How To Invest in Data Infrastructure and Data Projects That Create Business Value with Matthew Kelliher-Gibson of Rudderstack

Eric Dodds 00:05
Welcome to The Data Stack Show. Each week we explore the world of data by talking to the people shaping its future. You’ll learn about new data technology and trends and how data teams and processes are run at top companies. The Data Stack Show is brought to you by RudderStack, the CDP for developers. You can learn more at RudderStack.com. Welcome back to The Data Stack Show. We are here with Matt Keller, her Gibson, and we have so many exciting things to talk about. Matt, welcome to the show.

Matthew Kelliher-Gibson 00:34
Thanks for having me.

Eric Dodds 00:35
All right. Well, I have the privilege of working with you every day. So I know a lot about you, but our guests don’t. So give us your background, or your brief background and tell us what you do today.

Matthew Kelliher-Gibson 00:47
So I’ve been in data and data science for a little over 10 years, I started as an analyst, data scientist, I managed data science teams, including an even ran the data science function at a public company before coming to RudderStack. And now I am a Technical Product Marketing Manager here.

Eric Dodds 01:08
We’ll definitely have to talk more about that career transition, because I think that’s definitely an interesting one. Well, very excited to chat today. Matt, one of the

John Wessel 01:19
Things we talked about before the show were driving business value with data. So I’d love to hear more about that. And then whatever other topics you want to cover. Yeah,

Matthew Kelliher-Gibson 01:28
I think that’d be a great one. I think also just looking at, you know, why it’s what the process is like trying to buy data tools and kind of why it’s so hard and what you can do about it. I think that’ll be another great topic to dig into.

Eric Dodds 01:43
Yeah, definitely. All right. Well, let’s dig in and talk. I think the business side of the conversation is going to be really interesting. So let’s dig in. Matt, there’s so many topics we want to cover today, especially around sort of the business side of working in data. But first, how did you get into data and actually even know, we work together? But I don’t even know if I’ve asked you this question. So I apologize. Yes.

Matthew Kelliher-Gibson 02:13
So it’s a little bit of a roundabout story. You know, the funny thing is they teach you like when you come out of grad school, but always have been very linear, everything has been leading to this moment and getting this job when you’re in a job interview. And that’s not at all what it’s like in the real world. So I actually started off. I went to my undergrad in sports management, which is a business degree. Really, I was going to try to work in Major League Baseball in the front office. But I graduated at eight when the entire world fell apart. So I almost got a job with the Seattle Mariners. I interviewed for a few of their internships that next year went to spring training, got 209. And there was nothing there. Like it was just dead. So I was like, I’m not doing this, because trying to work in sports actually costs you money. Because we meet so much. There’s so much demand to want to work there. Like minor league teams are like, we will pay you nothing. And by the way, you have to pay for college to get credit so that we don’t have to give you any benefits or anything like that. It’s I mean, it’s just, it’s all run on the back of unpaid labor. It’s an interesting thing. So I did that. And then I weirdly spent a year actually in politics. So I ran at 24. I ran for school committee in my hometown. And that’s a longer complicated story. But I ended up not winning, but I lost to the chairman of the board by less than 200 votes. So I came very close, which actually led to me running a state Senate campaign, which once again, we didn’t win, but it was a first time candidate. And he got about 40% of the vote, which was very impressive at the time. For comparison, two years later, someone else ran in a more competitive district in my town and got 28% of the vote. So I feel like I’m pretty well there. That rolled into the decision to go back and get my MBA, particularly because I had an old professor who was like, if you’re gonna go to grad school, that’s the degree that will have the most currency in the long term. So but my one condition with that was I needed to go somewhere where it was going to be hard. Because otherwise I would get lazy because that’s what happened during my undergrad. So I originally only applied to places where I could do all majors because I needed wow really hard. So I eventually got into grad school at a university called Bentley University just outside of Boston. Then in the Waltham area, which is like a well regarded, like regional business school, when I came in there was a MBA masters of finance dual degree. And then I went, they had a concentration where you go and talk to everybody and the different advisors and the guy was like, yeah, there’s no jobs in finance. And we were talking and he’s like, You should really go talk to the business analytics, concentration advisor. I was this little French woman, who was one of the math professors there. And she was describing what business analytics was because there wasn’t a lot of good language around this field back then. It was kind of like one of those sorts like wheat suits, like, all the parts I like about some of these other majors, but like, none of the bad parts, because it was like, you know, accounting is interesting until you get to like Philo Lila, all that stuff. And I lose interest. Right? Like, yeah, finance was kind of interesting. And again, do a lot of balancing stuff. And I just didn’t really care. And this was like, just the numbers part I like to do. Yeah. So I ended up switching to being just an MBA with a business analytics focus, took a bunch of statistics courses, what was essentially a SQL course and like a PhD overview of machine learning. So we had to read papers and teach them to the whole class, so that was interesting. Wow. But my assignment was actually neural nets. So I got a pretty good foundational understanding of them back in, like, 2012. Oh,

Eric Dodds 06:33
wow. Okay. Yeah. I have to start to interject really quickly. But so major league baseball, and politics, like, I feel like both of those while like running for a race, you know, or like running a campaign, I guess, running yourself. But like, Were you already predisposed to statistics, I mean, Major League Baseball and then like, running a political campaign, to me feel extremely predisposed to statistics, because both of those, you know, have huge statistical underpinnings.

