#190: I See Your Data Layer and Raise You a Metrics Layer with Benn Stancil

“Modern art” was a terrible label because, ya’ know, time would pass and here we are 50 years after the end of that period shaking our heads at what a short-sighted semantic gaff that was. We share that observation for no particular reason. On this episode, we sat down with broad, deep, and entertaining thinker Benn Stancil from Mode to talk about one facet of the modern data stack: the metrics layer. What is it? Who’s thinking about solving for it? What is a monthly DAU? These are questions to ponder that, hopefully, won’t leave you impersonating a piece of modern art.

Resources and Platforms Mentioned in the Show

Episode Transcript

0:00:05.7 Announcer: Welcome to the Analytics Power Hour. Analytics topics covered conversationally and sometimes with explicit language. Here are your hosts, Moe, Michael and Tim.

0:00:22.6 Michael Helbling: Hi everyone, this is the Analytics Power Hour, and this is Episode 190. We like to think of ourselves as fairly modern, and this is a modern Analytics podcast. Moe, how is the modernity?

0:00:38.7 Moe Kiss: The modernity is great.

[chuckle]

0:00:41.9 MH: As opposed to the maternity, which is a different thing, all together.

0:00:46.6 MK: Totally a different thing.

0:00:48.1 MH: Yes. Tim even as quintessential as you are, you still stay pretty up-to-date on things. How are you doing?

0:00:55.8 Tim Wilson: Definitely not modern.

0:00:58.0 MH: Yeah, I mean, and I am Michael, and I even own an NFT, so I mean, I’m doing alright. Well, you know…

0:01:03.4 TW: You really are proud of that?

0:01:04.8 MH: I’ve owned a number of them in theory, I guess. Okay, let’s not talk about that anymore. We’ll do a whole podcast, of sort about how I raise money out of stupid stuff. Alright, you know, last year we renamed the podcast, and actually it was just so we could talk about topics like this one, the modern data stack is upon us, and unsurprisingly not everyone’s problems have been completely solved with a cloud-based data warehouse and ETL tool and a reporting platform. It turns out what a lot of business users have wanted all along is consistent definitions and usage of metrics across the entire business. To which we said, “Well, why didn’t you say so?” But we thought we should add a guest, someone who could give us maybe some more insight into the ever-changing Hydra, that is the modern data stack. Benn Stancil is the chief analytics officer and founder of Mode, an analytics platform designed to help data analysts and data scientists analyze, visualize and share data. He has also held analytics role throughout his career at Microsoft and Yammer. He has an excellent newsletter, you can subscribe to it, benn.substack.com, that’s B-E-N-N.substack.com. And today he is our guest. Welcome to the show, Benn.

0:02:21.7 Benn Stancil: Howdy, I’m glad to be here.

0:02:25.0 MH: So Benn, maybe to kick us off and to get the conversation going, how did you find yourself as an advocate or a voice in the modern data stack problems, because that’s what brought you to our doorstep and why we’re talking today, but we’ve all really enjoyed reading some of your writings over the years, and so it would be good to hear a little of your background, maybe.

0:02:51.1 BS: Yeah, I appreciate that. So we started Mode, me and a couple of folks started Mode now about eight years ago. So we worked with Microsoft. Yammer before that, Yammer was acquired by Microsoft, then Microsoft and ended up building some internal tools at Microsoft that we thought, “Hey, this might be good products to build for the rest of the world.” And that’s kinda where Mode came from. My story in that was me and the two other people who started it, one of them was our CEO, he is good talking to other people, he is presentable, he is nice, people enjoy hanging out with him, so he was the CEO who was out talking to potential customers and investors and things like that. Our second co-founder was an engineer, who he was responsible for being locked in a room to build the product. And my background is as an analyst, and if you’re an analyst to the three-person company with no data, you don’t have a whole lot to do. And so, I basically started writing stuff on a blog at that point, this was a while ago, it wasn’t about the modern data stack, which wasn’t even a term at that point, it was more about pop culture things that were related to data.

