you.
Juan Sequeda [00:03:38.020] Going back to school. All right. So we have our warm-up question. So what is the most interesting thing you've done with ChatGPT?
Anna Bergevin: [00:03:45.941] Oh, well, I am a big gardener and I live in Utah. It's very snowy in the winter and I very much get seasonal affective disorder. So I have my like happy lamp and I start gardening as soon as possible. Meaning I have a rack with like five levers, five levels, all these grow lights and I grow like a million seeds. But it's very complicated to figure out when you're going to plant them outside and then back up to how early you need to start them so i went to ChatGPT i give it a list of all the seeds i wanted to start and when my last frost date was and it made me a little plan it was amazing.
Tim Gasper [00:04:18.501] That's awesome, i love that, great use case.
Juan Sequeda [00:04:25.281] How about you Tim?
Tim Gasper [00:04:25.281] How about this I'll give you one personal one and one professional one so my personal one is my kids love ai generated stories. And so I've started to get more sophisticated with my ChatGPT generation, because the stories were getting a little too formulaic. So I needed to like kind of feed in some like, I need to tell it like make the story weird. And like, you know, like things like that. So that way, like, like, I tell it like, break up the structure of usual story structure and stuff like that. It actually does it like make some more creative. So anyway, that's my I could I could talk to you all day about using chat GPT to make kids stories. Work thing. I was spending a bunch of time this week trying to like, explain the difference between a sales play and a joint solution. Yeah, for those of you who know what that is, you know what I'm talking about. But those of you don't, you're like, what are you talking about? Well, it was kind of hard to define those things. And after like struggling with it for about two hours, I was like, you know what, this is something that chat GPT is probably really good at. And it was it had beautiful definitions for those things. So, you know, great use case for using chat GPT for work.
Juan Sequeda [00:05:31.541] I do the, so on the story side, my wife is kind of for fun, been writing some, some stories about chickens and pets. So we have a little bit of a dark humor stuff, but then we were like trying to go make creative or versions of it from with chat GPT too. So that was an interesting one. Actually, part of the thing we want to go do now is like, maybe we could have the chat GPT version of a book, all the content and the images versus the one that a human's done and then actually sell it. And then our metric here, what is better is what makes more sales. I don't know, but that was it. But anyways, we're going to dive into more chat GPT stuff on using, but let's kick it off. Honest, no BS. How the heck did he get into data product management?
Anna Bergevin [00:06:17.495] Product accidentally, as many product managers do. Yeah, so I was an analyst for a really long time. I was kind of at the top of the analyst career ladder. I had some product manager friends trying to decide should I become a data scientist, an analytics engineer, like do I become more technical or do I go the people leader route for analytics? Or should I just like jump career ladders completely over into product? And I just thought product was really interesting. I got I got asked to help out with one of our overseas engineering teams temporarily after Reorg was left without a PM. And I love the stand-up. I love working with the team. I loved tackling things as a group. And the thing we were working on was very data-heavy, which is part of why I was asked to help on it. And so I switched to product management. My first product was not data-centric. It was the comms platforms. We were doing email in-app notification and mobile push notification. communication. There's some experimentation, A-B testing metric kind of stuff around there, but data wasn't like core to the product. And I was really missing data. And so a friend recruited me to come be a growth analyst. So I went to go, not growth, I was growth PM, which is very data centric. I went there, then after a layoff, I kind of had to look for a job for the first time ever in my life, like in that way of like, hey, everyone, I'm job hunting. And a former friend who was a head of data said, hey, I need help on the data platform. I need you to bring product and meld it with data. And I landed and had no idea what that meant to bring those two things together and spent the last two years figuring it out. So that's my story.
Tim Gasper [00:07:59.335] Do you have it all figured out?
Anna Bergevin [00:07:59.335] No, I figure it out actively every day.
Juan Sequeda [00:08:06.755] People don't have this figured out. or you think they're so anybody who says they know what they're doing, it's BS.
Anna Bergevin [00:08:13.048] Yeah, I like, no, there are people who are further along than me. I think that no one's like got, there's no great books written on this. There's no like extensive learnings. There's lots of people like scattershot sharing things. But if you're like, like every product podcast you go to is talking about customer facing product and you may build a data product that is customer facing. It's hard even to find content on platforms, let alone building a data platform. So yeah, I've just been piecing it together. And that's why I started posting and writing more was to connect with all the other data PMs who every time I get on a coffee chat with another data PM, they're like, yeah, me too. I have no clue what we're doing. What are we doing?
Juan Sequeda [00:08:56.328] Well, so one thing I want to get into this on this side is from the timing perspective. And the reason why I wanted to kick off with this question is because it's something that I kind of, we've been hearing from folks kind of like asking, oh, we're getting into this product management this time seems like such a cool role but is it's new and then now you're starting to pop they're starting these people pop up around that stuff i really want to kind of like have that honest discussion about like how how you said it's you've been doing it for two years like how long have people actually been thinking about this stuff and what are the things that you would be that you would consider there is no book yet but if there was a book right now like if you could outline what are the things that are like already kind of kind of figured out now and what are the the ones that like, I talk to people and they see it this way and the other people, people see it this other way. Curious about your point, your perspective. Yeah, I think that there's always been,
Anna Bergevin: [00:09:45.948] Yeah, I think that there's always been, you know, chief data officers, data engineering leaders, people who had a broader vision for what it meant to like, get value out of data or to extract value from data. But what's happened in a lot of businesses historically, right, is they struggle to operationalize that You have PMs who are like, oh, we want to build an analytics or a data product, but they don't have a data background and they struggle to piece that together. Or we have...The data vision that these engineering leaders have, but then they end up just getting all these requests from the business. And it's hard to work on the strategic work because you're just being reactive and acting more like a service desk. So I think that what we are learning, those of us who are trying to figure out this space and what it means to be product principles is we don't just build whatever we're asked to build. This is strategic. It should go with the company strategy. We should be prioritizing according to what's needed. And some of that means building what's needed today. And some of that means building a platform that can deliver what we need a year or five years from now, right? So none of this happens overnight. So it's a mix of knowing your customer. The best thing about being a platform PM or an internal PM is all your customers are easy to find for customer research. The bad part is they can all DM you every day asking where their stuff is. So it's just like managing that, managing that energy of like, what do you need? Tell me why this is hard for you. Like finding the friction or the pain points in just getting the data, getting it together, especially in my industry, which is very regulated, doing all this in a compliant way that protects privacy and protects the business's legal risks. So a product manager...Can do all that a little better because it's their whole job, right? Is to talk to the customers, to prioritize, to strategically connect things, to say no to things, to protect their engineering leaders' times. I think companies that don't have data product managers, you don't have engineering managers who enjoy this and just who kind of are doing data product management, even if it's not their title. Or you get engineering managers who are just frustrated that they're being pulled into too many meetings. They have too many asks and it's not pleasant. So I think most companies should have data product managers, but they don't.