Matthew Kelliher-Gibson 07:13
Yes, so I’ve always been predisposed to math, so to speak, I actually wasn’t huge into statistics when I was trying to work in baseball, because this was like, post moneyball. Everyone was trying to do that. So I tried to be a little bit of a jack of all and kind of ended up being you know, I didn’t have something to hang my hat on. But then, when I was actually running the state Senate campaign, I had access to the state voter Vault, which was run by the party, which had all this information on who voted and if they had answered previous, like phone surveys and stuff of affiliations and things like that. And I actually, just using Excel I had made, like, I tracked previous elections. And based on that, I was able to make some predictions of where I thought turnout was gonna be, which the interesting thing was, every like, like, politically oriented person in the district I was in was like, oh, no, that’s not gonna be the case. Because we had just had, I was up in Massachusetts, and there had been a special Senate election earlier that year. And I was like, I feel like, you know, it’s 2010, it’s an off year, it’s going to be somewhere between like, oh, six, and like a special election, everyone’s like, no, it’s going to be a presidential election. It was like, smack dab between the two. So I got that one, pretty much, right. And I was able to then use it to basically be able to say, there was like, one district that was like, 40%, there’s one town, it was like, 40% of the whole district. And I’m like, if we don’t do at least like 40-45%, here, we have no chance of interest, which ends up being true. So that was kind of a thing, and then also kind of pushed me when I part of what pushed me when I went to get my MBA was because everybody thought they knew what drove voters. And they were all wrong. And then if you would talk to people who had run like stuff for the state party, they would tell you a very different story, right? Like they actually knew more about what it was. But it just it put me down this path of thinking about how there’s like what people think works, and then there’s what like, reality actually reflects and data was a way to get to what that reflected reality was

Eric Dodds 09:21
super interesting. Okay, I love that we could keep going down that path, but we still have several years of story to cover. Okay, so yeah, you sent us your MBA, you did, you dug deep into neural nets, you know, 2012 So sort of at the beginning of mass modernization, like very early innings there, I guess he uses an appropriate analogy. What happens from there?

Matthew Kelliher-Gibson 09:50
So from there, I ended up taking a data analyst job with a marketing consulting company that was focused on things like direct marketing, which one of the big reasons I took was because I would actually get to build models right off the bat. So what kind of models, they were like response models. So think like, think like checking acquisition campaigns, right, we send you a letter, it says, if you open up an account, you keep at least a $500 balance, we’ll pay you $50 After. Interesting, okay. So that was kind of what I did. So I mean, I made like, over a dozen models in the one year I was there, including, I think, in my last month, I made three or four for Medicare Advantage companies, which was interesting. But that was also kind of like a crash course. And there’s school and then there’s like, the real world, because like, I mean, they were processes in place that did not work. And where there was like, they had these like, special proprietary, this is how we, you know, can project who’s likely to, to respond. And when you dug into it, it was actually like there was a ton of data leakage in it. So they actually weren’t that good. And then they had a layer on top of them with all sorts of filters and stuff like that. And the tough part was explaining to people because whenever you’re talking like, Hey, you’re taking future data, and you’re putting it in your training data. It’s very hard to explain that. But so I did a lot of that. I mean, I also that was my first time when it’s like, okay, we had a process to match customer files to our consumer file. It was like, alright, so where’s the standard? This is how we do it. And for the first time, but unfortunately, not the last, I got the answer of Well, everyone has their own version of it. And I instinctively was like, that doesn’t sound good. Yeah. So I became an early proponent of like, we should all have like, this seems to be standard. That way, if it fails, we all fail. And we all know it fails. And we don’t have these hidden failures to live in, which was a problem we did have there. But so I did that for a year, which was just, it was a good experience. Overall, it was, it was a lot of work. It was a consulting company. So you were working a lot of hours. From there, I went and worked for a company that did construction data. So it was like, kind of a tech company, but not really in a lot of ways. And I was there for five years, I was there first. I started in marketing. But then I moved over to what they call the data team, which was really like a research team. I was their first data scientist, that was also where I became a manager, and did some interesting stuff there. We did a lot of stuff around kind of like we were kind of almost in internal consulting the team I was on. So anytime people had issues that were like it with efficiency, or optimization or any thing kind of manual stuff, we would try to work to make it more automated and things like that, including, and we did some work on just helping move away from 100% like human research for like material like construction, material prices and stuff like that, which sounds probably a little more glamorous than what the actual results were we were dealing with were. So I was there for five years. From there, I went and worked for a subprime auto loan company where I was a senior manager there. And that was really, that was a big one because it was the first time I was not on the modeling team. I was managing a risk team. And it was a place that did automated decisioning for auto loans. So I did not make the risk model. But my team used it to determine what the pricing should be like, do we accept this deal? If we don’t, what do we counteroffer? Because it was all automated? So I owned all that production code, which funnily enough was all written in our Oh, wow, not necessarily something I would recommend. But it was an interesting experience. So I did a lot of work there. Like the big projects that I worked on there had to do with like, kind of optimizing a lot of how we our processes and how we got stuff done. I oversaw a complete rewrite of our production code, because it was just very interconnected, very spaghetti code ish. When we did a whole refactoring and rewrite of it, we found like 22 minor bugs and like 10 major bugs, including like a test that was giving people better pricing that it was like no one could figure out. No one was using the test. No one could remember when, like when they stopped using it. So they’d had a couple of years of getting better, like artificially better pricing for no apparent

John Wessel 14:47
reason that I have to ask you. Where did that start as far as running this, like production load? How when you came in, how was that being run? And then like what did you guys move to just from an infrastructure Random. First of all, I think that it’s always interesting to hear what people inherit.

