Data and Analytics FinOps with Mark Stange-Tregear, VP of Data, Babylist

Elif Tutuki: Hello everyone. Welcome to Data-Driven Podcast. I’m Elif Tutuk, Global Head of Product at AtScale. Today we have a very exciting topic to talk. It’s, the new strategies for managing enterprise analytics cloud spend. And together with me, I have Stange-Tregear, who is the VP of Data for Babylist. Hi, Mark.

Mark Stange-Tregear: Hi. Good to meet you.

Elif Tutuki: Yeah, great to meet you. So, as we are the, with the, all of the data gravity shifting to cloud data platforms, enterprise analytics teams are adapting the shift to consumption-based pricing models.I think one of the things that we have seen, with pandemic, just not only for data, but all of the services moving to cloud, but in our topic as it is the data-driven podcast, like we will be more focusing on the, the reasons for the, and how to help organizations to manage and monitor, the data and analytics related, cloud consumption. So I think you have been living and breathing as the head of, data for many organizations. And now with PayPal Paper List. do you want to briefly talk about, you know, what is finops, why it is important, why we are seeing this trend more and more now Mm-hmm.

Mark Stange-Tregear: . Yeah, certainly. So, I mean, finops goes well beyond the data space. I mean, it, it refers to a pretty general sense of trying to make sure that you are controlling, understanding, monitoring, managing your, your online expenses. obviously the name is a little kind of parallel to DevOps. it it’s that idea that you’re really taking control, that you’re not just spending money. and I think for me at least, where I sort of really start to grow was, you know, people having negative experiences where these super powerful cloud platforms came along and people started using them, and then people started using them a lot, and all of a sudden these surprise bills arrived on people’s doorsteps. and there was a, you know, an immediate sort of reaction like, what happened, , what, what just went on there And I, for me at least, that’s where the journey started.

Mark Stange-Tregear: I think that’s true for, for a lot of people. Like, how do we get control The, these tools are incredibly powerful. They open up capacity in a way that, you know, 10 years ago just wasn’t really there. But how do we, how do we control it How do we manage it How do we stop those kind of runaway expenses I think at the more sophisticated level, it’s a matter of, not just cost cutting, but how do we make sure that we’re spending money in the right places and in the right ways So it really becomes a return on investment calculation, and, and process. When you get to data, more specifically, I think the, some of the principles are very similar. You know, managing your A W Ss s cost, whether it’s for kind of data or whether it’s managing production environments, you know, your APIs and production databases.

Mark Stange-Tregear: A lot of the principles apply. I do think that within the data world, there’s, there’s some, some features which I’m maybe not exactly unique, but some things were a little bit more nuanced or, or require a different approach, or at least a more specialized approach. So maybe the one I taught and thought of, thought the most about would be managing, snowflake costs. same could be true for Redshift or be query depending on what you’re on. I happen to, I happen to be on Snowflake. Then how exactly do we monitor, control and optimize what we’re spending with Snowflake how do we keep an eye on it How do we, how do we rationalize that back into the business and make sure that we’re spending what we wanna be spending in the right places

Elif Tutuki: Yeah, that is, that is great. I think one of the things that you have mentioned, like ultimately, you know, it’s about making data informed decisions. Like we have been chatting about, and I said data-driven, and you correct me. It’s more about data informed, which I love that. and maybe one of the reason seeing this type of peak in terms of the, you know, cloud-based analytics consumption is not only those cloud platforms are very, very, very, you know, strong and performance scale. Mm-hmm. , but also like now there’s also the trend where, there’s more power users sitting together with the business users. And, you know, one of the trends that I’ve been seeing is this, change of E L T to E T L E T L to E L T, where, there’s more users doing data transformation to make that data business ready. so there is more hunger for data, and that is also the other reason. And that goes back to your point, maybe if you can comment more, you know, how you balance it out. Like you don’t want to, stop people using data that is ultimately not the goal, but how do you manage that and what are the, you know, some best practices that you can share

Mark Stange-Tregear: Yeah, definitely. I mean, I think, I think that that trend’s, right and I think that in part that’s because it’s become much, much easier to move very large volumes of data and store it relatively cheaply. I mean, there’s multiple different approaches there, but it, it seems like overall that’s been a, an ongoing thread within the data environment. So in terms of thinking about where you’re spending your time and resources, it doesn’t make sense to spend for most organizations, I would argue at least, it doesn’t spend a ton of, make a, a lot of sense to spend a ton of time trying to figure out, you know, what little bit of data should we move What should we store What are we gonna sub select It makes a lot more sense just to just keep all of it, right Trying to appropriately manage a, you know, so you’re not violating privacy concerns and whatever, right