0:03:54.0 BS: So the very first blog post on Mode’s blog, and it’s actually still up, if you go to our blog. It’s a post about Miley Cyrus and the VMAs, and it’s basically like a data-driven look at the VMAs and YouTube views of various videos, and Miley Cyrus’s video had terrible reviews on YouTube, and it was kind of like, “What’s going on here?” I don’t remember the answer to that question, but basically my view of that was like, “Hey, the thing that… One of the things we can do is, we can engage with the community in different ways.” And if I’m not building a product or out doing sort of sales and things like that, that was something that I enjoyed and worked actually reasonably well for us kind of unintentionally, we weren’t really had a whole lot of plan around that, it was just kind of like, “Alright, this will be a thing that we’ll see how we kinda build some audience with this.”

0:04:35.0 BS: Over the course of the last seven or eight years as someone who’s working a start-up, you find yourself in a bunch of different roles, you end up filling a bunch of different sort of holes and things like that, bouncing around from different departments, and so that was basically my course of things. As Mode has grown, and as the company’s got bigger and it’s matured. We’ve brought on more and more people who know what they’re doing, and so I basically slowly rotated my way through the company until getting replaced by people who are experts in the things that I was attempting to do. At which point some time, about a year ago, I kinda found myself back where I started this time basically, “Okay, go write some stuff on the internet.”

0:05:09.0 BS: But instead of it being about kind of data and pop culture, at this point, I had a lot longer of just like time thinking about the industry, the space tooling, things like that, the problems that we encounter, the problems we know our customers encountered, and so that’s where the substack came from, and it’s a little bit of a substack without a clear thesis, it’s more of talking about what this in the space, what’s happening in the space, where things are going, things like that, but a lot of it is just sitting around you know, “Hey, what’s going on?” Exactly as you’ve described in your open of like modern data stack is a difficult thing to characterize, exactly, it’s an evolving thing where people who have the privilege of being able to see a lot of that just from being vendors in the space, talking to a lot of teams, and so it’s basically my thoughts on that, which sometimes come out okay and sometimes come out less okay. And so that’s kind of where I came from.

0:05:57.2 TW: Well, in the modern data stack, it’s basically, it’s just DBT, reverse ETL and the metrics layer, I feel like those are like the… Those are three different things. I don’t know, when it comes to something like the modern data stack, I think that feels like something that people are comfortable, that people were trying to work through what that is and aren’t necessarily trying to declare this is what it is, or is it one that you feel like people are declaring? And I guess I’ll ask within that, the metrics layer it…

0:06:30.2 TW: Is that something that people agreed on kind of what it is and are now all trying to come with ways to solve for it and explain it? Or is it one of those terms that is now starting to spread out and people either… Like an analytics engineer, people assume they know what it means and therefore start deluding the meaning of it.

0:06:56.1 BS: So I think on the definition of modern data stack, I think that is pretty deluded and that it… You have all these different people kind of have different definitions of it. Some people are like, well, it’s the un-bundling of stuff, some people say it’s the same BI tools in the cloud, it’s decentralized, it’s the… I think Martin Casado’s partner, Andrew said, it’s the delamination of the warehouse. There’s a bunch of these different ways of putting it. My version of this, which is sort of tongue and cheek, but I actually kinda stand by it, is it’s analytics tools that were launched on product turn.

[laughter]

0:07:30.5 MK: I can see that.

0:07:33.6 BS: Because it’s basically things that came up in that era and that were aimed at that audience, ’cause it’s like, it’s the Oracle cloud data warehouse, part of the modern data stack. It’s like… I would kinda say, no. It wants to be.

0:07:48.6 MK: Yes.

0:07:48.9 BS: But it’s not really. And I don’t think most of the people in the community would say this, but kinda who cares. And I think to your question of, is it… The modern data stack to me is anybody can start a company and say, “Yes, this is a part of the modern data stack.” I don’t think there’s any, “Oh no, you have to join the club, you have to have so many users, otherwise you’re not a part of it.” I think it’s just kind of like that philosophy of tools.

0:08:07.0 MH: Well, yeah, the key is you have to start a company ’cause if you’re an existing company, you can’t… Microsoft can’t just say that Azure is part of it now. I mean it is but, yeah, I get what you’re saying.

