Skip to content

What do data scientists do at mobile app companies?

Data science is becoming an increasingly important part of a mobile app’s growth. But what exactly does a Data Scientist do at a mobile app company? How do they do to help mobile apps grow?

We sit down with AdLibertas Advisor and Data Scientist Andre Cohen who’s led data strategy for years for some of the most advanced mobile applications in the world to ask his take and experience.

Watch the full discussion below.

Audio Transcript Below

(lighly edited for length and clarity)

Adam Landis  0:03

Welcome Andre to a brand new blog post series that we’re going to be doing with videos. Those of you who don’t know, Andre, he’s recently signed on as an AdLibertas advisor. And he’s an old friend that goes way, way back. And we’re going to talk to him today about what exactly are data scientists doing at app companies? You’re a lot about data science in our industry. It’s a hot, sexy term, but what does it actually mean and what are companies actually doing? So welcome, Andre. Thank you for joining us.

 

André Cohen  0:39

No, thank you, Adam. It’s great to join you guys. It’s great, because we were once at the same time for a brief moment, both having startups. So we have like a lot of feelings about being in the trenches of startups and building a company and the product and data and mobile space.

 

Adam Landis  0:58

Sure didn’t seem like a brief moment It seemed like a long time.

 

André Cohen  1:02

Well, yeah, it was a long time. But it was also a long time ago for me now. And, yeah, it’s great to participate now, from a different perspective.

 

Adam Landis  1:12

A little bit, before we jump into that, a “different perspective,” why are we talking to you about data science — and for the people that don’t know you –, who are you? And what’s your background?

 

André Cohen  1:22

Sure. Yeah, my background is that I was an academic for a long time. I was doing my PhD in machine learning. And that was way before data science was a term and even the iPhone was a thing. But halfway down the track, I decided that, you know, be cool to be an entrepreneur for a while and see how is this to build a product, and so on, and so forth. So I started out randomly, almost making iPhone games. I had two titles. We never got everything right. At the same time for the game, either we had the monetization, right, or we had the user acquisition, right. Or we had the game, right? For a matter of fact, however, we never got all three things together, which is a typical story in the game space. And then we, two years later, we’d switch to my company switched to building an analytics slash game optimization service. And that’s where I learned a lot more about data science in the applied world, specifically, also how to, you know, sell data science and how to package data science into a product. And that went well. That’s how I met Adam, also, between the gaming and heavy startup.

 

Adam Landis  2:37

And the startup, what did you guys do?

 

André Cohen  2:39

So Gondola was game optimization as a service. So our goal was to find things that could be optimized by a computer that would not require humans, and they often the traditionally was with a human. So an example of that is dynamic pricing. So how much do you pay for an IP in a game or an app? Right? Most people are have static numbers, you could technically sell for anything, a sword. And again, because it’s virtual, there’s no cost of producing swords in the game. So we’re doing these optimizations of you know, very quickly iterating through prices or numbers and finding the best number that works for the game with a user. The challenges with optimizations is a data science, this is like partially, the issue with data science is that you can have the right solution. It doesn’t mean that you have the the appetite from the customers for that kind of solution. Right? They’re–

 

Adam Landis  3:40

doing data science 5-10 years ago, it’s too early.

 

André Cohen  3:44

It was too early, definitely way way too early. But it also is a it’s just something that people have to get used to, right. It’s the same thing with like, self driving cars. Everyone likes the concept. But if you could go into a gas station and ask the gas station to rip out your steering wheel and install the self driving mode to your Toyota Prius, most people would freak out because you have lost all controls of your car from one minute to the next. And that’s kind of our service was it was “Hey, you have a game that’s functioning many nearly right now. Do you mind if we just rip out one of the key components to your car and automate it for you?” So the company eventually got acquired by Tilting Point two years ago. They’re a game publisher, so I got a sense of what it is like to to be on the other side of the fence making games and publishing names.

 

Adam Landis  4:40

So indie, service provider, big game publisher.