Mark Stange-Tregear: Obviously that’s an important thread. And then work with the full set of data and think about how exactly do we go from that to something that we want to use, right You’re sort of pushing the, the conversation downstream a little bit. You’re less likely to end up in situations, which happened all the time previously, where it’s like, well, we have this question. Okay, but that data doesn’t exist ’cause we threw it away ’cause we didn’t know we needed it. so I do think that that’s true. I think it’s pushed the use cases more into kind of the modeling closer to what the business actually needs. That’s where a lot of the work is happening. the, the challenge with that, or one of the difficulties with that is you can end up, in a situation where you are getting a lot of more ad hoc workload, you can, end up with really sort of splintered kind of pipelines or modeling processes, which can come with, in the data space.

Mark Stange-Tregear: I mean, if you’re dealing with any amount of data that’s gonna reprocessing all of those separate models is gonna start to rack up those costs. If you’ve got 500 Tableau workloads all running large extracts, that’s gonna rack up your costs. If they’re all inactive use, if they’re all being used within the organization to inform the decisions that you’re trying to make or to improve your customer experience in some way more or less directly, then great. Right That, that’s a very sensible use of time and resources to maintain all of those workbooks and all of those models. If you are using 10% of them and you’ve got a whole bunch of stuff out there that’s just refreshing for the sake of refreshing, and it’s just costing you, you know, 6, 7, 8 figures a year to produce this data or produce these models or refresh these workbooks that no one’s looking at, or maybe someone’s looking at, but they’re not actually making effective decisions or driving the business forward on, like, that is not a good use of money.

Mark Stange-Tregear: The right, and ideally, when presented with this information, and I think most organizations would say, yes, let’s not spend that money. Yeah. Right. Where it becomes tricky and where the discipline of finops comes in. I, I mean, the principles are fairly straightforward. Like it’s a good financial sort of oversight and, you know, let’s make sure that we’re spending money where we wanna spend money. How to actually do that becomes incredibly complicated. I mean, if you are running, I don’t know, a hundred thousand snowflake queries a day, which is small actually for a lot of organizations, right Like, how are you supposed to go in and understand what are we processing Is it really worth it And it gets even more complicated. Like even if you can figure out if someone’s technically using it, are they using it effectively Are they actually using it to drive some kind of positive return

Mark Stange-Tregear: Becomes much more complicated. And you get, you go from this somewhat kind of pure technical problem into a much more complicated problem where you have to try and combine relatively nuanced technical detail about Snowflake costs and optimizations, or you know, exactly have Tableau or Sigma runs queries. Then you have to translate that into the business context where, well, this person’s a marketing channel manager, they’re responsible for managing x, you know, million a year in terms of acquisition costs, and they’re using this information to try and drive those decisions. So how, but it’s costing us x to produce the information for them. So are they actually in a position to optimize enough out that it’s worth providing the information And I think for years the direction was just more data, more data, more data, because the systems were somewhat limiting, right Whether that was, I mean, going way, way back, whether that’s an old like SQL server machine where literally, you know, you bought the machine and you did you the most you could do with that machine, and it was about sort of access time.

Mark Stange-Tregear: Or if it’s Hadoop clusters and you’re buying racks of these things, right And you’re putting millions of dollars down to run them, buy them and run them. Or if it’s in the cloud, you know, you, you have this limiting factor where there’s sort of a SEC capacity. And the question is, well, what do we do with that capacity And you’ve probably got a technical team, which is thinking about what that SEC capacity, you know, they, they, whether rightly or wrongly, they’re in control of how that capacity gets used, right Mm-hmm. So if that marketing person wants even more data, you got a team of data, data engineers often sitting there, often within the technology organization making the call as to whether or not they get it with Snowflake. With the, with BigQuery, that’s not true at all. I mean, depending on how you’re set up, that person can just fire off queries as much as they want, and they get whatever data they get back. And if they spend $50,000 tuning a hundred dollars of spend, yeah, well, they get, they get to talk about the a hundred dollars of spend that they tuned, they just don’t mention the $50,000 that they spent. And if someone else isn’t watching it, it basically just rolls up at the end of the year into, oh, look how much we spend on Snowflake. That that was a lot of money.

Elif Tutuki: Yeah. I think just like, kind of like, this is how I look at life as well. Like you, you enable people and then you kind of pause the monitor, like in terms of what benefit and what they are doing, and then you shape that. So one thing that’s, you know, we have been thinking, at within, at scale as the product, like as the semantic layer sits in between the all and alters consumption of an organization and the cloud data warehouses, like the IT scale sematic data platform sees all of the analytics conversations of an organization. Yep. So like we have this active metadata mm-hmm. . And then with that, as you said, like now you are the, you know, as the head of VP of data, now you, you can actually kind of realize, okay, you know, these business users are really rocking like they are doing data informed decisions, or like you can actually see the business impact that they’re generating, let’s say maybe, you know, there’s, you know, awesome work that they are doing around the customer churn.