Matthew Kelliher-Gibson 15:04
So right before I came, they’d started the process of moving to AWS. So I kind of came in right at the very beginning of like, they’ve just gotten, they’ve been using this like custom built system for all the pricing and loans. And I got in there, right as they were in house, rewriting everything into AWS. Right. So it’s all like Lambda functions, and you know, called processes and stuff like that. So by the time I was touching it, it was just, it was all in the cloud in AWS. During my time there, they actually did, probably the only successful, on time transition from on prem databases to a cloud database. Right? Because I remember, like, my second week, there was someone and they announced like, Okay, so we’re gonna we’re doing this project, and we’re going to migrate it over to Redshift. And we’re going to be done. I think it was like eight months. And I just thought, yeah, good luck with that. But amazingly, they actually pulled it off. So I even got finance to move over, which shocked me to no end. Yeah.

Eric Dodds 16:14
That’s very impressive.

Matthew Kelliher-Gibson 16:16
So we were doing that. I went from there, I had an opportunity to go be senior director of my title is Senior Director of Advanced Analytics, I was really like it was this data science for a public company. And that was interesting, because that was really where I went from. It was in aftercare, aftermarket auto care. And it was really like, this was a company that was like, we’re trying to transition to a lot more digital and data, we’ve got some stuff. And we really want to put our accelerated belief, honestly, we never got very far there were so many data issues. And it was really eye opening, also to just how I had always been able to attach what I was doing to business value. And this was like, the farthest I had ever been from it, and just how hard it was to try to push us in that direction. So I did that for about a year and a half, and then came to RudderStack. So just always a weird transition to explain to people. Yeah,

Eric Dodds 17:23
well, just give us the brief, just give us the brief explanation, because you went from, you know, a decade as a data practitioner, and to the product marketing role.

Matthew Kelliher-Gibson 17:36
Yeah, so there’s probably like two or three main things that were going on there. One of which was, I was just kind of getting burned out. I was, you know, I was in a fully remote role. And I had wanted that, so that I could have certain things in my life. And it was like, I looked back, I got about a year and a half an image back and I’m like, I sit at my desk in front of a screen on meetings for five hours. I have less freedom than when I had to go into an office. And it was just, I was just getting into like just a certain amount of burnout with that, too. Another reason was, so I’ve done kind of like I’ve built dashboards, I’ve built tables. I’ve kind of done a little like, parts of everything and kind of the data ecosphere. But I don’t have a formal, like, I’ve never had the title of a data engineer, even though I’ve done some of that stuff. And going into work for a company that makes tools for data engineering, I thought would also be kind of a good step to round out some of my experience. And then finally, you showed me, I remember when we were talking before, like I officially even like, like kind of applied and you were showing me like kind of profiles and stuff. And I was thinking this is the tool I needed two years ago. It’s like if I can help get this out here, I will help do this.

Eric Dodds 19:08
Yeah, I love it. Man, so many things to talk about. I do want to return a little bit later in the show to your question about increasing distance from business value, right and use like a, like working in data. And that was really difficult in your last year before you came to RudderStack. But I’m gonna ask a very leading question, to dig into a topic that I’d love to get your insight on. What is your job? I guess actually, I mean, I’m somewhat curious on a personal level, because I know you nerd out on software a bunch like I do. But in terms of your role as data practitioner, Data Team Leader function owner, what’s the best Most software purchase decision you’ve made in your career, like when you look back, and you’re like, that was awesome. It made a huge difference. Like both for the company like it was good for you optically, like in your career, does anything stick out or any like one tool sticks out?

Matthew Kelliher-Gibson 20:18
Well, first thing you got to realize is that for the first five, six years, I had no paid tools, no one was giving me paid tools. Everything was open source. And I remember trying to get a license for something that was gonna cost like 250 per person per year. And they were like, I don’t know about that. Like, that was hard. They

John Wessel 20:47
think about how big a company is out of curiosity.

Matthew Kelliher-Gibson 20:50
People are revenue. Are you there? It’s like $100 million revenue.

Eric Dodds 20:57
Not a startup, not like a little startup.

Matthew Kelliher-Gibson 21:02
Not a startup? Not a start. Well,

Eric Dodds 21:04
yeah, I mean, that really, actually, that’s such a good question, John. Cuz that really puts it in perspective. Yeah. I mean, not conscious, but like at 100 million in revenue, you know, a couple 100 bucks a year for a seat seems?

Matthew Kelliher-Gibson 21:21
Yeah. And then there was the internal frustration from it, because it felt like HR rolled out some new platform every year. And we’re like, how did they have money for this? And we got to reusing, like, R and Python? And that’s it? Well,

Eric Dodds 21:36
actually, so stop there, like and John, you probably have questions about your TV, like, Why do you think that was cultural? Right? Because I mean, of course, we talked about on the show all the time, like, it’s second nature to us that you would like to invest in data, trying to make things more efficient, but like, what was the dynamic at that company where you think that was the case?