0:08:21.2 BS: So I think that part’s a little bit fuzzy. On the metrics piece, I think it’s still new. It feels pretty well defined as to what it is supposed to… The high level goals feel pretty well defined of, “Oh, we understand kind of the scope of the problem, we understand roughly where we wanna go to fix it.” I think there are now a lot of new ideas emerging about how to exactly solve it. The folks at DBT have a particular flavor of it, Transform has a particular flavor of it. Looker is becoming one with the way that they’re seeming to strip Looker now out of the product and put Tableau on top. There’s Malloy, which is the Looker spin-off thing by the founder of Looker, that’s kind of a metrics Larry thing. There’s a bunch of versions of this now that are kinda in that flavor, but I think the concept of just like standardization of metrics that isn’t traditional BI and how it does it is a fairly… The walls around it are fairly well defined. I think that the walls around observability are not well defined. Nobody knows where observability really ends ’cause people define it much of the ways. I think the metrics layer is pretty clear where that starts and stops. What you do in the middle and how it all works, I think is still a lot to be figured out.

0:09:31.3 MK: Can we just take a step back though and start with the problem, because a lot of the concepts that you’re talking about and the technologies are things that we do use and can. In fact, your blog was like an actual description of exactly how we’re working at Canva, it was really funny to read. But I’m not sure maybe everyone follows what the metrics layer is trying to solve for, right. ‘Cause there is a problem there that all these companies are trying to tackle.

0:09:58.7 TW: I feel like a lot of people here at first, will hit our face level description and say, “I know how to solve that,” and that really means they don’t really understand the problem that’s trying to be solved. Yeah, I’d love to… If you say it’s a pretty well-defined problem, I would love to hear to that articulation.

0:10:13.5 BS: Yeah. So my definition of it is, the way we used to do this was kind of a two-step process. You’d have a bunch of data in a warehouse that was all raw or what… It was just like you dumped it in the warehouse, and then BI tools did everything. And so you would define relationships between that data and BI tools, and then you would define metrics on top of those relationships. So you would have a giant list of raw customers like purchases. Say you’d have a list of purchases, the BI tool would define relationships between purchases, and customers, and line items in those orders that would create these data models to say how you join stuff together. And then on top of that, you’d have formulas basically, that would say, “Here’s how we compute revenue from a giant list of orders,” whereas like, do you include orders from gift cards? Do you include orders that get returned? Do you include sales tax or not sale? How do you make those adjustments? Those are all decisions you have to make when you’re actually defining this metric. And so the BI tool would be the place for the logic of the joining and the logic of the metric would live. And then people would have a little drag and drop tools to be able to say, “Show me this metric by this dimension,” and they’d build stuff and underneath the surface, it’s all just like a big data cube.

0:11:20.9 BS: DBT basically pulled the first part out of that out. So DBT said, “Okay, what if instead of defining those relationships in a BI tool, we define them in a way where they just get derived into tables in your warehouse.” So that we now have a nice clean dimension purchases table, and we have dimension customers, and hopefully those joint keys become relatively straightforward, but you clean up a lot of the sort of modeling logic now in DBT before you actually get to any kind of consumption tool. And then those consumption tools would just sit outside of the warehouse, automatically integrate with DBT is not really an integration, but the output of what DBT creates is tables that is just readable by anything. And this was one of the subtle brilliance of DBT was they could slide under any tool and automatically the values of the things that they created would integrate with everything else without having to actually build that integration. So then we end up with like, okay, we have these tables but, how do we actually turn these purchases table into a metric about revenue?

0:12:17.7 BS: There’s still logic that has to go into that, and ideally, we have a dashboard over here, and some analyst does some analysis over here, we’re using the same logic for that. But we didn’t have another solution for that. The way that we’d have to do that now is, you’d have a BI tool like Looker or something where that computation would get put in or we’re in Tableau would be computed like a calculated field or whatever, and so you’d have your nice dashboard. But then have an analyst wants to come along no matter SQL query where it’s like, “Oh, our revenue went down, what happened?” The first step might be to start recreating that and some other tool to be able to do richer analysis by writing code against it. You’ve gotta recreate that metric, and that metric logic doesn’t live in a place that’s accessible. The metric logic lives in a place that is… It’s now just accessible to the BI tool. And so the idea with the metrics there is essentially what if we put the metrics logic…

0:13:06.7 BS: Further upstream from consumption, such that all of these tools that use it, whether or not they are analytics tools or BI tools, but also you could do it with reverse ATL, you could do it with operational tools, you could do it with NL platforms, whatever, those things have access to the same kind of core governance logic that defines the metric in the same way that they have access to the core governance logic that defines the tables through something like DBT.