Elif Tutuki: and then like, again going like by leveraging the activis, active, metadata, then we can expose you like, okay, what metrics they have been using, what queries they are running mm-hmm. . So that can actually justify the cost of that business unit because, but then, you know, also they are getting the r o i Yeah. Versus like just seeing that a group of users, they’re running queries, they’re using these metrics, but they are not generating that type of business value. Maybe that is a conversation. So in those cases, maybe, if you can maybe talk about that, like, first of all, how, what do you use, how, like in terms of tooling or the practices or the people skillset, like how do you, provide that type of monitoring And then once you realize something, what’s your way of engaging other stakeholders or within babylist Like do you have already a, an an organization that is responsible overall for finops

Mark Stange-Tregear: Yeah, I mean, we, we do, babylist is not a huge organization, so the, for us, it’s not actually too, too complicated yet. but I think the same principles that we’re employing can, can scale. And I’ve used this at bigger organizations as well. I think there’s a couple of key fundamentals that you have to get down if you’re actually gonna get into a finops practice. number one, and this is probably blindingly obvious, but I’m gonna state it anyway, is you have to have sort of executive buy into the concept that this is something important that we want to talk about and look at, and it is worth spending some degree, right And that can mean a number of different things depending on your organizational size, but we’re gonna spend some time and effort thinking about this, right I haven’t seen that being a major hurdle just ’cause the bills from things like a w s and stuff.

Mark Stange-Tregear: Exactly. So big. And people I think at this point have the, there’s a natural fear, especially from finance departments of, hold on. So people can just spend as much as they want here, and I’m just gonna find out later. So usually that part isn’t too, too difficult in my experience. I think once you’ve got past that, the next, the next stage, and again, this sounds kind of obvious, but I, I think this is one of the cause that a lot of people haven’t really done, is to think about the scope of what you’re gonna monitor. So a lot of people I’ve talked to in this space, particularly for data, are focused on Snowflake or BigQuery because it’s sort of the biggest and in a lot of ways, most flexible. Totally. But there are other elements. I mean, one of the, that can get much more complicated.

Mark Stange-Tregear: So A W Ss for example, are you looking at how much you’re spending on a w s to process your data or on Google Cloud or on Azure, right Wherever you’re, wherever you’re doing that processing, if you’re running kind of Kubernetes loads or, or modeling load, like are you taking that into account A lot of organizations, the A W SS infrastructure is not set up in a way that you can actually get visibility into that. So the answer is sort of by default, no. Like we kind of know roughly what we’re spending on A A W SS or whatever, but you have a go good visibility. And then there’s other things, like if you are a consumer application, you are probably doing some kind of event pipelining coming off your application letters. Those bills add up, and they’re also very fluid, they’re very flexible, they’re very spiky, right

Mark Stange-Tregear: Like a developer implements a new, you know, impression viewed event somewhere on your application, and all of a sudden your event volume doubles overnight, like the cost per just moving depending on who you’re with. But say you’re using your segment or mParticle or one of the standard tools there, all of a sudden you’ve got a bunch of, you, you are into a whole different tier of costs there. You’re probably getting more e t L costs as well to move the data, because that’s done on a by rob basis. So the, the second step is to really think through, well, what are the tools I’m using and which ones am I going to consider as part of my finops practice I tend to go everything. I’m very clear on what tools I’m using, what resources I’m using, and either we are watching them or we’re on a direction where we will be watching them really some of the time. And that’s the, that’s kind of the third step is once you know what tools to monitor, are you in a position to monitor them And are, are you in a position to monitor them in a way that you can actually sort of break down the costs in a way that you can connect back to the business

Mark Stange-Tregear: Often not, right Like if you, if you’re not doing this pretty deliberately, if you are running, say you’ve got a Snowflake infrastructure and you’re running, you know, 10 large warehouses that are called warehouse one through 10, and different people, you know, different workbook, different tools, they’re just connected to different ones. And then someone comes in and says, okay, well how much Mark, how much of that is marketing

Elif Tutuki: Yeah, right now you don’t have result. But again, just perspective, like if you, if you have a semantic layer in between that kind of knows all those conversations, then you know, this is, you know, not kind of makes me awesome. Like, like happy about this, like that it excites me that, that possible to provide that, which we are actually just really a, feature in the product to enable.