Matthew Kelliher-Gibson 21:54
Well, I think part of it was that they kept investing in different tools that were supposed to do, like, you know, like training and like, you know, development and stuff like that. None of them actually really accomplished much of that. I mean, it was always, it was the classic problem, that kind of like, one of the things I always tell people is like, there are no tech problems. They’re all people in process problems. And these were like, VP of HR coming in and being like, well, we’re going to solve all of our like, you know, internal development with this platform. Lad didn’t work. Okay, well, we’re going to do it, we’re going to change our quarterly reviews and our process of checking in because we’re gonna buy this tool. And we’re going to tell everyone they check in four times a year, nobody checks in four times a year, right? Or they put in like, talk to so and so things going well submit, HR Get off my back. But I think there’s this idea of like, we can fix this problem if we just get the right tool. And it’s like, it’s not a tool problem. It’s a process problem. It’s wrong. That, yeah,

John Wessel 22:54
That was my experience with Eric. And one of the things one of my memories is with my first data job, is they went and spent over a million dollars on this fairly advanced at the time custom analytics system it was for, for a contact center. And it was like, well, this tool is supposed to do XYZ, so no blanks. So then anything outside of that as well, we bought a tool to do that. And then any explanation about, like, why the tool didn’t work, or why, you know, it was like, No, we bought a tool to do that. Use that tool. Right. So I think there’s that kind of a sunk cost fallacy that comes into play, where it’s like, Don’t we already have the fill in the blank of the ERP system? Or do we already have Salesforce? You know, I think that’s a component of it. And it’s hard for data team members to articulate why they need it when there’s that perceived overlap with something that you already have, or with like, say, an open source tool, like, like Matt was saying, Yeah,

Matthew Kelliher-Gibson 23:59
I think there’s also like, if you go back to like, 2012-2013, there was this little narrative of like, oh, look, you can leverage all of this data, and it’s going to be cheap and efficient to do it. And like, once you started digging into it, you found out like, Okay, well, no, once we’ve kind of done analyses after we’ve hit your database enough time, and we need to actually scale stuff out. Like this is not a no cost thing. Yeah. But there was kind of a narrative of like, oh, look, if you can get a data scientist in here, they’ll just work magic.

Eric Dodds 24:30
Yeah. Yeah. Interesting. Okay, so the first five years you’re using open source tooling, it’s really hard to get a budget. Did you get a budget at some point?

Matthew Kelliher-Gibson 24:46
Kind of like how I just shared a budget with the rest of the data. Org. So it was a little nebulous, exactly. Who had what at times, unless you got a line item epic that when we were doing our annual budget, and so that was your time. And if you wanted something, you could get it. Then probably we go back to like, what was the, like, the best purchase I made? It was, we were trying to, we needed a way to kind of like productionize. Some machine learning models that we were building, and we’re anticipating building. And we were looking at platforms, we looked at a couple of them. And there were some good ones, there were some not so good ones, but it is going to be so hard to get it in because they weren’t vendors, there’s gonna be a new process, we had a new CTO who was putting in any process and so I ended up being on GCP. And I went back to Google, it was kind of like, Alright, fine, I’ll listen to what you guys have to say, because I was kind of dismissive of them at first. And we ended up using their vertex AI platform, which was like, we basically met with them and made the decision to kind of go with it. And we started like, the next day. And within 30 days, we had productionize, a major model that was used by one of the one of the major business units at the time. So that was probably one of the biggest ones, because it’s pretty sure it’s still being used there to now like it still runs every day, and still informs a lot of the decision making for them.

Eric Dodds 26:18
And why were you dismissive initially, like because you had done your own research and like you wanted dedicated, like, sort of ml ops type platform, or

Matthew Kelliher-Gibson 26:29
it was a part of that. It was also like, I felt like every time I talked with them, they were trying to sell me on something new, like, oh, you can use this service. You just get tired of hearing that, like, you know, we have a weekly check in and all you’re doing is trying to tell me about these other things and asking me about that type of stuff. I think it was also because it was really unclear. Like what this was, if you looked at Google’s documentation at the time, they didn’t have good documentation for vertex. And all the stuff they did have was basically Well, here’s how you stitch seven GCP systems together. You’re out. And it’s like, right, I’m not doing that. Yeah, nope. Hard pass.

Eric Dodds 27:12
Yeah, I mean, yeah, I mean, there’s some great things about, like, GCP platform, but that is sort of a, you know, I remember being part of a startup that was like, building something on top of Google ads. And it was like, when we really got into the API’s It was shocking is like, wow, there’s like, teams who are building parts of Google ads that are literally completely separate. And like these parts of the same system, literally don’t even talk to each other. Yeah.

Matthew Kelliher-Gibson 27:40
Yeah, it was. There was a lot of stuff. And they were to their credit, they were building it up really fast. But the problem was, there was none of their kind of documentation and even like the people who were supposed to be helping you, they quickly couldn’t keep up with the speed. They were developing the platform.

Eric Dodds 27:59
Yeah, so. Yep. Okay, that. So let’s dig into why I’d love your thoughts on why buying data tooling can be difficult, right? I mean, I think it’s worth acknowledging it at some companies is probably less difficult, right? Depending on the context, right? If you’re a startup and, you know, 2020, with a really good growth rate, like buying tooling was probably extremely easy, right. But it’s getting hard again, I think, for a lot of companies and a lot of data teams, no matter what type of company you work out, but like, what are the dynamics that make that difficult?