Tim Gasper [00:11:56.206] You get to be more stakeholder facing and really engage with where these data products need to go. So, you know, I think that's interesting that you sort of started in data. You got into product management. And then obviously you kind of now are really focused on the combination of those two things. Do you feel like as a data product manager, you're more product manager or more data person?
Anna Bergevin [00:12:22.886] Oh, that's a really good question. I would say I probably am more of a product person by process or approach. When I was an analyst, my bias was much more towards solutioning what's possible, right? A question comes up, oh, we have this data or we could do this. It's very like that brainstorming or that creativity around what could happen. And as a product manager, I've had to learn to sit back and listen more and try to probe more to understand the problem before we jump to a solution. So my approach has differed. But I would say a huge part of product management is having empathy for your user. Of course, talking to your users helps develop that empathy. But also, I was an analyst. So I was like, oh, yeah, remember when we had data in two different warehouses and I I had to bring it all in locally to my machine. And R was like, this is too much data and we can't store it. And it was a miserable nightmare. And then everything got on Snowflake and my life got easier. So I guess I keep my... My data person at my core, but like, I act more like a product person.
Tim Gasper[00:13:28.004] Now, that makes sense. And, and yeah, the more you have expertise and empathy in an area, as long as you be careful, the expertise can sometimes be tricky, because sometimes you want to follow your own intuition instead of follow what you're learning and what you're discovering from your stakeholders, right? But like those things can help a lot. And obviously, that data experience becomes invaluable. I like that you also brought up and when you kind of explained getting into product management, you know, data product management, that sort of there's a difference between customer facing data products and internal data products. And I think that's super true. And it even feels like there are certain companies that, you know, are more like they sell data, or they have data products and services. And so they're more like customer facing data companies versus, you know, companies that are looking to manage data. And, you know, of of course, they need to sell better and market better and, you know, be more efficient and things like that. But those data products ultimately are serving internal customers. And, you know, can you talk a little bit about how you see those two things being different? And it sounds like you've mostly been focused on like internal data product management.
Anna Bergevin [00:14:39.204] Mostly, yeah, there's there's been some early discovery that I can't probably get into detail about around stuff that might be customer facing. But I would say, I think, let's just say traditional SaaS product managers who want to bring a data product into the product or as an offering, right? There's a lot of value in data. Often your customers may pay extra for access to more raw data or exports or analytics is that it is not as simple as let's just pull the data in and throw some pie charts and bar charts up. That's not what it's going to be. And so I think that often people get an idea around monetizing data and they have no idea how much tech debt and governance and things have to happen before we could do that. And so I think that it's just external data. You have to think about what's going to be valuable, what people are going to pay for, all of that aspect. But you also have to appreciate that you even understand the current state of your company's data. Is it it even ready to do this? Because what's going to happen, you start doing a scatterplot and you have a bunch of outliers. It suddenly becomes much more obvious that there are data quality problems, right? And so... Um, yeah, I think, I think you think about it differently. I would say I am thinking there's overlap in that both myself and an external phasing product manager are thinking about the end user state or the experience of consuming this data. What are they going to do with this data? What format does it need to be? And what questions does this data model need to answer? Or what, what shape does it need to be in? So the BI layer or the application can consume this? What legal reviews all these risks that have to be managed, but the difference is internally, Internally, I'm also managing a platform which has capabilities around how we're going to handle GDPR requests, how we're going to do other things. And a product tends to be just the interface and the data itself. So they overlap, but at the end of the day, there's different amounts of time spent on different aspects is probably the way I would think about it.
Juan Sequeda [00:16:39.270] So one of the things that we were talking before is that this work, all this management work, you don't necessarily argue that there has to be a data product manager per se. That work is happening already or it should be happening. So I want to dive into what is that type of work and who is doing that work today and when do you realize, when do organizations realize, oh, it's time to actually go hire somebody specifically for this? Or maybe you don't ever have to.
Anna Bergevin [00:17:11.130] Yeah, I guess, I mean, you may have data engineers who love to sit with customers and be in more of like a consultant type role of like really understanding the problem and what do you need and what can we build. But that may not be the best use of your engineer's time, nor what they enjoy the most. They might enjoy receiving more fully formed requirements and just doing the final design and then the tweaking and handling of edge cases. Like they're going to engage with the users or the customers either way. But my role, a lot of my time is sitting in a meeting, trying to understand the business context, trying to understand the assumptions that the user has, trying to help the user understand what they actually want. I don't ever build exactly what they ask for when they come. They come in and they're like, I need this thing. I'm like, oh, but do you? Do you need a dashboard or do you just need an Excel that connects and you can do pivot tables off? Why are we building a dashboard when all you're going to do is build a table in Tableau and click CSV download? Like why so a lot of it is me pushing them to be like don't tell me exactly what you need tell me what business problem you're trying to solve what skill set you have who's engaging what's the workflow so anyone could do that a business analyst could do all that context gathering and bring it to the data engineers the difference is um if it's my job to own it end to end and my job is to make sure we are utilizing the data engineers time for the right things my engineers Engineers don't have to worry as much about probing business leaders about company strategy or telling them, well, I can do one of these two-bit projects. Which one do you care about? It's just about resourcing. So who you currently have on staff. Currently, there's only two product managers on ResNet's global data platform. And I'm spread very thin. So is my partner, Roger. And we have a group that really wants to move fast. And they said, oh, how can we partner? What do you want? And I'm like, you can go interview all these business leaders, figure out exactly what they actually want and bring me more drafted requirements that I just go validate, right? That's how you speed things up, not by building it parallel to the data platform. It's that we split up this job of discovery. So discovery has to happen. Prioritization has to happen. If it's a product manager, it's my whole job. So the question is, does your business have five different people each doing 5% of their job on this and a bunch of it doesn't ever happen? then that's commonly what I see when there's no PM, right? Everyone's doing a little bit of this, but it's no one's job to own it wholly. And some of it just never gets done.