Mark Stange-Tregear: Yeah. Yeah. That’s right. And there, there’s definitely tools some of the time. and tools like ADSCALE can really help. I did. I think the tools are starting to be there. Some of it you just do have to go back and restructure though, right And you have to think about, well, snowflake’s a great example. ’cause there’s multiple approaches to doing this. Like, do I have separate warehouses Do I actually push into separate accounts Do I push even into separate organizations There’s multiple different ways that you can carve things up, to try and get this visibility. But if you’re not thinking about how do I, how do I translate my cost back to my business units so that I can try and get that return on investment, or I can understand that cost in a way that makes sense for, to put into the rest of my business context.

Mark Stange-Tregear: You’re probably not gonna just default into that. You’re probably not gonna just fall into it. even if you do for something like your cloud warehouse, can you really do that for all of your data tools, including things like your E T L tools Probably not the Yeah, yeah. There is effort. Yeah. And yeah, the, the, the final thing I always say is, who’s responsible for this And that’s actually a really pretty complicated question, right Like, because in a lot of ways, if you have a sales team that’s spending a bunch of money on Snowflake or on reporting, you want to argue that it’s the sales team, right They should be explaining why they’re spending that amount of money. On the flip side, they probably, most sales organizations wouldn’t by default have someone sitting there with the inherent skillset to tune Snowflake processes or D B T models to save money, right

Mark Stange-Tregear: Like, so you, you get forced into this kind of cross-functional world where you’ve got sort of business use or even more technical use, which has to be connected back to specialized skill sets to actually then optimize the cost. Yeah. That is what I’ve found is that if you’re not very clear about that, that can really confuse ownership. Like whose responsibility is it And the answer at the end of the day is it’s probably a joint responsibility. But that’s where a lot of even organizations that have got some visibility, because they haven’t stitched together that level of responsibility, they haven’t done the work to talk to the business teams about what this really means and how it can be managed. Actually, not a lot happens. You get the visibility, but without any true ownership.

Elif Tutuki: Yeah. So like what you’re saying, mark, just make sure that there is executive buy buying, and which I agree that there is, because now everyone is realizing, but then is it more, having, a program fin a program that’s been, driven maybe by the data organization, but then like making sure that you have stakeholders at the business side, because at the end, those stakeholders, like you are kind of the, even the news to them, Hey, this is how much you’re using enhanced costing this much mm-hmm. , and then it is the business to justify that cost. And you mentioned that there are some challenges on kind of really verifying that justification Yeah. Which we can double click further more on that. But then once you know there’s an agreement, yep, we can decrease that cost, we should decrease. And then that comes back to your teammate right now, like to kind of, make sure that there is the right optimizations that can be done. Is that correct

Mark Stange-Tregear: Yeah, that’s, that’s the general outline. that’s the general outline. And I think, the, the way to see how that can be different is imagine the, for example, you just, you signed a million dollar contract with Snowflake for the next year, right And it’s just up to the data team to manage that. And it’s like, well, that’s where you get into trouble because the data team really in a position to actually make the fine grain judgment calls about whether, you know how to divide up that $1 million or whether actually that just wasn’t a big enough contract in the first place, you really should have signed a 2 million contract and it is just AFI that you’re gonna blow through that and go and buy some more. That’s very difficult to do if you put it back on, on a sort of a more traditional like, tech centered data team without that connection into the, into the business and the rationalization.

Mark Stange-Tregear: I think the, the other thing that people really struggle with is this, it is an optimization, not a cost cutting exercise. Yeah, that’s perfect. Right So, and a lot of people miss this, right It’s about how do you, how do you bring your snowflake cost down How do you bring your big query cost down How do you bring your segment cost down That isn’t necessarily the right thing. If you’ve actually, if you’ve actually done, I mean, sure take out sort of obvious waste, but once you’ve taken out the obvious waste getting the next $10,000, if it’s gonna cost you six months of a data engineer, that’s a ridiculous use of time and energy. It’s better to pay the 10,000 ’cause their salary and lost opportunity costs is so much higher than what you’re gonna return. And I’ve actually been in situations where the, I’ve encouraged the business team to spend more ’cause actually, right

Mark Stange-Tregear: And we, and push the bill up because when you think of as an optimization, if the data is really being used, if it’s being used effectively, and if you’ve kind of already tuned the models, and then you get to the point where you’re saying, well, okay, but the, if this is too slow, like if it’s not responsive enough, if, if we’ve got too long a wait times, you know, imagine a, you know, a sales team that’s having to wait a minute for reports to load, they, they’re meant to be using on calls with their clients, well, that isn’t gonna work. So if you’ve already tuned this thing as much as you possibly can do, and you’re still on a one minute wait times, and the only way to speed that up at this point is to throw more power and more money at the situation, well, it’s probably a really good idea to throw more money and more power at the situation, even though that’s gonna increase your bills. It’s, and you’ve gotta have that, so full view and the full sense of return on investment, otherwise you actually potentially end up hurting the business. Trying to penny pinch.