Matthew Kelliher-Gibson 28:46
Yes, I think talking to your point about it gets, it’s easier in some instances, and harder. And others there is this point at which there’s kind of a push to be more kind of like formal and professionalized, usually hits you know, it has to do with either like cost controls, or it has to do with, like, security and making sure that you’re doing that because, you know, there’s a point where you’ll hit where suddenly, there’ll be this corporate thing of like, we don’t want to use a lot of SaaS platforms, there’s too many security risks with it, stuff like that. So I think that’s part of it. You just, there’s a lot of kind of, for lack of better term like bureaucracy that goes into it, too. There’s a lot of choices like just a lot of checkpoints. Are they an approved vendor? If they’re not, okay, what’s the process we have to go through for that we had to fill out a form this whole like, set of forms they got to do, it’s then got to go back to your InfoSec team, you have to do that. And it just becomes this thing of like, do I want to spend the time to do it, right. So that becomes one part of it. The other part is, like, you know, I’ve been at multiple places where either from a financial point of view or or from a PE investor point of view, they started looking at it and one of their metrics was, how many licenses do you have per employee. And they had a goal, interest down to a certain number, like that was a way that they can attract or divest. So, especially when they were getting ready to sell the number of vendors and the number of licenses per employee was a big deal to them.

Eric Dodds 30:24
Like across the entire spectrum of software, you know, like Salesforce data, whatever, like the whole thing, whatever it was,

John Wessel 30:34
and not dollars, like we’re just talking number of licenses, regardless of dollar value, right?

Eric Dodds 30:39
What a fascinating metric. I’ve never even thought about that.

Matthew Kelliher-Gibson 30:44
Yeah, it was literally for a good reason. Yeah. I mean, it was one of those things where, you know, you’re partially cleaning up, like, hey, we have 20 licenses, and only five people use this, why are we paying 20, right, you’ve got some of that going on. There’s also just kind of the feeling of like, the more vendors we have, the more overhead it takes to kind of manage all this stuff, the more kind of risk you’re exposing yourself to, because quite frankly, it’s hard to keep track of where it all is. I mean, that’s the other thing is you get all of this, you give autonomy to like every department to buy all of their own technology, you end up in the state where you’re like, well limit, you know, where it or where security, we didn’t know that people were using this, that or the other. Yeah. And so there’s this move of fewer vendors, and fewer licenses. I think it’s also because it’s kind of a proxy for like, what’s kind of your ongoing costs for technology per person? It’s going to be an easy proxy for that too. Yeah,

Eric Dodds 31:42
yeah. I mean, yeah, it’s, I don’t know, I’m not an expert in like, you know, PE, like cost efficiency, but you can see how it would serve as a proxy for that, right, like, under utilization is certainly a problem you’re paying for, you know, half the seats, and you’re not using them. I guess the thing that gave me pause is that pricing models are so fragmented across SAS companies that that can certainly only give you part of the picture. But that is super interesting. How much conviction? Do you need to have John? And I actually, well, I have a question for you, John, after this, but like, so how much conviction? Do you need to have it in order to go through that process? Right? Because it’s not like you have all this extra margin that you can spend navigating all of the, you know, your procurement process and all those sorts of things like, you know, because if to some extent you have to go convince the company, which creates risk for you. Yeah, right. Like you want to go spend money, you want to like, allocate more budget, like, those are risky things for you to do yourself from a career standpoint.

Matthew Kelliher-Gibson 32:53
Yeah, and especially if it’s going to have a long implementation, like the longer the are more complicated, the implementation that riskier, it’s going to be I mean, I’ve seen, you see several of them. And I think you learn quickly in large companies where it’s like, Yeah, that guy, put himself out there, and, you know, was going to buy whatever thing and it was going to have a five month implementation schedule, and we’re now on month 12. And we still don’t have anything. And that’s usually when that person takes a job somewhere else before it gets to be too late. And so it becomes a thing you kind of internalize like, Well, I’m not doing that. So I think part of it is, yeah, you talked about, you gotta have conviction in it, it’s got to be something that like, you are like, we cannot live without this, it can’t just be that it’s going to make your life a little bit better. It’s kind of like it has to make your life like five to 10x. Better

Eric Dodds 33:49
interesting.

Matthew Kelliher-Gibson 33:53
People will put up with a lot to not have to go through that order not to put the risk out there for it. Yeah, when I was running the data science function, we had a, like a 200 grand line item for a graph database. They were like, tell us when you think it could be useful, and you know, will, they’ll use it, and I never used it. Because I was like, I am not paying for this, when we’re like, there is no chance we will be able to fully utilize this, I’d have to train people and how to use it. We don’t have a good use case for it yet. And it was like that money was already allocated, and I would not use it.

Eric Dodds 34:30
Interesting, like, have you already purchased? Like, have you already procured a graph database or it was just sitting out there someone had,

Matthew Kelliher-Gibson 34:38
it was a budget line item like, you know, 200 grand graph database, and it was like, Well, when you’re ready, you can go and procure it, I was right.

Eric Dodds 34:47
But you don’t want to Yeah, you don’t want to hitch your wagon to that. And then you have to justify 200 grams of database spend. You know,

Matthew Kelliher-Gibson 34:56
they’re like, Well, what are we getting out of this? And it’s like, oh, well, you know, we’ve started training people in the query language. And it’s totally going to be useful next year. So yeah, I’m not doing that. Yeah,

Eric Dodds 35:09
That’s super interesting. John, you have had the CTO role before. And did you control all the tech budget?

John Wessel 35:20
Yeah, I did. We had some small things, like you mentioned, like HR systems, like a lot of companies have those small systems. But yeah, the vast majority of it, we had centralized into a one tech budget.