 

André Cohen  4:43

Exactly. And so that gave me a new appreciation for something I always overlooked, which is the data engineering and data platform side. When you own your own service or your own game, you don’t see how interconnected data is for all the different people in a company. But only when you get into a company that has a portfolio of apps, do you realize how integrated and how you need to integrate all the different data sources together. I always tell people when we were interviewing or when I interviewed people, you know, an app has an average 30 different API’s or services that it uses daily. And the the challenge in data is, how do you bring it all together?

 

Adam Landis  5:26

And we’ll get there in a minute. But what were you doing for Tilting Point?

 

André Cohen  5:30

So I was the head of data sciences.

 

Okay, from the the publisher, the game publisher, large, recognizable, you were the head of data sceince. So you have some authority in in the subject of what app developers are doing with data scientists?

 

Yeah, in many ways, I mean, it’s like, the data sciences, like, it’s very misunderstood. It’s probably the most misunderstood title. And it might even be a fashion.

 

it’s hard to understand, because I did not aim my career to be a data scientist, and don’t feel attached to the title in any way. So I suspect it will be a fashion that will go you know, become something and will disappear.

 

Adam Landis  6:12

Well, let’s let’s put a stake in the ground of what what is data science to you as you define it.

 

André Cohen  6:17

So data science is at the tip of a pyramid of things that you need to have in an organization. And it doesn’t mean that it’s in the tip, because the most important one is just, there’s a structure– it’s a corner of it. It’s at the very bottom, you have data, and you have analytics as a second layer, data science sits on top of that, meaning that it depends on everything below in order to operate. So if you think about what does it take to build a company in the app world or having an app, data science is probably one of the very last hires to do, because it’s just every other position is necessary ahead of before it. You need an app, you need data, you need to ingest the data, you need to clean up the data, you probably need an analyst before even, you need someone to count the revenue , so you need an accountant. Data science is very, very much at the end of the tail.

 

Adam Landis  7:18

So break those apart. So you said this in the past, but I thought was super interesting. I’d like to bring up. So data analysts, data engineering, data science, what is the difference between those roles?

 

André Cohen  7:30

Yeah, I mean, one, just to go back to the previous topic, maybe it’s a different way of explaining because the pyramid is a little abstract is data science is also another vertical of revenue generation in a company. That’s how I like when you are ready for data science. It’s saying, okay, we have the app, we have an app to monetize this great, then you then that’s a vertical, right of a company, perhaps the second one might be Oh, we added ads now Oh, okay, ad monetization is another vertical. You could if, if the app is big enough, you can say oh, we have subscription, there’s a whole subscription team specialized in like managing subscriptions, but that’s another vertical. And you could even like if you’re in the in the game space, have an IP vertical, which is like, Oh, we go to Hollywood, we find great movies, and we put them into their game. That’s a vertical like it makes money by the sheer fact that is walking dead or a Star Trek game. You don’t need to have the best game to make money from a Star Trek game, you just need that license. And then there’s data science, that’s what I would claim. data sciences specializes in extracting value out of data that’s sitting there in ways that other people by themselves couldn’t. So you’re depending on algorithms and methods from academia, usually, to find new things in that we’re just there, right? It’s like finding gold in a haystack, which is why it’s so difficult.

 

Adam Landis  8:54

So data science is the process of finding answers out of data?

 

André Cohen  9:01

Yes.

 

Adam Landis  9:03

And how does that relate to a data analyst?

 

André Cohen  9:07

So the difference I think, in Data Science from other positions that are similar, like BI analysts, or data analyst or any other analyst is. Data science is the career path and like the purpose of a data scientist is to create methods that extract value from data. Now, you can find data scientists that are very happy simply, you know, picking things off the shelf, using them and creating something out of it.

 

Adam Landis  9:43

When you say things, things off the shelf–

 

André Cohen  9:46

Let the simplest example is image recognition is a data science problem. Let’s say let’s say that’s what a real data scientist is gonna say. Well, we’re talking about an app That’s about dogs. So let’s not use an off the shelf method for finding dog faces in these photos, because it’s too difficult, like it’s not the best solution. The best solution is if I write a new method that takes into account the hair, the fact that the dogs have a, you know, a very specific kind of nose, it’s not a human nose, it’s a diagnose. There are different than cat noses, you don’t want to, you know, you want to probably filter those cats out of the pictures. And that’s what a data scientist does. He goes deep into trying to deconstruct what are the features in the data? And how to create a method from those features. Now, an analyst is really incentivizing getting answers today. So an analyst his his reward is like, Can I figure this out today? So it might be something like, well, how many? What’s the like, what’s revenue is a great common one, right? What is the revenue of this app? And let’s forecast the for the next 180 days?