0:13:30.7 BS: And so that, to me, is why it’s a fairly clean… To me, metrics layers do not derive tables, they don’t do fundamental business logic modeling of those tables, nor are they about some consuming data. They’re mostly just a framework for saying, “Hey, we have a bunch of inputs that are tables. What is the formula that we wanna compute on top of those tables that will spit out a consistent metric? And then, what’s the API essentially for being able to access that? Is it an interface? Is it like actual code? Is it SQL? Is it some other API that’s like some REST API?” That’s where all I’d like to think that the implementation details get decided. There’s also a lot of stuff I like to transform, do it differently than DBT. Yes, they do it very differently. How does it actually work? But fundamentally, it’s, I wanna be able to request a metric to say like, Show me daily active users by day, by region, by quarter, whatever, and that to be reliable regardless of which tool I actually make that request from.

0:14:23.8 MK: Yeah, and I mean I really like the point you made in the article about the alternative, is that you end up basically, like you have your, I guess, model data right. And then you end up with report tables that are trying to cut by different segments, and then you end up with hundreds of report tables because someone wants to look at it by country, someone wants to look at it by customer segment, someone wants to look at it by length of customer duration or whatever it is, and so you could potentially otherwise end up with a gazillion different report tables all trying to aggregate that data for some kind of data visualization tool.

0:15:03.1 BS: And Looker got this right. Looker was the first one. Looker is architecturally the same as what BI has been forever, but they got the idea right of, okay, we build this, what essentially is a metrics layer that then allows people to do all these derived aggregations and stuff so that you don’t have… You’re not creating a giant cube that is a million dimensions wide, sort of is the synthetic cube that the whole thing happens on the fly, which with a cloud warehouse, it’s fast, you can now do… But the problem with Looker was it was… I mean, the problem, it was obviously a very valuable company that had a good outcome and makes how much money, so it wasn’t really a problem, but the thing that I think, architecturally, Looker was going to run into problems three or four years from now, and they kind of already are seeing that because it is a metrics layer design just for the Looker BI tool.

0:15:50.5 MK: Yes. Totally.

0:15:51.6 BS: And at some point, I think they want that to be more open and obviously with the Tableau integration, which I think we could probably argue was done for a few different reasons, but one of those reasons might be a recognition that actually look, NL is more valuable as a a generally accessible thing, as opposed to just a backend to a single BI tool.

0:16:10.1 S5: Alright. Let’s step aside for a brief word about our sponsor ObservePoint, the industry leader when it comes to governing the completeness and the integrity of the tagging on your site. I know this isn’t the sponsor spot where we do a quiz normally, but I decided to make this one a bit of a test, just for fun, so Moe, you’re up. Fill in the blank. ObservePoint can be used to automatically blank your data collection for errors across your entire website.

0:16:38.4 MK: Audit!

0:16:39.8 MH: Correct. Okay, Tim. Now, it’s your turn. Fill in these blanks. ObservePoint can be used to blank your blank blank blank and blank blank for blank and blank blank blank.

[laughter]

0:16:52.6 TW: This doesn’t… This doesn’t seem fair, but I’ll give it a go.

0:16:55.6 MH: All right.

0:16:56.0 TW: ObservePoint can be used to test your most important pages and user paths for functionality and accurate data collection.

0:17:03.5 MH: Oh, I am so sorry. Well, Tim, your statement about ObservePoint is correct. The answer was only supposed to be the blanks, so the correct answer was test most important pages, user paths, functionality, accurate data collection. So Moe is the winner!

0:17:18.8 MK: Woohoo!

0:17:20.5 TW: Jeez, give me a break.

0:17:23.0 MH: Well, better like next time, Tim. Maybe study a little harder.

0:17:26.8 TW: I did study. I actually read ObservePoint’s 2022 Digital Governance Report, which was a survey of 230 marketing analytics and privacy professionals, and I was all set to go with tidbits from that report about how companies are thinking about things like proving server side for marketing and analytics tech grappling with privacy challenges and maintaining effective data governance.

0:17:50.3 MH: Oh wow, I mean, that sounds pretty interesting. Too bad that wasn’t part of the quiz, but if you, dear listener, wanted to check out that report and/or learn about ObservePoint’s many governance capabilities, head over to ObservePoint.com/analyticspowerhour and do that. All right. Let’s get back to the show.