Eric Dodds 35:35
And how did you like it? What was your experience with this? Right? I mean, was it so hard to buy software? data infrastructure? Sorry, software is a really broad term. Sure,

John Wessel 35:47
it was right. It wasn’t a huge company. So there weren’t a lot of really, like, drawn out formal processes for the procurement. But each purchase was like me as an investor, like putting money into something and like, this better go up in value. If you make enough of those, like it, like hiring people, right, like each hire and each procurement, like both of these need to produce value, or like, eventually you’re done. If that doesn’t happen long enough. Probably my riskiest one that turned out really well, as we were on a custom ecommerce platform that has been developed over a lot of years. I think they had upwards of 80,000 pages out there on Google, many of them were hand coded HTML, lots of effort over like a day should have gotten onto this site, lots of legacy SEO. And for various reasons, security reasons, scalability reasons, etc. I decided to move to Shopify. This was seven years ago. So that was like, I felt good about Shopify. But that was a super high risk thing. Because if, you know, something had happened with Shopify, like, well, I can’t do this crucial thing. Or maybe Shopify is having scaled problems, and not as stable as we hope works. There’s something that happened along the way. I mean, that would have been a huge deal. And to be honest, there were definitely some bumps in the road as far as SEO with that money, you know, pages out there, that from the legacy app, but all that eventually got worked out. But I mean, it was a huge deal. And I can think of several other examples where like, I’m investing in this technology, like being good, and this working, and then continuing continually improving, especially if it’s a newer technology, and not getting bought and then killed or stuck for whatever reason. And

Eric Dodds 37:50
startups are an actor, right? Like, right $20 This company, and if they run out of, you know, VC capital.

Matthew Kelliher-Gibson 37:59
So even rather than have to deal with, like, you know, you get someone on your team, and they’re either like, I don’t know why we can’t just buy whatever, or they pitch you. I think we should do this, and you kind of have to be like, I hear you, but no.

John Wessel 38:15
Yeah, all the time. Yeah. Yeah. You know, I especially don’t want to take on marketing people, but especially marketing. That will come up a lot, because this

38:27
is the right, this is the right venue to do that. This is the right venue. I know.

John Wessel 38:32
And I’ve been involved in marketing groups, I lead a marketing team for time. So I ended up with some marketing, for sure. But yeah, especially in the marketing side, there’s always a new tool right of the month that somebody was excited about sometimes for good reason. Yeah, but working through that, like progression with people like okay, and all that was like working through like what I’m going to have to go do with our CFO of like, okay, what are we going to use it for? What’s the expected ROI? Do we have tools that already do this are the free tools that can do this, you know, just working through that progression with people usually ends up in a fine place. I didn’t have too many situations where somebody who was just so diehard, like, we have to use this email tool or whatever tool and like nothing else will do. I don’t remember too many situations like that. But after you got through that kind of progression of justification and value creation, people, people typically got it.

Matthew Kelliher-Gibson 39:31
Because they I know at least on some light teams, you get some frustration when they’re like but my life would be easier if we could have whatever. And it’d be like I hear your life could be easier. But if I could just flip a switch, that’d be great. But you’re talking about a three four month process that may or may not end up successfully and that I now have to invest political capital into right once I burn it, it’s gone.

Eric Dodds 39:56
Right. John, before the show, we were talking about things with Matt a little bit about, like business value that you even mentioned, like distance from business value. What’s would love to know what questions you had around that specifically, because I think that’s really in a cost cutting environment, which we are in right now. That question is way more, you know, there’s way less appetite from the CFO to gloss over things that, you know, maybe we’re, maybe we’re like, sort of drawing the story out a little bit on ROI for this data infrastructure? And

John Wessel 40:33
yeah, I think, Matt, one, one piece of that I’d love to hear your perspective on is you’ve got data teams in this cost cutting environment right now that I think a lot of them are scared, right? Like they’ve they like you mentioned earlier in our conversation about that kind of data science Savior mentality of like, you’ve got all this data, like hire data scientists, the bill, you know, fix all of your problems, create all these predictive models. And now you’ve got like, you know, various AI things. I think people are past that. I mean, I’ve even, there’s a lot of people out there calling them like, recovering data scientist is like a thing I’ve seen out there. So now we’ve gotten past that, like data science, these data, scientists are kind of going to save the day. Now, I think, because that hasn’t materialized and lived up to some of the hype, you’ve got all these systems that were installed and put in place and a lot of, you know, investment in infrastructure and things like that. It’s like, we have to get business value out of this, or we’re going to kill some of these programs, we’re going to, you know, downsize some of these initiatives. So like, tell me about your experience with that. And like, what are some tips if you’re somebody in that situation? Like, what, what can they do, to try to, you know, connect the data to business value. And you wouldn’t know for everybody’s circumstance, but you know, speak to someone in your past circumstances.