 

The quick answer, I don’t, I don’t care about the fact that there’s seasonality, I don’t really particularly care about the fact that there might be a competitor coming up in the market. Like I just need to have some best, you know, explosion of data quickly. The cool thing, I think, in the in the world of apps is that these two titles are merging together. So analysts are becoming more data scientists, in the sense that now you could ask like, hey, bi analysts, I really need to know how many dog users there are in my app versus users that have cats. Can you figure that out? And you can find algorithms from Amazon and from, you know, Google that will do that kind of exploration very quickly. It would never satisfy like the itch for data science. This is so easy. It’s just off the shelf. And that’s I think the difference.

 

Adam Landis  11:47

So to break that down: analysts get answers today. And data scientists develop methodologies for highly specialized, difficult answers?

 

André Cohen  11:59

Exactly. We are the data scientists is really trying to come up with something new. That’s very specific. An example of this might be LTV, that’s something every app and game does, you can create an LTV algorithm using off the shelf tools. It’s not maybe that’d be the best, but it will do something. The alternative is to spend, you know, six to 12 months developing an algorithm for LTV, that is very specific to the fact that you have a I don’t know. I don’t know a bicycle app. Right? Which has very specific, you know, what kind of bike you have here, here, but it depends on the bicycle, though. Because if it’s a kid’s bike, summer’s peak. You know, a long racing bike might be like, Paul, because you know, 50 year old men only really liked the bicycle when it’s comfortable, interesting. There’s all these things, then you have to combine it all together, right? What kind of tires are they using? Kind of like the running apps, right? running apps are very specialized. And with a lot of data science behind it. You know, what kind of

 

Adam Landis  13:09

The data scientists will come up with a highly specific, specialized, LTV model that applies to that apps in that methodology for that reason, and it might be have a slightly or very different outcome, whereas the data analysts in that case would be okay, we’re just going to do a simple regression projection, for instance. Yeah.

 

André Cohen  13:32

And by the way, there’s no nothing that says analysts did not go, you know, iterate, if you give enough time. An analyst becomes a data scientist very fluidly. If you say to a data analyst, hey, for the next 12 months, just focus on LTV. You know, I think the answers are almost the same. I think it’s really about expectations. And sometimes,

 

Adam Landis  13:52

yeah, probably more analysis versus science would be yes. Okay. So did engineering get thrown out there a lot, too. It’s a title. It’s kind of I never heard of it till recently. But what is it a good engineer?

 

André Cohen  14:05

Yeah, so the mistake that has happened in the last 10 years, like even when I wasn’t a data scientist that I didn’t use the term very often is that the analysts in the app space, and games and data scientists in the game space and app space, were hired to solve problems, right, it goes back to the tip of the pyramid that I discussed above. And people forgot that they’re not self sufficient bullet that no one in a company is self sufficient. Right? If you look at a game developer, you have game programmers that program the game, but you have graphic artists, right? You have effects people, you have a whole bunch of cinema, like making videos for the trailers or the interstitials in the game. You can’t expect the programmer to do all of those things. And the same unfortunate thing is that data has And so data scientists and analysts are very common. But now, after 10 years of iPhones, and so on, people have realized you actually need data engineers, data engineers are a unique new, like development in the world of data. Because it’s people who are specialized in ingesting data from third parties, where the data is not trusted and not formatted. So it’s a matter of how you put rules and logic and correctness to data and assumptions, enough space that has an so it’s really like, it’s like the department that takes the world’s input. And it makes it understandable for the internal company to understand it, right? I see.

 

Adam Landis  15:50

So they’re in charge of quantity– data quality– in a lot of ways.