Tim Gasper [00:19:37.985] Interesting. So it sounds like, you know, this work has to get done. And, you know, a data product manager creates a more efficient and a more focused person who's helping to really figure out what are the right things to build that are going to solve the pains that are going to provide the most value that are the most prioritized. And also that is the actual experience that is what somebody wants, right? And when you were going through that, I always think about the old adages about like, you know, a customer asked for a faster horse and what they really wanted was a car, right? Or the other flip side of that is customer asked for a unicorn and what they really wanted was a car, right?
Anna Bergevin [00:20:17.405] Yeah, and just like what's realistic with where the business is at. So like any business, I mean, you can talk to lots of data people across the business will be like, why can't we get finance to stop storing everything in spreadsheets right the spreadsheets are stressing everyone out and and so a lot of people are just like let's get them off the spreadsheet and i'm like but the spreadsheet is like really useful to them like there's a reason i use that spreadsheet i did a whole round of interviews um last quarter i think i talked to all these spreadsheet users i'm like tell me what you're doing with the spreadsheet and um some of it you know is just reporting some of it is like basically data pipelining like insane things in spreadsheets and i'm like so what if we we, you know, what piece can we build, right? Could we build data cubes that you connect to from Excel, you can connect to like Power BI data cubes, for example, and run pivots in a report. So when you need next month to get that report, you don't go back and do a CSV download in this whole experience, start with raw data and reproduce this report, you just change your filter on your right. So the question is, that user left spreadsheets. So I'm going to help them do spreadsheets better rather than just deciding I have a vision that they should never use spreadsheets. It's like being realistic with what users want, what they need, and how do we stepwise move forward rather than what am I going to do? Send everyone who's in an analyst-y type role and be like, you all need to learn Python. You all need to be writing scripts. You all can only use the BI tool I chose for you. That's going to be hard to do, expensive, slow, all those things. So it's also just, I don't know, being realistic with the evolution, having a vision for where where we're headed and not just being reactive.
Tim Gasper [00:21:53.125] Yeah. It's not just about taking, taking orders. It's also about trying to build out a roadmap of where things need to go based on what you're hearing.
Anna Bergevin [00:22:02.105] Yeah. People don't know how things could be better. Right. Like, I didn't think I needed a digital whiteboard. I kept hearing about these digital whiteboards. I'm like, I don't think I need a digital whiteboard. Like I got paper. I'm fine. And even then we're all using these a little bit during the pandemic. And then I started using Miro. I'm like, this is so awesome. Miro like makes my life so much better. Like, right. So people also don't know kind of how things could be. And you kind of have to demonstrate what if you did things this way. And if they don't want to move to it, you figure out why. And maybe build something different or maybe just they need more enablement.
Tim Gasper [00:22:36.095] No, you're solving the pain and the problem, not just not just asking for the solution, because, yeah, if you just ask for the solution, nobody's going to come up with digital whiteboard. But when you're like, hey, I'm having trouble collaborating when we're not in person together. And, you know, but it needs to be real time. It needs to be all these things. Now, all of a sudden, you know, the product starts to make sense. You know, it sounds like you're working with quite a few different personas as you're doing data product management, both from more of a consumer perspective, as well as like the people that are helping you to build out those data products. Products you know I think about traditional software product management where you tend to have UX designers and you have you know engineers and usually there's like an engineering manager and you know they're they're partnering up with the product manager to execute on that product what does that look like for data products you know is that similar or is it a little bit different?
Anna Bergevin [00:23:33.135] It's a little different at least how we're currently staffed is right I I don't have one single engineering manager partner. I have two directors. One's working on more of the ingestion. One's working more on the analytics side of things. And then they have their engineering managers report to them. And those managers work on slightly different tech stacks, slightly different systems they work with, et cetera. And my job is really, I'm working with the customer and they have to bring it together. So I'm not working just with a tightly coupled trio. I'm working with more like a small team of engineering managers who then go execute with their their individual engineers so that's a little different because we like as far as ceremonies and cadences and like how you practically work also ResNet is very very globally distributed like I have an engineering manager in KL in Malaysia I have one in Sydney we have people in India we have people in Halifax we have people in San Diego so like just on a practical level time zone wise it's tricky. Um, but I think that it's similar when I was a traditional product manager, it's similar in that ideas come from everywhere. Um, ideas might come from your designer. Who's like, Hey, I was fiddling around in Figma. I don't have a designer right now. That would be pretty cool if I got a designer, but we don't have designers allocated. Um, but I still work a lot with my engineering partners who are like. The ideas come up all over, right? My engineering director says, we need a semantic layer. I'm like, what do you mean by semantic layer? Because no one on the internet agrees what that means. And he's like, well, I think we need this caching layer and this BI agnostic layer. And we're talking about it. And I'm like, okay. And he might send me some tech and we're talking about data contracts. We might send things back and forth. So it's organic, it's fluid. And the ideas come from anywhere. We brainstorm those things. We roadmap those things. I build requirements. um they do the technical design i guess what's a little different is um just the scale right usually when i was in b2b sass even right we're building for like one individual user on the platform sometimes we're building for one individual user sometimes we're building, complex systems and pipelines for like data quality monitoring other things it's just a little, it benefits the end user but it's not all about an end user interface where someone's clicking and pointing, which is often B2B SaaS at the end of the day. It's about creating digital experience. And I'm creating a lot of invisible experiences.
Tim Gasper [00:26:02.741] Yeah. Yeah. No, that's interesting. A, maybe we do need data designers. That sounds like a cool role. So something for us to explore as an industry.
Anna Bergevin [00:26:14.361] We definitely need more design training for analyst building dashboards. I can tell you that. I'm just like, I mean, you go to Tableau conference you watch those like iron viz guys and they build the most beautiful thing in like 30 minutes and that is not the analyst most of us are working with no no shade on the analyst i was not that analyst either when i was an analyst.
Tim Gasper [00:26:34.161] Most of the analysts watching that are like that is beautiful how did you do that.
Anna Bergevin [00:26:39.121] I will never be this good at tableau, but i think yeah there's someone has to think about the end experience and and product isn't just about i mean design's not just about interfaces it's also just about user experience the mental model and i would say if If you don't have a designer, the product manager has to manage that usability risk through interviews and feedback anyway.
Tim Gasper [00:26:58.341] Yeah. Well, and it sounds like there's quite a few, even though the people are a little different, you don't necessarily have that trio of engineering design and product. There are a lot of similarities, you know, from a, you know, ideas can come from anywhere. You got to collaborate with your engineers. You got to collaborate with your, your key stakeholders. Do you feel like the process is, is similar or different? And, you know, just to tie in like a personal experience. So when I first came to data.world, actually, one of the first teams I managed was our data team. And, and one thing that I found that was difficult. So that was my first team time directly managing a data team, I had usually done more like working more in a product capacity with software products. Um, I found that was really hard to adhere to agile kind of deployment methods. Like it was a little hard to be like, oh, we're going to do a two week cadence. And you know, we're going to launch things every two weeks, because the work was pretty unpredictable. I don't know if you see something similar, or you kind of figured that out.