Elif Tutuki: No. Yeah, that is, that is, I love hearing optimization. you know, in some of my conversations with customers and the, you know, head of, CDOs, it’s usually like, so far we have been thinking about, you know, traditional data governance, especially when you think about semantic data. But then now with the age of cloud, cloud data analytics, it is not only data governance, but governance of, of cost and optimization of performance and cost. Yeah. So like when it comes to that conversation, I always also kind of refer as optimizing the performance and cost. Yeah. And that is where also like, again, coming back to, you know, I will be always talking about semantic layer , just in terms of like how semantic layer can also help, optimize that because as, as it knows about the usage, then it can actually understand, okay, these are the most common query patterns that let me create an aggregated materialized view back in the data source without extracting. ’cause also that’s another, downside, but within the data source, and then, you know, let the semantic layer doing the trafficking in terms of okay, you know, we all, I already know that there’s a materialized view. Let me kind of push that query back to that materialized view. Yeah. So, and, and I think it’s just like we will see more and more technologies hopefully helping that optimization as well, that will make your job easier.

Mark Stange-Tregear: Yeah, I mean, I think Yes, I agree. I think, yeah, I definitely experienced that with AtScale on the, in the, in the semantic layer. I mean, for years actually, lesser, lesser babylist and more in previous roles, where I used the AtScale product, I was heavily focused less on, some of the semantic kind of joins that had done on the fly. Like we actually ended up, like we had stable out enough models. We actually did that as part of our E T L process in a lot of situations. But the smart aggregation, right, like one of the quickest and easiest ways to rack up a huge amount of cost is to have a bi like a BI tool that’s pointing at a really granular table where you have repeat use that requires aggregation.

Elif Tutuki: Yeah.

Mark Stange-Tregear: And aggregation, it’s a, is one of the quickest and easiest ways to blow your costs up. other than given a data scientists access to a fourex Excel warehouse , that’s another really quick and easy way to blow up Yeah. Blow up your costs. But if, if you have, again, right Like, and it’s good like sales teams, customer service teams, marketing teams, right And you give them dashboards or data to help them make the kind of ongoing decisions that they have to make day in, day out. And there’s a lot of them, and they’re doing this repeatedly, and you are set up in a way that each one of those interactions is pushing queries down to large tables over and over again. I mean, some of the tools out there, you know, like, like Tableau and like Snowflake are trying to tackle this.

Mark Stange-Tregear: They’ve got caching, they’re relatively small, right They’re trying to hold things in memory or re-serve things. But with aggregations, that becomes particularly tricky. Or if you’re passing down custom parameters, right It, it’s tricky to get that right. ’cause you can’t just cache the results that the previous person looked at. And that’s, I think where yeah, tools that sort of sit in the middle of this and get, can get more sophisticated about anticipating patterns and sort of, be clever about, well, we know there’s gonna be this repeat pattern that can be served much more efficiently if we just do this one time run. Versus pushing this down 150 times in the next three hours, right Yeah. Is a huge win. there’s just a, there’s a lot of room here. I, one of the other key areas is, looking at E T L or D B T kind of patterns and think of, you know, like DB t’s great, like building models in a very stretched kind of way.

Mark Stange-Tregear: But then how often are we seeing end users actually combining kind of mark tables or end tables, or going back to raw source tables over and over and over and over again. Yeah. Sort of working effectively around the data products that you’re building. And how do you, within all of your snowflake logs or your big query logs, how do you identify that kind of repeat a pattern because it’s pro, especially if it’s coming through SQL and not through a BI tool. ’cause the queries probably look something different every time. And then how do you actually anticipate that and translate that into, oh, we should, we should really just build a model there, create that at 7:00 AM in the morning and we will knock out, you know, this 1000 custom queries a day that are kind of killing us.

Elif Tutuki: So in those cases, like if you don’t have a system that does this automated data engineering based on the usage patterns, like do you have a, practice in place where you maybe, classify different user groups Like, okay, this is the set of users that are power there at Hoc, you know, that they will be running those types of ad hoc sequels that, and it’s okay because part of the job, they’re going after the hypothesis. so like, maybe like segmenting users is one way. Like what other methods do you use or,

Mark Stange-Tregear: Yeah, segmenting users. it, it in some ways the most obvious way. like, so my standard approach to this, particularly with Snowflake, is to separate, is to separate, have separate warehouses with the different organizations within the department, within the organization rather. So we would have separate sales warehouses to marketing warehouses. You lose a little bit in terms of kind of data caching, but you gain, like I would argue significantly more in terms of ability to optimize. I often have warehouses set aside for the data scientists or the analysts who are more likely to go in and poke around at stuff. but where you can also better, you know, more understanding of different sizing and costs, right So you just sort of pull that, pull that out when the organization gets big enough. You may have, okay, this is my sales data science warehouse . Yeah. So you know exactly who, who you’re going after, you can kind of get back into, you can back into it depending on the number of users you’ve got by looking at the user logs and then mapping the user’s de departments. But that means that you then have to implement some kind of metadata layer that translates users into departments

Elif Tutuki: . Yeah. Yeah. Help . Yeah.