 

André Cohen  15:54

Quality is definitely one. Data Governance is a term that gets thrown out all the time.

 

Adam Landis  16:01

Data governance, meaning like, PII and making sure you protect the policies you— GDPR–

 

André Cohen  16:08

For instance. it goes into data quality data assurance, like is it there? When you want it? Yeah, what are the things that guarantees that we that you have for data, like Apple revenue? it, you still have to put rules around that, even if it’s Apple, right? Like, it doesn’t come up? You know, it’s not, it’s not in real time, the data, there’s a 24 hour, plus or minus–

 

Adam Landis  16:33

we see a lot, we saw 12 hours, big data, batching, you’d have to schedule it, you have to refetch it, you have to add delays, and you’d have to dependencies. So the engineer, the data engineer, gets the data in, the analyst gets fast answers, and the data scientist develops methodologies for tough answers.

 

André Cohen  16:55

Right. I guess what we have not really like the only thing that’s missing in this, if you lay it out, is a data platform or data architect role in this. So data engineers do more than just ingest, they also create progressively more  interesting data out of right, where they have to their responsibility to merge data together from different sources, but also becomes their responsibility–

 

Adam Landis  17:23

ETL, ELT?

 

André Cohen  17:24

Exactly. The reason that becomes a data engineering role is because if you let a analysts or data scientists do it, they will only have enough visibility to do for the very specific problem they’re trying to solve today. So if they want revenue from Google and an apple, in a spreadsheet, they will create date, Apple, Google, and that’s it. Which will be perfect for this one analysis. But if tomorrow someone says, Yeah, Andre, we really need it by country, as well as start over. New Python file, or SQL, you just do it all over. And it’s fine. If you’re a small company of 10 people, if you’re 1000 people, you just made a project out, took a first a month, you know, added another month to the project.

 

Adam Landis  18:11

So we obviously are very interested in data platforms for obvious reasons. What is it it a data platform? How’s it playing all this?

 

André Cohen  18:20

So data platform is the– in my view at least– is like it’s a holistic understanding of data, right? It is about having enough visibility, and collaborating with data engineers and data architects, and also the down the road, you know, analysts and data scientists and even product people. And what are the rules? And what are the what is the strategy for data? It can be anything from what things we want to ingest data from, what things do we want to give trust to in a way? And which things do we say we do not trust, we can simply not trust it. Because if you blur the line of what is trusted and not trusted, you end up with situations of multiple data sources, you end up with inconsistencies. And the classic inconsistency is just naming conventions. Do you call it UID code or user ID.

 

Adam Landis  19:18

One thing we’ve seen a lot of is how the political affiliations of the organizations who define the country code schemas, you’ll see that bleed through a bit on on exactly how they, how they standardize on country names. Are they using the UN? Are they using the ISO standard either using old fashioned names of countries that no longer exist? It’s kind of funny.

 

André Cohen  19:41

It’s fascinating. And it’s it’s always like the trivia for understanding like, how much time have you been in this space? Because it’s like NA, country or not? It’s like almost the game.

 

Adam Landis  19:53

Anonomous proxy. I like that one.

 

André Cohen  19:54

Yeah, there you go. But even like what is no like empty, what is the what the “empty” mean, there is every service player has its own definition, and the company needs its own definition. So that’s what goes on in the data platform. And the idea of the player platform is that it’s “productifying” data.

 

Adam Landis  20:15

Something that I have noticed, when I talk to people– there’s gonna be people that watch this, and they’re gonna nod with everything you’re saying, because they know it, they understand it. And there’s gonna be people that are kind of like, I don’t understand what data they’re talking about. Like, when I talk to people about what we do as a company, they say: “Oh, I rely on my MMP. So appsflyer, Adjust, Kochava. I get my data from from there.” Now, why would– why would there be multiple data sources? What is the value? And what kind of data are we talking about? Why do these all need to come together?

 

André Cohen  20:50

So everyone needs a specific kind of data to do their job. That’s how it is, so when I was making games, me and my business partner, we were wearing multiple hats. And so for us, it was very easy to use a single data source for it’s because we knew the tools, but also we we, because we were so much together and like the work on the day to day work. We modify tools to adapt so that we could do everything from a single tool. Right?