Anna Bergevin [00:27:57.215] Oh, man. We are always... I mean, I would say ResMed's business has been around for 35 years. So any business that's been around that long, there's going to be spots of the business where those teams are cutting edge and they're pushing to production every couple days and things are like... And then there's these poor people who have this super old monolith code base and they're like, look, it's not our fault. We can't push to prod every couple days. So it's very variable within ResMed. And we're definitely trying to move in the direction of more agility. But at some point what some of these data projects are so massive and so many elements have to come in together when i part of what made us move fast in my first product team is every team had a distinct bounded context and it was very decoupled right so you had to publish data streams out for people to take a dependency on them it was just like individual code bases uh data is like one internet connected web of code bases. And you can try to componentize and there's things you can try to do. But at the end of the day, the bigger scale the project is, the more coordination there is. So there's some aspect of it that's always a little bit waterfall, especially with things like migrations. And planning wise, we plan in quarters. And I would say we try to leave agile bandwidth within there. We're always handling certain emergency ad hoc. Things are always coming into the data team. Team but i also just try to protect my team and when requests come in i'm like i'm sorry that sounds really interesting and important but um we're midstream like we can't just constantly shift our priorities some things just take you know six months and um we also don't ever just have one work stream every team always has multiple things going on but yeah it can't be perfectly agile and it's unpredictable you also just don't know how bad it is till you get in that's how bad it is that's very negative sounding but like if we all start from an assumption that the day will clean and interoperable and everything's going to be great. And then you get in there and fight every edge case, every weird historical bug. And it just takes longer.
Tim Gasper [00:29:54.715] Yeah, I think that's such a good comment. And, and that's okay. Hopefully, that's not too scary to the audience here. Like being somebody who's been in product for a very long time. It's like, of course, you know, you know, dragons are everywhere. And you run into them, you find them right, things are harder than you expect to rarely does a project you go in, you're like, Oh, my gosh, this is, I thought it was going to take a week, but it only took me an hour, right that's that's usually not how it goes so um but you know that this is just part of the process and the more that your team kind of learns that you know the situation that you have you learn how to work together you try to be agile the better it gets.
Anna Bergevin [00:30:31.874] Yeah, I think I think it's different about data is we don't control the inputs into our product, right? Like the inputs are coming from all over the business. We're usually a secondary use case, not the primary use case. And so we are making do now there's a lot of work going on to try to help connect those primary stakeholders to us to be like, hey, there's a machine learning model built off your data, you should probably be aware of that. And how do you get that context to persist over time and making for stable, but I think that's part of what's different about data is you're working with components and pieces that you don't control end-to-end, right? You receive them and then have to make them work. And in software, at least the teams I've worked with, you're more building in experience, right? You're engaging with the user, you're capturing data, you're reflecting things back. At least in good software organizations, things are as decoupled as they can be. You just can't do that really in data.
Tim Gasper [00:31:22.374] Yeah, that makes a lot of sense.
Juan Sequeda [00:31:27.234] That comment is something that I think people don't realize when there's so many people talking about just data and stuff. So I think you always hear people like, oh, this is what they do at the big tech, at the fangs and all that stuff. And it's like, they're doing all these things. I'm like, yeah, and those people control the product and the data all together. While you are working in an organization where you just get that data and you have zero say. I mean, you're getting data from Salesforce and from whatever it is, right? Someone's like, you cannot change that. And I think those are two different worlds, which one of my pet peeves is when people make these statements out and they make it in a way that is applicable in a generic manner. And then other people just eat it up and like, yeah, that's how we should do it. It's like, but that doesn't apply to you. And by the way, you're not a Google. You're not a Netflix. You will never have those types of problems. So I think, I mean, anyways, I'm kind of ranting here.
Tim Gasper [00:32:23.234] You're making a good point. And I think, Anna, what you're talking about, I think is the real, it's the raw truth, especially of like enterprise organizations, right? Because like, we've had a few folks on the show who sometimes have talked about either directly or indirectly data product management, but like they control all the inputs, right? It's like a software company where like they control all the instrumentation and they can manage all the data contracts all the way through the chain. And it's like, well. That's nice. You know, that's nice that you could do that. But most companies don't have that level of control.
Anna Bergevin [00:32:54.644] No. And if you want to work, sometimes I think I chose the hardest possible use case to start data product management, which is healthcare in over 140 countries. It's cool. It's not complicated at all. um but at the end of the day that um the problems that are most interesting to solve are most cross-functional most complex like like but i meet people every day who are like oh you work at ResMed i have a c-pap i love my c-pap it changed my life and i'm like and they mean that because they were not sleeping like it is actually life-changing to start sleeping and so um but yeah you have to be realistic that if you're going to join one of these enterprise companies in healthcare and you want to change the game of healthcare and make it more connected, more data-driven, all these cool things, and you want to bring that, what we see in digital SaaS and bring that to the enterprise, it's going to be painful. It's going to be slow. You're going to face a lot of problems, but it's really important. And that's really, for me, very satisfying. Not everyone belongs in the enterprise. My husband is a founder in a startup. He would lose his mind in corporate America. He hates meetings. He is just like founder all the way. but i think we don't hear enough i think um online from people work at enterprise companies and so it sounds like oh all the cool stuff's happening in silicon valley i'm like no there's cool stuff happening everywhere.
Juan Sequeda[00:34:19.344] I'm so glad you're saying this and i want people to really realize this like this is also i want to like on the podcast give the voice to folks like you who are like you may not you know yeah you may not be working but in the most shiniest kind of visible name that everybody else was but one you're doing really impactful work that's impacting people all over the world everybody everybody's lives around that and it's freaking hard and you can actually have one could argue that it's harder than what what these other companies make it saying because it's just completely different perspective you don't even listen you don't even know about and obviously the the people coming from the big tech in silicon valley like they they're the ones who have a louder microphone and stuff so honestly like Like when I hear people coming in from the fangs and like, well, this is how you do data engineering work. And then I come like, yeah, I get it. But come on, you got your corner. There's another part of the world here, too, which is probably bigger than you. And we also need that change. We need to know how to bridge that. I'm not saying it's incorrect. Obviously, there are great things there, and they're the ones who are driving and pioneering. We just need to understand how to go translate that to the rest of the world.