0:18:15.1 TW: So, something that I’ve seen in a digital analytics and BI context creator time and time again, pick your digital analytic tool, be it Adobe, be it Google, pick your BI platform, be it Tableau, be it Domo, you name it. And I’m struggling to figure out if the metrics layer helps with this or if just faster processing, but it’s… I’ve watched BI teams say, “Let’s just bring in… We’re gonna bring in the digital analytics data.” I can go back 15 years with Cognos saying, “Sure, we can handle your web trends data. It’s just data.” And the core thing that BI developers can never grasp, and I’m curious as to whether a metrics layer helps with this or not, is that whole de-duping, metrics that need to be de-duped, which I know we use daily active users, is similar to this, but it’s like… Oh, we’re gonna bring in a flat table.

0:19:13.1 TW: We’re gonna put ten different… We’re gonna pull this out of our BI tool by day, by channel, by device category, by whatever, whatever, and we’re gonna bring in visits, page views, visitors, conversions, orders, or whatever. And so it winds up coming into the BI tool with dimensions and metrics, that some of them, aggregate and aggregate, fine, ’cause you can… Revenues is revenue, it’s all gonna add up, but it’s something like visits or visitors, it starts cratering, ’cause they say, “Oh, well, now when I roll up and try to look at total visitors, my number is huge because they’re spread across multiple dimensions.” And the one trying to explain that to the BI tool owners who seem to be conditioned to working with data that that’s not an issue. And the only solution I can think through is, well, you have to be able to query that raw data on the fly. If I have a metrics layer, I could do that, but now am I not relying on… Every time I do something, I’m having to do a massive query to go through and do all the processing to get to that?

0:20:28.0 BS: Yeah, this is a real question, I think, and it’s something that is still to me an open issue. So one, I very much agree with you that, there’s a couple of things I can just tie to that. One, there’s a question like this for metrics bloat, of just like we have tons of metrics, we thought we organized them already and we have right things, now we have a 1000, what do we do? I think there are some technological solutions to that. There’s a handful of companies out there that are doing these kinds of metrics layer things, they all have their slightly different variant of how they differentiate themselves. I know of one, for instance, that is focusing on exactly that issue, that is focusing on, how do we help if you create one metric and somebody else creates and that metric looks very similar, can we try to say, “Hey, these are the same, don’t do that?” That people have built in their platforms, basically, they try to do this and try to say, “Hey, these two things look pretty similar, can we make that a different… A smaller problem?”

0:21:19.3 BS: I think the real solution to that is probably just discipline, it’s discipline by the teams that you have to figure out ways to keep people from doing that, somewhat, somewhat… I think DBT also sort of figured some of this stuff out, I don’t think this was a DBT-specific solution, but they figured out how to, to prevent some of the bloat around tables, by actually just putting a bunch of friction to create them. There used to be a… Facebook, for instance, used to have a way that was a sort of internal like DBT type thing, where anybody could write a query to materialize this as a table. Effectively, it’s DBT without some of the fancy DAG stuff, but it was essentially, I could just create a job to say build a table out of SQL. It was a disaster, in the sense that people created thousands of them and nobody knew what was good, and it was a disaster in the sense that Facebook is worth a trillion dollars, but it was also a disaster in… It was really, really hard to manage and nobody knew what to trust, and so everyone just had their little wall gardens.

0:22:09.6 TW: And if somebody spun up a table, could then other people could see that table? So it’s like… Okay, so you were just throwing… Okay.

0:22:14.6 BS: Yeah. Yeah. So you’d be like, “Oh, this looks interest… This is useful, I’ll try this,” it’s like, “No, don’t do that kind of thing.” And DBT sort of solve some of that by putting friction in the way, and they would probably argue that part of that is good practices, peer review, that sort of stuff. Honestly, I think the bigger thing is it’s just like, gets kind of hard, a lot of people don’t wanna learn it and not that many people are going to, and so you automatically constrain how much stuff you can create.

0:22:38.2 MK: Ohh. I’m like…

[laughter]

0:22:40.2 MK: I work in a company with like 110 analysts, and it terrifies me how much stuff can be created, because every single person on the team is expected to be able to use Git, so.