 

Adam Landis  21:25

I mean, that’s… Excel.

 

André Cohen  21:27

But no, no, the tool, let’s say, my day, let’s say Mixpanel, right was an analytics tool that was very popular mixpanel can technically record any piece of data in your game, or app. So when we had added IAP to one of the games, it was very easy to incorporate that into it. And not only that, but we knew the value or the the, we wanted to make accurate decisions about our games via Mixpanel, and revenue. So we did not, you know, we made sure that the revenue lined up exactly with Apple’s to the closest 5% difference. Now, as you expand a company, sometimes you can’t do that you can’t be in sync, an example of that is five years ago, we definitely couldn’t get ad revenue on a per user level, click-level, in your analytics, because it was just not data that was available, you would have to get the analytics from a player from one source. And later, you’d have to download or extrapolate the revenue per click from a different source.

 

Adam Landis  22:28

Total average and a total waste of time.

 

André Cohen  22:30

But then you can mirror that at the end. And then you know, you come up with a with an answer about player value, or LTV. Now, that is way where data data science data engineering, data architecture comes in. That process, how do you serve, like, now we just talked about two data sources needing a whole, you know, team of people to, to merge it together correctly. But again, you know, that’s not to position. So if you’re talking about revenue, that’s like a monetization role in many companies. And we talked about ad revenue as a different as a different title in the company. Now, if you look at an average app, there’s many other things going on. There’s marketing, there’s social marketing, or there’s user acquisition, there is mailing lists that you might have

 

Adam Landis  23:17

CRM…

 

André Cohen  23:19

Yes, CRM, there might be leaderboards, or things that are, you know, encourage players to play the game. In a King situation, subscription is a different thing altogether, then, you know, you might realize– that’s what happens a lot is like– oh, you know, firstly, like it used to be iPhone, oh, we should also launch the app in Android, that adds a whole bunch of new things that you have to consider. Then you might say– which is happening a lot now–“Oh, the Mac is very similar to iOS, it’d be a very small hop for us to take our, you know, dog app, picture sharing app to the Mac”

 

And now all of a sudden, none of the tools we just spoke about work on a Mac. The there is no the push notification maybe works.

 

Adam Landis  24:01

So it’s a whole different set of tools and services…

 

André Cohen  24:03

–No ad monetization, so you’re gonna have to do some other things.

 

Adam Landis  24:07

And the Huawei store. So if someone says, hey, yeah, just clone your Android app and put on Huawei. It’s a totally different revenue source.

 

So what you’re saying is, as a company gets more sophisticated, releases more versions, or releases more apps, they spawn up different datasets that need to be merged, and that’s where it gets complicated.

 

André Cohen  24:26

Exactly. It gets complicated because of that, or the other way that I thought it’s not that the apps growth but by even by your company’s growth, like the day that you hire an ad monetization person, there will be needs from that person that cannot ever be really translated into the data tracking that you have.

 

Adam Landis  24:43

You did not talk about what I think is one of the one of the biggest consumers of data and that is the product people. The product people trying to decide from the user’s analytics journey, what to do to drive the application, user’s retention, and impact and engagement.

 

André Cohen  25:00

You’re completely right, there’s a whole world of AB testing, I think is one, right? You often end up with AB testing as an afterthought like, oh, man, we really need to test something. And having the right solution is kind of hard to duct tape into a game or an app. And to be fair, there’s very few service providers out there that do that. Specifically, right. So that’s a great example.

 

Adam Landis  25:29

But without without being too self serving, you always talk about the easy way, the intermediate way, and then like the advanced way of doing things, so I was thinking about if you were to give advice to the “x” app publisher, and they’re, they want to get more sophisticated about their data, they want to get more answers into their data, they want to make their data more useful. What would you say is kind of those three camps? Like what’s the easiest way just like make it work? Forget about it, move on. And then when you start getting more sophisticated, how do you do that? And then the penultimate, A+, what are those stories look like?