Anna Bergevin [00:35:25.464] Yes, exactly. Exactly.
Juan Sequeda [00:35:28.564] So you have, throughout this conversation, Tim and I are kind of back channeling here, it's like, you've seen a lot. And what I'm really curious is to know what's missing right now in the data industry today. From a, from a, from a people process technology perspective and the flip side, it's like, what do we have too much of? And can we just like, please calm down. That's it. Like, we don't need more of that.
Anna Bergevin [00:35:55.781] Yeah um i will say the um trickiest problem and i'm not seeing. Like uh the trickiest problem we're working on right now which is like we know we need our data to go from less structured to more structured in order to power certain things where we're going with uh like people are like oh i think in three years you won't need to know sql because computers will write it for you i'm like how do computers know anything about these systems it's not written down like it's like how do you teach a computer like oh we have three different erp systems one in each of these regions and this is what belongs in each of them like so a real problem i'm facing right now is how to get that context out of the heads of people who've been working for 17 years in a role and they don't even know what other people don't know because to them it's all intuitive. How do I capture that? So that's how I started playing with chat GPT was building these custom GPTs to try to help my SMEs like brainstorm their metric names and their business vocabulary. So we could start like documenting in more structured ways. And what I learned is like, I really need like a multi agent system, like a team of taxonomical documentation experts who organically interface with the non-trained in these ways, SME. That doesn't exist. I don't know anyone who's working on that. We all know we need to get to this target state. And data.world, for example, which is our current catalog, I can have people go into a UI and document it. But it's very hard to get someone to be like, can you go document these 600 fields that you are a steward of? And they're like, yeah, eventually I'll get to that. And they never get to that so like, Is there a chat experience or a different interface or a way you could harvest through a corpus of all their documentation and their OneDrive and all these things and say, hey, these seem to be terms that matter. Help us make sense of them. That doesn't exist. There's lots of things that don't exist, but that is a current top of mind pain point for me.
Juan Sequeda [00:38:01.634] Thank you for being very open around this stuff. I think it's the stuff that I'm realizing right now that the whole GenAI and the LLM wave is getting the focus more on structure, on context, on knowledge, on semantics, which is, well, I'm thrilled because that's been my entire life. And I'm like, now people are paying attention to that. that but i feel that we're like in this in this hype moment and you're like people are saying oh yeah systems will write sql for you don't learn it so then we're like oh yeah i'm just gonna wait for it to happen but then no no there's work we need to go do so and if we don't do that work then in like one or two years from now we're like we're waiting for this stuff oh that lai stuff that never worked and the pendulum swings right so i think it's it's it's this moment where we have like folks like you who are really pushing for this and this is why i'm really excited and And I mean, confirmation bias, I'll acknowledge that. I'm like, I'm excited to meet people like you are like, like, like, that's what I would say too. But, but yeah, I'm excited.
Anna Bergevin [00:38:59.134] Yeah. So I mean, if we're wrong about this hypothesis, worst case scenario, I have a better documented data set for my analysts to refer to. Like, like what's, what's the downside? I guess the opportunity cost of what people could work on instead of documenting that. But I can tell you in a business as big as ResMed, it's 10,000 employees and growing and globally distributed. We need, I mean, when I came, everything I would ask someone, they're like, oh, go hop on a call with Mary, go hop on a call with Bob. I'm like, I can't just like sit on calls back to back. It has to be async ways for me to answer questions or to get context. But if documentation isn't part of people's natural workflow, it's the task that always is going to go to the bottom of the list.
Juan Sequeda [00:39:37.894] Yeah, so we just need to make documentation work.
Anna Bergevin [00:39:41.921] Intuitive, easier, fourth. Like we're talking about things like, do we have our software engineering teams push data dictionary as part of a schema registry and they're not allowed to push changes without that, right? That might be controversial to the teams who don't want extra checks in the CICD workflow. But if we're like, would we rather you push it to us as a YAML file or would you rather have to go into a UI and enter it all manually? And then questions of like, what's the source of truth? Probably the YAML file. Or is it the YAML file that then writes tags against Snowflake? Like, I don't know.
Juan Sequeda [00:40:15.381] But that goes into the whole, let's push left. And I think that's a good, I mean, what are our incentives? I mean, at some point, like that is the context. If you like, we're talking about data lineage and all this stuff, but you want like the knowledge lineage and like, where does that go back? It goes back to the person who actually wrote that piece of software and that team that come in with the requirements. Like, that's the ultimate source.
Tim Gasper [00:40:38.181] Who created that column in the first place? That person probably should share what the definition is of it.
Anna Bergevin [00:40:41.921] Yeah, I was interested. We've been talking about lineage a lot. People often talk about it in a backward-looking way, like this asset I have, where did it come from? But something else we've been thinking about a lot recently is this thing I have, if I change it, who's going to be impacted? The downstream side of lineage is something I'm thinking a lot more about. It's really tricky, especially when things hit the BI layer because the lineage breaks or it becomes a service leader. And then you have to say, oh, and then that asset, who's viewing it, who's consuming it. And so, yeah, that traceability of we need to replace this thing. Who are all the stakeholders who have to pivot? How much can we stop the breaking change? And how much do I have to change manage people to change new assets? So, yeah, that's another problem of change management over time and mitigating impacts impacts. There's some innovation happening in the data contract space. I think it'll be interesting to see where that goes, the open standards, the private products. But what's clear is we need producers to be aware of secondary use cases of their data. They're just thinking about it in the application context. And we need that context to be there for them at the time they're planning a change and going to make a change and that it's a contextual alert, not just this will break a dashboard. Well, is that Todd in supply chain's dashboard he built three years ago and it's still refreshing, but no one looks at it? Or is that the CEO's mobile dash where he checks every morning? Those are different because we can't necessarily stop all development because it might break something somewhere. Well, we probably need to stop it from breaking the most important things.
Tim Gasper [00:42:15.220] No, that's a great, great thought process there. You know, I was just seeing a comment coming here from Raja, how to improve data quality in an organization as you're thinking about like as a data producer or as a data product manager, you know, by the way, I think it's a sign of maturity of your organization that you're not just reacting. It's not just the people building reports who are looking back at like root cause analysis, right? You're being proactive and thinking about as producers and as curators and as data product managers, how you're affecting people downstream. How do you think about quality and governance in the context of the work that you do for data product management?