Mark Stange-Tregear: Yeah. Which you can have. it just means that you’ve gotta put that in place, right So you’re gonna have to do some work somewhere along the line, to be able to actually manage this effectively. But yeah, you can choose where to do it in some cases.

Elif Tutuki: So Mark, I know you have spent a good amount of your, like, you have great experience with the, like retail, sector mm-hmm. . so maybe if you could briefly talk about the analytics use cases for your current organization, like, and also just overall for online retail sector. Like how does analytics drive business value mm-hmm. .

Mark Stange-Tregear: Yeah. I mean, so I think Babylist is, Babylist is a really interesting product, for those, for those people listening that don’t know, it’s a, it’s an online baby registry. You can go in, and we have our own store. So you can, create your registry online, as you maybe would’ve done in the olden days on a piece of go into a store and do it on a piece of paper and you can add things from our store. But we also, were just universally, you can add products from anywhere else, and effectively we will consolidate it and link it all nicely for you and make it visible in one sort of easy to access place for anyone who wants to go and get you gifts. we have, very strong consumer focus. We know our audience very, very well.

Mark Stange-Tregear: we’ve been around for a while. we’ve got also a lot of connections and very strong relationships with vendors in the kind of baby good space where, and there’s a lot of, there’s a lot of vendors in that space, and they can be relatively specialized. So we have the, we have these multiple sort of internal facing thing, internal facing teams, but we also, we do have the consumer experience. We’re a large scale consumer app. We then also have multiple B two B type business relationships, whether that’s, sort of a vendor supply chain type relationship or more of a media type relationship. And we use, we need data in all of those different places. So, the general philosophy that we have, about the use of data, you, you mentioned it, we don’t actually talk about data driven, is data informed . Yeah,

Elif Tutuki: I love that. Yeah.

Mark Stange-Tregear: The idea being right. Sometimes because we’re close to our use cases, we’ve got a pretty good idea of what’s going on with, with, without data. We always like to bring data to the situation where we can. but it, it isn’t, we don’t derive from the data necessarily, but I think we have a lot of the standard use cases, the, the most kind of consumer, consumer applications have. We, we need to understand how our consumers are interacting with our product. It’s important to us, to try and improve that. And obviously if we can improve the product, we expect that people will ultimately spend more and users us more and refer us to their friends more. And, you know, it will, it will help our business grow. And, in the long run, that takes quite a lot of crunching, right We can crunching those consumer applications.

Mark Stange-Tregear: We have to monitor our, our sales volumes. We have to monitor our registry volumes. We are, you know, we’re running C R M programs, we’ve got big email and push campaigns. And then on the, on the other side of the business, you know, we, we supply a lot of information back to our vendors and our partners, helping them understand where the product demand is, helping them help us to make sure that their products are in stock and in the right place, and that we’ve got the right volumes, or that we’re, you know, putting experiences out there that can, that can help the consumers to find the products. The interesting thing about the baby space is it’s a lot of discovery. Mm-hmm. , if it’s your first kid, like you probably don’t know the brands, you probably don’t know the vendors. You probably don’t have a preferred stroller brand or bottle brand or whatever it may be. Yeah.

Elif Tutuki: As a consumer, you are using actual data to kind of Yeah. Inform and make decision as the purchase. And you are using data for your consumers so that they can get the best ultimate, experience out of that. yeah.

Mark Stange-Tregear: Mm-hmm. . Yeah, that’s exactly right. And that fi that powers a lot of kind of, thought about d discovery and how do we, how do we help people discover from a very often a very cold star type of position, which is an interesting consumer, you know, it’s an interesting consumer position to be in.