 

André Cohen  26:11

Yeah, no, it’s, there’s two ways and I think a lot has to do with like, how, how ready Do you want to be from day one, right, as well. Like what I always joke about, like, I’m from Brazil. So in Brazil, we learned to play soccer without shoes, and that like with a piece of rock, on the dirt, and on the street, that’s how we learn soccer. If you go to the United States and you see how kids learn soccer, you know, they have a full set of gear at age five, they have the Olympic, you know, authorized soccer ball, the goal is in pristine condition, like you have a goal. Let’s like start there, in Brazil, you just have to do a little rocks, you know, and kick through the rock, which is more than sufficient for kids. So it’s kind of like that for this too. So keep that in mind. Going like you know, with a limited budget and starting out when your small, I think the less services used, it’s always better. And in the case, for apps, it’s super easy, because Google is actually a very decent service. And you can probably– with the with the right knowledge and the right team, almost use it your whole life.

 

Adam Landis  27:18

And that’s Firebase?

 

André Cohen  27:20

Firebase. It has the AB testing component has the analytics component, you can hack your way out of you know a lot of things by having BigQuery and Data Studio integrated. So you get, you really can go from, you know, lawnmower setup for analytics and data for app all the way to kind of like a Prius, like a pretty decent Prius car, if you use a car analogy, with just a couple tricks, and you’re in priuses, electric, you’re ahead of the curve, like it’s pretty neat. It gets the job done.

 

Now, what the big step up from that, however, is data into using multiple service providers for this, right, we’re using the best, that would be level two. And I almost say like you can skip to level two if you really feel ambitious. Unfortunately, if your app or game doesn’t really monetize quite yet, you’ll feel very expensive. But I mean, I would always say that tech generally is less expensive than humans. So take that into account. Like there’s tools like Amplitude that do amazing analytics. There’s tools that do great AB testing for campaigns, for instance. They’re expensive. If you’re if you’re starting out, and you ask for the price, or if you find yourself asking for the price as an element of decision making, you’re probably not there. Go down to step one. Unless you you’re funded or something and you want to go big. That’s like giving the five year old the full set of gear.

 

Adam Landis  28:57

I mean, I think it’s probably fair to say I mean, I think Firebase has something like 90%, penetration and all of that. So it’s fair to think that everyone has experienced Firebase. That’s made it this far through the the talk.

 

So if we’re talking about the intermediate: that’s multiple data sources that’s trying to get different sources of information that go that Firebase may not include, or, and what what are some examples of data sources that aren’t included in Firebase?

 

André Cohen  29:31

Well, Firebase and Amplitude, they’re only as good as the user. That that’s usually the challenge in our space. The difference between them is not what data sources you get from these tools, but rather, what new features can you get out of it to extract value because data is relatively cheap, what’s hard is the knowledge that you get out of it. So you know, as we said, like you have BigQuery and you have DataStudio from Google with Firebase. So you’re down now to SQL queries for the bulk of your work.

 

Adam Landis  30:08

One thing I will throw at this is, of course, pretty self serving. So take this with a grain of salt, but BigQuery makes it difficult to kind of pipe data in really well. So impression data, your IAP from like a revenue cat, it’s hard to get all that transformed and in.

 

André Cohen  30:25

Exactly.

 

Adam Landis  30:26

Your own data engineer using Google query, or Google BigQuery as a platform, and then you have to do your own analysis, or data sceince on top of it.

 

André Cohen  30:34

But Adam, you’re hinting at level three, almost, because that is level three. If you find yourself at two different pieces of data and saying, this is really hard to do with Firebase, and you would say the same thing with Amplitude actually in Amplitude’s case, you would say, I can’t do it unless an engineer helps me.

 

Right there’s an SDK.

 

That’s where your level three, level three is, we’re going to create new data sources based off of another service, or we’ll build our own data service or services. Like, you know, if you go to EA, or Zynga, they all have their own analytics backends. And level three is: we’re going to roll your own final data sources. So if you really care about the, you know, having user tracking, plus ad revenue, you’re in level three, you doing

 

Adam Landis  31:26

Are you doing that for cost? Like, why wouldn’t they use the best of breed for like an Amplitude of something?