Anna Bergevin [00:42:48.580] Uh i mean we talk about it all the time um there's quality for like there's so many different layers of quality um there's just like for example one very strict is like when we bring in a source system to the warehouse did we get the whole thing right did we do a good job like is this an actual good replica that's one dimension of quality of course we have services to manage and do things things around that then there's like quality in the um is in the eye of the beholder like it might be fine when they set up the erp that they let that be a free text field but now it's not fine um like someone's like no we need to govern it um for example um a data value came in it was from a third-party data set and it was for bmi body mass index and it was like a four digit value which is not possible in bmi and so someone's like whoa how is this happening i'm like Like, I don't know, we can go investigate, but. You can't put data quality tests on everything. You can't put alerts on everything. It's going to kill your compute costs and no one's going to look at it. It's just like a waste of time. So I think quality is always in the context of like, okay, this data point trigger kicks off an automation over here. So it has to be valid. Well, what does valid mean? Valid means it can only be one of these two values. It can only be within this range. It can never be null. So we want, what does it mean to quality test that? Someone needs to get alerted if something's out of range so they can fix that upstream so they can handle the edge case so you can clean up historical data. There's also a difference between we have some ungoverned history here and some messy data. Do we go rewrite history and clean it up as if it never happened? Or are we just stopping the bleeding from here on out? And then once we clean up a set, I might do a static quality analysis and send that to the IT team that's over that Salesforce instance or that ERP and say, here's all the data quality problems we're seeing. You can go put validation in here. You can go make this field required. Like you can prevent some of these issues and how much of it is just alerting. So I guess the question is, why do you care about quality? For what use case? What's the ideal outcome? Who needs to get alerted? Who needs to be aware? Who needs to remediate it? Who's going to be incentivized to remediate? I'm not going to put a bunch of quality tests in place and report on it. And then no one's incentivized to improve the quality because then we've just spent a bunch of money for no reason. So to me, it's figuring out that whole... When people say we have data quality problems, that's where product management comes in and that muscle is saying, tell me more about that. What do you mean? How are you experiencing that?
Juan Sequeda [00:45:16.253] I am going to give you a huge round of applause for this. This is a very important segment for when people say, we need to improve our data quality. I'm like, okay, let's open up that can of worms and listen to this particular segment that you just said. That was super spot on in how you went through all these things. Because I think that's it. People think about it as, oh, it's a one thing. we need to have high quality whatever just give me a green or whatever numbers i'm like no no this is so much and you very eloquently described it so thank you so much that was great, And I think this is a quote. Data quality. I like this part. Do we go back and rewrite history as if it never happened?
Anna Bergevin [00:45:54.544] I mean, there's some point where it's like, let's say, for example, there are dates that I found a date field. We'll just say this isn't very inside baseball. I found a date field in one of our databases and the date was 1910. And I was like, hey, ResMed didn't even exist that long ago. Like, this is not possible. What's going on? And of course, there was a whole story in Edgecase. So then the question is, if my architecture is that my database should replicate source, that's what it says in the source, but it's not true. That's a data quality problem. But do I make the software engineers go fix it? But can they overwrite it? Do I impute it as null? There's a whole bunch of decisions to be made. And that depends on who's the user and why do they care?
Juan Sequeda [00:46:32.644] I did the same thing like data. I had a column called years. So it's a string. So because they only have years in there. Okay, there's a string. So then you're doing an aggregate over that. and then suddenly you have some null values or you have strings whatever and then it like you're trying to get the average age of something well then this is completely off right or then there are some other numbers which is like 1800 something well that was just somebody who put in that means i don't know whatever right so yeah yeah all the qualities but anyways before we go into the lightning round one more thing because we started off with our world question about GenAI and GPT everything and you there's a lot of stuff that you've been posting on But just quickly, how are you using ChatGPT on these Gen AI tools to help you in your job, in your data product managing job?
Anna Bergevin [00:47:17.684] Yeah, I mean, I use it for everything. Like every day. I only started using it six months ago. ResMed had to do their due diligence. I used it a little bit before then, but I'd been very careful about every input that it was only public information. So it was very fragmented in how I could use it until they got it at an enterprise license. I use it for my productivity, for research. I still use Google. I think it's funny some people don't use Google. It's like the intent behind the query matters, right? If I need to find out what's the address of this restaurant, I'm not going to ask ChatGPT that. But for example, we were looking at semantic layers, right? And I hear the word semantic layer. So many vendors love to say semantic layer. I don't know what any of them are talking about. So I tried to just be like, what is the semantic layer? And it was like, again, ChatGPT's answer was just as confusing as the real world. And I was like, Like, you also don't know what anyone's talking about. But my engineer was like, hey, we should look at kube.dev, which is a version of a semantic layer that is, you know, a way to model kubes on top of the warehouse, push them to the BI layer. And I was like, okay, but I don't really get it. Like, how does this, we're modeling data in dbt. So why are we also modeling in kube? Like, I couldn't figure out how these two things came together. And yes, I could have just called my engineering partner and been like, please explain this to me. But he is busy also. so. So I went to ChatGPT and I said, are these tools like duplicative? I'm confused. And ChatGPT is like, actually, they're really complimentary. And this is how people use cube and this is how people use dbt. And I was like, oh my gosh, this all makes sense. So sometimes like research like that. But I build a lot of custom GPTs for repeat tasks. Like I have a PRD creator. I know there's some tool out there called ChatPRD. But when you work in an enterprise, you cannot just use every SaaS solution that's out there. Your InfoSec team would not love that. So I built my own that formats math things in Markdown, I copy and paste it into Confluence and it's formatted the same every time. It makes it very easy to read. I also have done, the thing I posted about the most is like taxonomy experiments. I'm not a taxonomist. I have a friend who worked on a taxonomy team in a prior role. I have been an analyst who had a bunch of tags that became the taxonomy later. So I was thinking a lot about controlled vocabulary, how you manage a data set and all this stuff came out and I thought, okay. If we want to be able to ask robots about the world, we have to teach it about what our lens of the world looks like. And we can't just, I don't care how, you know, the U.S. government conceives of the medical world. I care about how ResMed talks about the medical world. So I started doing things like throwing documents into ChatGPT and having it do entity extraction and give me lists. And just, I feel like the thing about experiments like this is sometimes they save me a bunch of time, sometimes they don't. But I always learn about what are the bounds or the limits of these LLMs out of the box of that additional training you're like oh yeah it was not good at that or you know what's really good at, generating descriptions if you give it vocabulary terms with a little context for what domain they're from it actually generates very decent descriptions if you try to get it to extract from free text or from metadata from a database mixed results and all of it at the end of the day still needs a human it's basically just getting short draft lists to a human but a human still has to to make decisions about what goes at what level of taxonomy how how do we want to define relationships between nodes like what are the standard terms we want to use stuff like that so i just keep playing around with it i just keep throwing things in there and see what happens.