Elif Tutuki: That’s, that’s interesting. So as you are talking about these analytics use cases, which is awesome, like what is the cost allocation Like, do you have a, a central cost, you know, is it the finance department who, who is the kind of managing the all cloud costs Or is it, do you have any chargeback structures if you can maybe talk about that some more. Yeah,

Mark Stange-Tregear: So we’re evolving the practice out. I would say at this point, so my team manages all of the data related cloud costs. We’re fully aware of what they are. We negotiate all of the, we ne we are on the front line of negotiating, negotiating all the contracts and dealing with all the terms and conditions, and then sort of visualizing back what the usage is and what, and, what the usage is. And we’re starting to build that kind of practice of allocating to allocating the cost back to certain teams, back to certain departments. And we’re, we’re ramping up those conversations. I would say, you know, we’re, we’re maturing in this space. We’ve got a sense of where we’re going. It takes a little while to get there, right Like, it is not inexpensive to reorganize your infrastructure to be able to monitor and to be able to associate the cost back. So, I would say we’re doing more than chipping away at it. That sounds too slow, but it’s, it’s a process that,

Elif Tutuki: That’s, that’s, yeah. Go ahead. Sorry

Mark Stange-Tregear: For nd vendors that are listening. So at scale, I think is ahead of the curve on this. For anyone else that may be listening in, this is starting to become a factor in my, as I’m thinking about what tools we’re using and procurement, this is becoming a factor in how I think about things, right Like, can I stretch this in a way that I can actually understand where my costs are coming from, that I can explain this and, frankly, in a way that is understandable and usable with my sort of business peers and business stakeholders, not just to the technical side of my team.

Elif Tutuki: That, that’s good. That is actually kind of ties back to my next question. I was going to ask about what are your challenges I think, when we were prepping for this call, we have been talking about, you know, from a scale perspective, you know, it’s all about antics consumption, providing visibility and helping you with that. But then you have ma made a good point, like, you know, there’s out of other services running on the cloud. So maybe if we can talk about like, what is the future trend What is the need that you see in the market to kind of help you with some of your challenges

Mark Stange-Tregear: Yeah, I mean, I think, I think just doing this at scale, I don’t think there’s too many templates in how to do this well, which is interesting. and I think that’s probably a sign that this is still an emerging line of thought. I mean, I know there’s some groups out there that are starting to talk about this, but I don’t see it, especially in the data space as a well established kind of, area where there’s been a lot of thought or where there is sort of the, the playbook, like, this is how you do this. if it exists, I certainly haven’t seen it. And I try and keep my eye open for these things. the, so where are the challenges that I’m facing Well, some are maybe a little, specific to Babylist. we’re a very rapidly growing company.

Mark Stange-Tregear: We’re innovative. We’re doing a lot of new things. and just trying to, that means a lot of new potential new data sources. So trying to balance like where do we bring data in Where do we spend time optimizing Do we double down on kind of, you know, with the engineers, analytics and data engineers, I do have, do we double down on tuning existing workload Do we focus more on doing a good job with new workload coming in right now Like how do we balance that And again, trying to connect that back to the business value so that we’re not spending valuable, valuable team resources in areas where we’re not gonna get the, the appropriate return. I do think that’s a more general problem, is there aren’t that many people out there that actually really understand this calculation and how to do this because it is a relatively new discipline.

Mark Stange-Tregear: And frankly, you need someone, whether you call it data engineering, analytics engineering, there’s other kind of terms out there. You need to find people who can understand the business use case as well as the technical side. Yeah. Because if you’ve just got someone who understands the technical side and is very focused on the technical side of things, it’s incredibly easy for them to go down relatively low value paths. It may be in Yes, okay. You tune that job, it totally wasn’t worth the time. Yeah. Right. So hiring the right amount of talent and then finding talent who can actually think comprehensively enough to truly do a good job of optimizing and not just sort of fix the technical problem is that’s an interesting challenge. I, I, I’m very fortunate. I work with a very strong team right now. but this is not an easy, this is not an easy fit. That’s another kind of key, key key challenge.

Elif Tutuki: I think you made an interesting, I’m seeing this trend, you know, all over the places. Like it’s just, you know, having the right mindset that understand the business and the value. And then tying back to technology like this is, you know, when it comes to also, you know, making, you know, better use of analytics, that’s why, you know, the businesses now is approaching more, more with a hub and spoke approach. Like there is no on only one central data team mm-hmm. , but it’s all about how we can enable, how they can enable, and also how, you know, providers like AtScale can provide solutions to enable the business units with that innovative data product, creation, but then without hurting the economic side of things as well. yeah.