 

André Cohen  31:31

That’s a great question. And that’s kind of what excites me about working with you guys. And even data platforms right now is doing it yourself is very expensive, error prone. It doesn’t scale well.

 

Adam Landis  31:46

–build versus buy that whole —

 

André Cohen  31:48

–there is that but it’s also, even in a portfolio company, if you saw this, you know, ad revenue with, you know, IAP revenue issue. There’s all sorts of little edge cases you don’t know about.

 

Adam Landis  32:03

There’s pain of exploration.

 

André Cohen  32:06

That’s data exploration. And once you solve it, the cost of solving these problems are only as great as you know, dividing it by the number of apps in your portfolio.

 

Adam Landis  32:16

Right. So what are you getting in return…

 

André Cohen  32:18

It’s a terrible return on investment,

 

Adam Landis  32:21

which is why companies like RevenueCat have come up — because so many different companies have solved the subscription challenge with the platforms. And then RevenueCat’s come out with a great product that can just pipe that data back into their their platform. It’s like a no brainer.

 

André Cohen  32:36

Yeah, I think you know, the, the dream that I have. And what I really look forward to is that level three will slowly be phased out. And to fit level two, where is your connection of services, all specialize in something really great, that delivers really great final data or perfect data for end users to use. And you might have three or four of those.

 

Adam Landis  33:03

You’ve used this analogy in previous conversations: back in the day people have built around Oracle Database data, data warehouses, they call them back in the day, they had their DBAs, they’re data architects. But then snowflake came along and said, hey, we’ve got the best of breed. It’s like a plug and play. So people are starting to migrate as technology changes back to these service platforms that are better, faster, cheaper.

 

André Cohen  33:26

When we when I started making games, there was only EC2 instances on Amazon. That was the only AWS service. So what happened, I built my own analytic service using rock machines with Ubuntu  to you know, as the operating system, I think about the operating system choices. And then eventually, they had SQL as a service, right? RDM RDB. And then we all closed down these instances, create our RDB instances, then you know, someone as you said, Snowflake, Oh, forget it, you don’t need a database. Use. We have Athena, put files in s3, and it’s done.

 

Adam Landis  34:05

Athena gets expensive, we know that firsthand. Anyone who’s going to Athena, talk to us first, it gets expensive.

 

André Cohen  34:13

Yeah, data is super expensive. Which is why you have to really know the questions you’re trying to answer.

 

Adam Landis  34:22

So we’re coming near the end of our time, is there any kind of closing thoughts that you have for an organization who’s looking to get sophisticated that is wanting to know more how to utilize their data efficiently?

 

André Cohen  34:36

So I think knowing the right questions is super important. And actually, most people know the right questions already, right? Like we we spoke about ad revenue, how that’s interesting. If you have ads, and then incorporating that with some other metric and solving that problem, you know, individually, it’s probably easier. As you get bigger, you’ll have more and more of these problems. And eventually that’s where data platforms comes in. But I think it’s really important not to get lost in all these problems, right? It’s very common to see people say, I really need to AB test things. But it skips all these other questions that are way more obvious that you probably already had, which is like, Am I monetizing correctly? Well, how accurate is your monetization recording? Maybe we can figure all your problems just by looking at your revenue data. You don’t need to go off this deep end yet.

 

Adam Landis  35:28

So start small with easy to answer questions.

 

André Cohen  35:31

Yeah, there’s a lot that can be done. That’s what I would say, with very basic data. And I think the limitations are even great. And we’ll work on that.

 

Adam Landis  35:42

Cool. Well, Andre, thank you very much for your time. We’ll do this again. If anyone has questions that they want Andre to answer, send them my way. I’ll put my email somewhere along this, video. And we’ll try to address them in future editions.

 

André Cohen  36:00

Yeah, this was excellent. Adam. It was great to talk and chat about like these different divisions of data that exist and really look forward to chatting more about what you guys are doing and also just about this space–

 

Adam Landis  36:18

You mentioned building an LTV curve, I want to talk about that that’s going to be the future. All right, everyone. Thank you very much for joining!

 

Transcribed by https://otter.ai