Matthew Kelliher-Gibson 42:03
So I think one thing to remember is, there’s two ways you’re going to generate business value, right, you’re going to either hide costs, you’re going to like efficiency, something’s going to be cheaper to run, or you’re going to generate some type of new revenue, right? Whether it’s making a process, an existing process, be better, or if you’re going to add something, you know, sometimes, like on the marketing side, you can sometimes add in new processes that will generate more revenue and things like that. I think the biggest, one of the biggest things is to always start with that in mind when you’re doing things. So like, don’t talk about what you’re going to do in terms of like, oh, well, the data will look like this, or, you know, we’ll be able to, we’ll be able to be this much faster as a data team or whatever, it’s really going to come down to like, we will reduce customer churn, right, or we’re going to reduce the cost of acquisition, or we’re going to increase the number of leads or something like that, like, you’re gonna you’re gonna handle it in that sense. I think the other thing with that is, specifically kind of with the infrastructure side, where I’ve seen people get tied up is where you get kind of this push and pull, right, you’ve got one group that’s like, we want to build stuff that is like, it will handle this one thing that the business is yelling at me about. And it’ll fix it. But it’s, you can’t build off of it. It’s completely non scalable. It doesn’t fit in with anything else, right? And that’s when you get into like, we’ve got 1000 different pipelines that all do one thing, and it’s impossible to manage, because we’ve got no, there’s no organization around it, then you got this other group that’s kind of like the we need to build, like the grand unified data model for everything. And there is a middle ground there, there is an 80%, where it’s like, what are the biggest needs that the company has? What are the biggest things? What are the things that are making everybody’s life harder? Or like what are their metrics that they have to obtain? Right? Every department has some kind of financial goal they have to obtain, figuring just talking to them and saying, what is that and figuring out where you can tie that into is where it’s going to be. And a lot of times you can figure out, hey, if we build a more flexible process, for this 20% It will get us value, right? And it’ll be visible. And we can see it, we can talk to it. And it’ll leave us open to be able to build off of it afterwards. So I think that’s part of it. Like one way to think about it. I think the other thing is to remember that busyness is not a business value. So it’s very easy to get caught up in the idea of like, well, but they keep asking me to do things and I keep doing that. Yeah. And when they’re talking to each other, they’re saying, I don’t understand what we’re getting from this team. I don’t see any real value coming out of them. Because they’re not really like a lot of times they don’t know what they want. So they’re going to ask you to do stuff that’s at least visible, right? Can you pull this? Can you go grab that? Can you fix this? Can you do that? And you’re going to, it’s harder. But you’ve got to do more of that work to figure out like, well, what is it you actually? What are the big things you’re actually working on? What are your initiatives? What are the places that are? Harder? That shouldn’t be? Right? Where is it far that it shouldn’t be? Where are you struggling to get this? And trying to work into that? And unfortunately, it’s like, there’s not a good one size fits all answer, like it’s clearly do this. But you gotta be able to tie those things. And I mean, you know, thinking back to stuff, I’ve, you know, I worked with projects, where it was like we got to people in finance, who are just wrangling these worksheets, and they do this for a week, every month. And we turned that into a 10 second SQL query. Well, there’s like definitive value we can draw from that in a savings for human cost point of view. I’ve also done things like, you know, there’s a lot of models that I built, where it was like, hey, look, we increased the conversion rate XML that leads to X amount of dollars, things like that. But it’s really tying it into those types of ideas. And I would also say not trying to bite off too much, there’s a temptation to go big. And going big usually means you go home.

Eric Dodds 46:18
So like, but I see what you want. Yeah.

John Wessel 46:24
Good. So one of the things that I always came across was the business value. Question is, you’re right on, like getting in the weeds with the spreadsheets with the business users. But how do you do that? In a way that’s not threatening? Where can you get real answers? Because it’s a really delicate situation, sometimes where people throw up walls, people don’t like to be afraid of their jobs. It’s like, you’re gonna automate my job burn across that a lot, like in new strategies for people that are, they want to create the business value, but they really want to do it in a way where they’re, you know, not freaking people out and not losing their job.

Eric Dodds 47:02
Yeah, job automation. That’s super interesting.

Matthew Kelliher-Gibson 47:04
Yeah, I think part of that is, don’t come in and be like, Oh, this is selfish, we could just replace this with a SQL query. Right, like walls are gonna come right down. Right? Right. Like those walls are there, they’re not going anywhere at this point. I think a big thing with that is more, you gotta kind of come up alongside them. And so it’s really kind of getting an understanding of what their view is and what their pain points are. And that’s where I think like, if you talk to places, if you talk to people within a department, and you’re like, what is difficult? That shouldn’t be? Right, you’re kind of getting towards, like, you know, what they’re doing? Or it’s kind of like where, you know, you can even start with where you need help? The other thing I would say is that if you can, like places I’ve worked, my teams, we built up a reputation for like, we were good at getting stuff done. So we were typically brought in when things were going bad. Yeah. which frustrates the team to nail and because they’re like, Well, you know, if they would have brought us in at the beginning, this would be really simple. And I’m like, Yeah, I get that. But we’re going to kind of, we’re going to do more than we should, it’s not going to be the right way. But we’re gonna get them through this, and they’re gonna see how we’re helping them. And then they’re going to come to us next time, because they know that we’re here to help them. So it’s that it’s a people problem, you know, you have to go down to people, you’ve got to build up a kind of relational capital with them. And you need to build it up before you’re trying to kind of get them to do that one. I think that’s the other big thing. You got to go out there like this idea of I’m going to sit back and the businesses is going to come to me with all these kind of data projects like that doesn’t really exist in most places, you got to go find it, you got to go build relationships, you gotta build that trust with them. So that when they’re like, Hey, this is really hard. While we were talking to Matt’s team like John was really good at I, you know, I heard that John really helped this other group out, let’s go get them. Because everything you gotta realize is like, my first manager, Team Leader is this where it was like they announced it. We’ve got this team, they’re here to help you. It’s a center of excellence, blah, blah, blah. And both groups would go like, Oh, that’s great. I’m sure that people can eat it. We know how to do our job.