Tim Gasper [00:50:56.171] Humans need to be in the loop um and, you know, it's not a tool for everything. It can do a lot of amazing stuff. But like, if you're trying to classify something, then maybe you should use a regex or something like that. You know, like it's, I think that's one thing that everyone's starting to realize now is Gen AI is awesome. It does a lot of things that weren't really easy before. And there are some problems it's not the right tool for.
Anna Bergen [00:51:23.013] Yeah, like my first experiment, I tried to create a general taxonomy assistant that was like, Oh, I can hand this to me and have them just chat with it and generate me a list of vocabulary. So we're not talking about hierarchy. We're just talking about vocabulary lists. It couldn't even do that. The user just had to behave in more predictable ways. And then what I realized is, no, you need like one agent who extracts, like hand me a bunch of documents from your, you know, or screenshots from your dashboards. It'll extract values. Then you have to say, sort these into categories that make sense. Then you have to decide if you like those categories or like all these different subtasks. But at the end of the day, let's say at ResMed, but if we want to handle taxonomy centrally, we've worked with an official taxonomist. They're going to come in and bring a rigor to it that our SMEs, they tried to train our SMEs first to be like, this is what a dimension, like it was too complicated. The SMEs were like lost in a world if their brain doesn't work that way. So what the answer was, can I use these multiple agents together? Yeah, and it wasn't good out of the box. It could be good at subtasks, but then you need something that orchestrates those agents to show up in an organic way for the end user, which is why only I am using these and I am not handing them to untrained SMEs and having them get frustrated with me, at least so far.
Juan Sequeda [00:52:37.613] Wow, this is another topic we can go in. And again, my takeaway here is go follow Anna and go set. You're always sharing good experience. All right, so to start wrapping up, we got our lightning round questions, which I'm seeing them here for the first time. I'm going to go off. So number one, could all companies benefit from a data product manager.
Anna Bergevin [00:53:00.270] Hmm. Yes. All right. Oh, how about all companies over a certain point?
Tim Gasper [00:53:10.230] You anticipated the follow up question. We're going to be is it only at a certain scale?
Anna Bergevin [00:53:20.710] Yeah, I think that. I think until you I mean, you're going to need data engineers before you need a data PM, right? you're going to need a warehouse before you need a data PM. You're probably going to need analysts before you need a data PM. But yeah, if you're going to get your data headcount above a certain size, or if you want to build a product where data is the backend, so a customer-facing product, you should probably get a data PM with data background rather than try to have a traditional PM. There's just so many risks. Someone without a data background won't know. And so if you try to be too scrappy with a PM who's great at go-to-market, great at messaging, great at building user interfaces, but they don't understand data, there's a lot of risk there. So those are the reasons you'd hire a data PM. Your data team gets to a certain size or you need to build a data facing product that people are going to pay for.
Tim Gasper [00:54:04.310] That makes sense. Yeah. And it'd be hard to build a software product without an engineer as well, right?
Anna Bergen [00:54:10.150] Yes.
Juan Sequeda [00:54:10.150] Oh, next question. So are data product managers also the right folks to manage AI products?
Anna Bergevin [00:54:16.730] Um not necessarily, it just depends um uh i think all pms data pms and standard sass pms all need to get familiar with the capabilities of ai not just generative ai but like you should understand fundamentally how data works you need to be literate in these things so that you can engage efficiently with data science and your MLEs. But you don't necessarily have to be a data product manager to effectively use your data scientists. It helps. The more literate you are, the more background you have, the more efficiently you can work with that team. But I've seen people successfully utilize ML models in their product without having a data background.
Tim Gasper [00:55:00.330] All right, last question. Do you need to spend some time as a product manager to be a good data a product manager or can you like read some books and kind of self-service around it.
Anna Bergevin [00:55:16.069] That's such a good question. I think anybody, let me put it this way. If you're not going to become a product manager first, you really better have a good mentor. If you just try to figure it out just from books and podcasts and you don't have like, either you report into a director of product or you have an official mentor you meet with weekly for coffee chats, you're like, this is what I'm dealing with. When I first became a product manager, I had a mentor. We had a mentorship program through our business and it was a much more senior product person. And I would sit down with her and she's like, well, what are you thinking for your roadmap? What are you doing? And I started rambling at her at this coffee shop. And she's like, you need to center more on the customer. It's just that muscle or that training of you're jumping to solution too quickly. You need to sit with the problem longer. Those muscles and habits, if you've already been a product manager, you can bring all that to data. If you're a data person and you just learn product, as we said, everything outside that's written, it's going to feel very...Everything you read, everything you listen to on a podcast, you're going to think, but how does that apply in my weird enterprise platform world? It's all like SaaS customer facing. So could you without a mentor and without a background? Yeah, but it's going to be really hard. Hmm.
Juan Sequeda [00:56:28.078] All right. That's fantastic. So, Tim, takeaway time. Take us away with takeaways.
Tim Gasper[ 00:56:33.538] Let's do it. So many takeaways. I'm going to do my best to hit the top ones. So, you know, we started off, Anna, with how did you get into data product management? And you said that, you know, you were sort of at the top of the data analyst career ladder and you were trying to figure out what to do next. And you got an opportunity to actually be a product management pinch hitter, which I think ties a little bit into our lightning round section, which is that, you know, if you get an opportunity to do product management, that can be a really great accelerator. Of course, having a mentor or something like that can be great and valuable as well. You ended up actually having an opportunity to go into products, you were a product manager, you became a growth PM for a little while, and then you were actually looking for a job. And, you know, somebody was looking for help with data product management, and you got to step into that. So, you know, you were a data person, you were a product person, and then you got to merge those two things together as a data product person, which I think is a, is a really interesting and cool journey. Um, and then, you know, we kind of talked about, do we have it all figured out? And you were like, no, we don't. Um, but you know, there's no book for this. Uh, there's good resources, but, um, you know, there's a lot of good content out there that you can find. I think Anna, you put some really awesome content out there. So I think folks should follow Anna and see what you're putting out there. But I think folks are learning more about data product management. And that's important. But you know, if you were, if there was a book, what would be in it? And you said, like, hey, you got to don't just build what you're asked to build, right? You got to think thoughtfully about the problems that people are having, and what's the right solution, you should be strategic, you should, obviously, you have to, you know, resolve certain needs that are urgent, but you really are trying to develop a platform that serves what's going to, you know, becoming in a year, right? So you're thinking not just short term, but also long term as well. And, you know, you said that you kind of have to be slightly more product person than data person, in terms of especially the process and approach, you need to listen, you got to make sure you're not just jumping to solutions. But of course, data experience is core, right? That's what allows you to have that empathy and that perspective and really understand. Internal versus external product management, those two things are different, right? Usually, you know, internal product management, you're not as worried about monetization or, you know, some of the ethical considerations and things like that versus with external product management, you would. And, you know, we talked about, you know, how many data PMs and kind of what the process is at ResMed. You said you have two data product managers and And you could have more. But as you think about who does data product management work, some people are already doing this. They might be a data engineer. They might be a business analyst. But you get to a point where you're like, hey, maybe we should be consolidating some of that work, right? And being able to get more scalable about it. So I thought that made a lot of sense. So many other things. But Juan, what about you? What were your big takeaways?