Mark Stange-Tregear: Well, and I think that that’s a really tough balance. I mean, the, the hub and spoke model, I don’t actually subscribe to, we have a central team, data team at Babylist, and I think that that’s, the, the , there’s definitely different ways to go. What I’ve found, is one of the challenges in the data space, just more generally is the notion of career paths and development and management. I see a lot of analysts, you know, we, I hire, so I talk to a lot of people through recruiting process or just kind of peer to peer. And one of the, one of the incredibly common themes in the data space, and this is true for data engineers, analytics engineers, but it’s much more true for data scientists or analysts, is they get stuck

Elif Tutuki: Uhhuh

Mark Stange-Tregear: , right If, if you are a sales analyst who’s working and you are basically the only sales analyst, you are working on a team with salespeople, you’re reporting into a sales manager who knows nothing much about analytics or, or data more generally, how does your career go grow It can, right But how are you getting the, how are you getting these new skill sets How are you understanding things like finops and how that’s supposed to work Like’s a real challenge. So my preference, a strong preference is for, to actually have a central data organization to focus on kind of the mentoring, the training, the, and then, you know, have people work very closely in a cross-functional way with the business units and, but structure it so that there’s a more, more clear kind of center of excellence type type approach.

Elif Tutuki: Yeah. Right. And that is great. Like, I mean, I think if, if it is working for the organization, then that is awesome just overall, as you have said, like as, data roles or the people who are in the data roles as they grow in their career, like they become more business facing anyway. This is the kind of how they can grow. And, and that is kind of like, and also understanding these new concepts like finops mm-hmm. Just to kind of help them grow. I think, we are getting close to top, like, just, maybe final, thoughts like as you go through this journey, like, do you have any, you know, final suggestion to the audience, if they are starting new to their finops journey,

Mark Stange-Tregear: starting new, I think that goes back to the earlier question you asked as well. I mean, one, the fir probably the first thing is do you actually understand how much you’re spending Like again, talking about data finops, do you actually understand how much you’re spending on, on your data infrastructure, right Yeah. Including what component of your a w s costs or Google Cloud cost, is in that. If you don’t, then probably the very third, first thing you need to do is to actually understand that, whether that’s going back to whatever degree you can do and you may not be able to get it precise, go back and then go back to the finance team and take a look at what percentage of your total company cost that is. And just at a very high level, make a decision with kind of the exec team, with the leadership, with finance, whoever the appropriate person is in the organization.

Mark Stange-Tregear: Like whether this is worth, whether this is big enough that’s even worth caring about, because it may not be, you may be small enough and lean enough that it is not a good use of mind share. Right. if, if it, if it is big enough, if it’s, if it’s actually getting up there, it’s a significant cost with a company and you decide that you do want to get into finops, then there’s a couple of, there’s a couple of key things to do to start with, and then you can sort of, you can figure out the details from there. Often one is, who owns this Right The way I sometimes put it with my team is who’s getting promoted if it goes well and who’s getting in, who’s having a rough review if it goes badly, right Who is that person And you need a person who’s thinking about this.

Elif Tutuki: That’s great. Yep.

Mark Stange-Tregear: Even if you are a team. The second thing is, what are your goals I’ve virtually seen this, I’ve talked to people and I’ve seen people talking about this where they’re like, yeah, no, we’re doing finops. And what they’re actually doing is they’re just watching the data, which is definitely part of it. But, when you ask the question like, well, what’s your goal Like, oh, well how do you know if you’re doing a good job or not If you don’t know what your goal is, basically you’re just watching the information scroll past. So the other thing that is like, what’s your target Like if you’re currently spending a million dollars, is your goal to maintain that Is it No, we’re comfortable spending more than that. If we can prove the value, is it No, we need to cut 50% of that spend. Like what is it Then maybe different situation is gonna give you different goals, but if you don’t establish goals, you are playing at this. Yeah. You need to know what that is.

Elif Tutuki: That’s great. and I think just Mark, overall, you have mentioned this, like maybe as part of this is thinking about optimization. ’cause when you said, you know, goal, I was thinking, oh, is it cutting back No. And you have mentioned like the goal could be increasing that if you’re getting the value, so Yeah. Yeah.

Mark Stange-Tregear: I think, I think the reality is for most people it starts with a cutting phase because if you haven’t spent a lot of time on this, but it’s big enough as part of the organization that’s worth spending time on, there’s almost certainly probably a good amount of waste in the system. Yeah. Right So just generally, if this is becoming an issue for you and you and you are new to it, you are probably wasting money in there. So it is very natural to start with just cutting out the waste. And you can often do that pretty quickly. At a certain point you are gonna reach where it’s like, hold on, is it really worth spending like three weeks tuning this one job And you may be right that it’s not. And at that point then you start to invert a little bit. Now you’ve got all the, the low hanging fruit and it’s really start to think about, well, what, how do we optimize versus just cutting

Elif Tutuki: Perfect. Well Mark, thank you so much. Like thank you for sharing all your experiences, and then, you know, what you have learned during your job, journey with finops. So we really appreciate and I hope it was a, a great, session for the audience as well. So thank you for joining.

Mark Stange-Tregear: Absolutely. Thank you.