Eric Dodds 49:17
Yeah. Yeah. It

Matthew Kelliher-Gibson 49:22
was only when things were going sideways, that they’d be like, hey, all right. We really need help here. And we don’t and we’re willing to take it now. And those were the relationships you built moving forward?

John Wessel 49:34
Yep. Yeah. Yeah, that relational capital is really important. One of the things I found that was very effective, was trying to be working with analysts for different groups like financial analysts, marketing analyst, and being an enablement person for them, and caring about them helping them Yeah, helping them develop career wise, like if they wanted to learn a little bit of SQL if they warned and knows Some Excel tips or tricks, the more advanced stuff like most of them, especially in analyst type roles were pretty hungry for that type of thing. And that seemed to like work really well, as far as Yeah. When engaged.

Matthew Kelliher-Gibson 50:11
Analysts are a good kind of door into a lot of departments because they usually feel overworked, they usually don’t feel as supported as they should be. And a lot of times, when you’re coming in to help them, you’re not coming in saying, Hey, we’re going to automate your job away or something like that, we’re saying, we’re going to, we’re like, kind of get beside you. And either give you kind of like some tools or some knowledge, or we’re gonna help build something with you. It’s going to make your life better and lay your focus on things that you’re more interested in.

Eric Dodds 50:41
Yeah, super interesting, probably a good way to, like find not I mean, this is a good way to make enemies as well, but find potential people to poach onto your team, right? Like, someone who’s really. So

Matthew Kelliher-Gibson 50:55
simply. The opposite is also true. I always wanted my teams to be ones that were so competent that other departments wanted to poach them. Because every time they poach your people, and out there, you now have someone who’s in that department, who knows what’s going on, and knows your language and what you’re capable of. I mean, I had one instance where we embedded someone in another department. And it was like my knowledge of what was going on in that department and where they could actually use real help went up like 10 fold.

Eric Dodds 51:29
Yeah, I’m using really aggressive terms here. But that’s like the mole, you know, like you plant a mold. Okay, we’re close, it doesn’t matter. But Matt, what’s your so in your personal life, us, we nerd out about software all the time and productivity hacks or whatever? What’s your favorite piece of software or app or whatever that you just use that you use in your personal life?

Matthew Kelliher-Gibson 51:55
That’s a hard one. I mean, I try to not over optimize my personal life too much to be honest with you. So

Eric Dodds 52:02
That’s a wise, really wise disposition. Yeah, we

Matthew Kelliher-Gibson 52:07
get to be a bit too much with that. But I would say probably, and this is going to be really basic, I have been using the notion calendar app. And I have just because I can hook it up to I’ve got a different Google Calendar for like every person in my family so that I can kind of see everything there. And it’s worked better than kind of like the Google Calendar app of like, I can see that I can see work. I don’t feel like either one is trampling on the other interesting. He has a really nice feature. I have one calendar that’s called Work block, that if I need to put something on there that I’m like, Hey, I’m gonna be at the doctor. I can put it on my personal calendar, and it’ll automatically block my Work calendar for it too.

Eric Dodds 52:46
Nice. Oh, wow. And that’s like a notion feature. notion calendar. Yeah. Yeah. Very interesting. You know, what’s interesting to me about that bricks is gonna get mad at us for going a couple minutes on. But you know, what’s really interesting to me about that is that experience doesn’t map to a lot of things like marketing or the benefits of a market, which is basically that you can see, you know, your work notions like a knowledge repository, right? There’s all sorts of documentation and whatever. And bringing that together with a calendar is a very compelling value proposition. Yeah, that’s interesting. Maybe that’s because they acquired a company who did that? Well, under the hood already?

Matthew Kelliher-Gibson 53:24
Yeah. It’s basically they kind of rebranded that with, I think, some other internal improvements in there, but it has been helpful just to kind of get everything out there. Because, because, I mean, otherwise. I am one of those tech people that like, I have a notebook with me, I am writing within the a lot. I will not I kind of have this, like, digital

Eric Dodds 53:45
paper and pen, not like a Jupyter Notebook. Just to clarify.

Matthew Kelliher-Gibson 53:50
I don’t. I will, as both of you have probably heard me grant on there. I am not a Jupyter Notebook person here.

Eric Dodds 53:56
You are definitely. Just to clarify for the listeners.

Matthew Kelliher-Gibson 54:00
Yeah, no paper notebook. And actually, I use a fountain pen that someone who used to work under me gave me when I left the company. Yeah, I was very surprised when they gave it to me. I

Eric Dodds 54:12
gotta pay closer attention to your pretty cool script. All right, look, we went from buying enterprise software, to talking about analog, you know, writing within just a great narrative arc. And so with that, I think we’re gonna wrap this show, Matt, thanks so much. Man, what an insightful episode. And great insights from you and John on just the challenge of navigating an organization and acquiring and getting value. I think, more importantly, out of software, so thanks for the time. Thanks. Thanks for being here. We hope you enjoyed this episode of The Data Stack Show. Be sure to subscribe to your favorite podcast app to get notified about new episodes every week. We’d also love your feedback. You can email me, Eric Dodds, at eric@datastackshow.com. That’s E-R-I-C at datastackshow.com. The show is brought to you by RudderStack, the CDP for developers. Learn how to build a CDP on your data warehouse at RudderStack.com.