Juan Sequeda [00:59:27.160] All right. So on the team aspect, which is interesting, I think it's important to understand kind of what circumstances are where people are listening, right? Right. Where they work at like in your case you don't have a specific engineering manager to work with like which is usually what people have in software engineering design uh in ResMed you are very distributed teams right you don't have a designer but you work very close with your engineering counterparts but hey you like design is really important you think about the inexperience not just the ui but it's the whole usability uh on the process part i think it's it's some aspects maybe waterfallish because you have big migrations but some others are going to be agile i mean it It really depends on how old your organization is and how mature they are. Some parts can move really fast, some parts are going to be really slow around that stuff. What are the trickiest problems that need to be solved right now, especially in this era of AI we brought up? We need more structure in the data. People are saying, we're not going to have to learn SQL anymore, but hold on, the computers need to learn what that stuff actually means, and that means it has to be written down somewhere. So we've got to get it down. So there's the biggest issue is, how do we get the context that's in people's head? And you're like imagining a multi-agent system that you have like a team of taxonomical ontology experts who can work with the SMEs around that, right? So how do we make documentation easier, intuitive? Like what are the incentives around that? We're saying, hey, software engineering teams, they need to go push the data dictionaries. That can be a contentious kind of conversation there, but maybe those are the type of thing we need to start discussing. And that's kind of what's missing in the data world right now. Then we had this conversation about data quality, which is like, hey, everybody talks about it all the time. But like, for example, if we brought in entire source data, do we have a good replica like that? But maybe that good replica has some quality issues. But at that point, that was correct, right? Quality is in the eye of the beholder, right? It is fine at the moment that the ERP was set up to have some free text, but it isn't fine right now. Really, you got to ask about quality. Why do you care? What is the use cases? Who's going to remediate all of this stuff? A T-shirt quote to data quality. Do we go back and rewrite history as if it never happened? Ben, how do you use GenAI for your product management work? You started officially using it six months ago. A couple of things. You use it for research or definitions. So they compare how this vendor is talking about this thing. You built custom GPTs to create PRDs, so product requirement documents. You're using it for taxonomy, using control vocabularies to extract entities from documents. And you have to acknowledge, sometimes it works to save you time, sometimes it doesn't. But you need to experiment and try these things. It works great for generating descriptions and don't forget we need humans in the loop. Anything we missed?
Anna Bergen [01:02:06.281] I mean not that i can think of it sounded like a good story.
Juan Sequeda [01:02:09.681] Hey this is just this is you so yeah great to wrap us up uh three questions. What's your advice, who should we invite next, and uh what resources uh do you follow?
Anna Bergevin [01:02:26.701] Oh okay um what's my advice Um, my advice is to, um. To use writing to clarify your ideas before you start building anything. I said I use Gen AI to write PRDs, but like, it's a very conversational, like, it's not just me going, imagine this solution and build a thing. And then I often have to edit what the PRD gets written. I think often when I go to write something out, that's when it clarifies for me. What's missing? What assumptions are people not talking about? Hey, this doesn't make sense. Even if that artifact is just for me even if it never sees the light of day and even with all the ways I use AI to shortcut things I still end up writing and and rewriting even if I use AI to assist me to getting a first draft I always end up just cleaning it up and that process is very, thought-provoking so I think everybody should do that and especially anyone tackling a data project should draft that first what are the assumptions what do people actually want and ask more questions Try to get the core of the pain point and not just be reactive. The second one, who should you talk to? So there's another data PM you could talk to. Nick Zervoudis, I think is Greek. I don't know how to pronounce his last name. But Nick and I connected online. We've talked. He really gets this data product stuff. We're of the same mind. But, you know, he's working in a very different business, very different context. And I'm going to recommend my friend Adrienne Baker who is not very active on LinkedIn but I could connect you to who worked on taxonomy as a product at a business and what it means to like have a controlled governed data set like a data product have what it means to version that over time what it means to merge things and to have so many dependent use cases on a data set that you are actively managing and she's just a very thoughtful interesting person I hope she doesn't kill me for sending you her way. She's great. And what do I follow or listen to? Like I said, there's not a lot directly on data product that I encounter. I basically follow every data product person I can find on LinkedIn that I'm just like, tell me what you're doing. Tell me what's happening. I also follow data engineering and information architecture people. And I think these are very different thought circles and it's very helpful to see the ways that they are perceiving trends or things that are happening. So I think curate your little bubble online. I heavily, like I unfollow people with all they post is like memes and fluff. That's like nonsense, like influencery type stuff. I'm unfollowing that. I don't have time for that. But I also try not just to follow people who are just like me because I want to hear how are people handling things in a different way. I listened to a couple of podcasts I listened to, but they're producty. From Lenny's podcast, Lenny Rychitsky, and I listened to Melissa Perry's podcast, which is another product podcast. And I listened to a lot of like, random data podcasts, often picking episodes that sound relevant or interesting to me, a variety of them. And, you know, that's all I have time for because I have two kids and a full-time job.
Tim Gasper [01:05:42.226] That's a lot of stuff. And by the way, Melissa Perry's podcast is great. I love that one.
Anna Bergevin [01:05:46.586] It's really good. And it's a different flavor than Lenny. They just approach it differently. It's good.
Juan Sequeda [01:05:51.066] Well, Anna, thank you so, so much. Just a quick reminder, next week we have Tiankai Feng, who is the Data Strategy and Governance Lead at ThoughtWorks and just recently has this book, Humanizing Data Strategy. And with that, thank you, thank you, thank you so much. It was a fantastic conversation. As always, thanks to data.world who lets us do this now for year five.
Special guests