0:22:52.9 BS: And that’s… Yeah. And there are certainly ways that that probably breaks down…

0:22:57.6 MK: But I think that’s where the good practices come in, you have peer review. And like our data warehouse team have been ruthless, where you can create a temporary table and stuff, but they delete them very, very, very regularly, intentionally to stop people doing that. It’s like, you have to go through the proper process of getting peer reviews and actually pushing your code to our data warehouse, and if you build temporary stuff we’re gonna delete it, so that you can’t build things off the back of it.

0:23:26.4 BS: Yeah. And we… That’s something we’ve done at Mode but haven’t been as rigorous about, where it’s like we have Scratch pages, but if your stuff ends up there, and we drop the Scratch pages every Sunday and… That’s what you get. And so, I think there is some stuff like that. To your point about metrics though, and this re-computation every time, yeah, this is the… I can’t remember the exact term, the symmetric aggregate problem that Looker solved. Basically, you can’t say you’re trying to do something that you can’t just sum numbers together, like daily active users you can sum into monthly active users by just summing daily by month, but monthly distinct users, you can’t do that. You could sum the total users that are logging on everyday and say, “This is the total number of log-ins,” but you can’t sum distinct users ’cause if someone used it 30 days, you tripled 30 times count or whatever.

0:24:16.2 TW: So Looker actually has those sort of constraints that you can build in to the…

0:24:19.8 BS: Correct. And this is one of the really, the technological innovations of Looker is basically, “Oh, you can define these metrics and they will do these things in such a way that you can create fan out joins that it will then figure out where to collapse those down, such that you can count monthly active users on top of daily active users without just summing everyday user, like it will do the distincts. To your point though, the way that it does that is it pushes computation down to the warehouse, it can’t pre-aggregate everything up into a giant table and just… It’s not really just a cube, it’s having to actually push that computation back down to the warehouse. I don’t know how we solve this, and one way we solve it is we all spend a whole bunch of money on stuff, which is kind of what it seemed like…

0:25:02.6 MK: Hello, welcome to our lives.

[laughter]

0:25:07.3 BS: And so that is one option. First off, like loves it, but it’s not really a scalable thing. I think there’s probably some ways that we can get around… People will figure out some smart things to do. Transform does a little bit of this. Transform is built on top of… Inspired by tech that came out of Airbnb, where they did this kind of thing, where they essentially pre-aggregate a bunch of stuff once, and then once they had it, they potentially put it in memory store that would let you kind of fetch those metrics in a way that was a little bit faster so you didn’t have to run queries every time. But I don’t really… This is like getting out of my depth, I don’t really know how this problem get solved. I think it’s like, hit the performance problem first.

0:25:45.8 TW: Well, although it seems like you did… Part of it does come down to having some discipline around metrics. It’s… ‘Cause there’s a… And there was… I think there was… I’d read something where it was… The author was talking about, like it was Falkon that came out of Mona Acmo, Why metric should be a first class citizen in a modern data stack.

0:26:08.8 TW: And I was like, well, this will be interesting. And it seemed like her definition was well, metrics are basically this really tightly defined, they’re like your true business key performance indicators, and you’re gonna… Like if daily active users, if you’re all about daily active users, then you’re gonna put that on a pedestal and you’re gonna govern the crap out of it. If you want clicks on blue buttons, I don’t know what you call that, that would still be… You’re still counting something. So definitionally, it still seems like a metric. But I wonder if that’s part of it. If you get alignment and a little focus from what are the 5, or 10, or 20 things that we, you know, revenue, orders, active users, subscriptions, renewals, really nail down and say, manage the crap out of that. And this other stuff, let it be a little Wild West.

0:27:05.1 MK: To be honest, that was actually one of the reasons. So we use Mode and Looker at Canva, and we were considering using Tableau at one stage, and one of the reasons that we went with Looker is because we had all worked at companies where everyone was using calculated fields in Tableau and had created various slightly different definitions. And the idea of being able to use LookML to define the metrics and make sure they were standardized was a problem we wanted to try and solve very specifically.

0:27:38.8 BS: Yeah. And I agree with the idea too, that it’s like there is a distinction between a metric and a chart, like everything that you create is not a metric and shouldn’t be. To me, businesses should have a handful of metrics because that’s all anybody can actually understand. And that’s all that anything anybody can really meaningfully make decisions about. If you’re trying to optimize 20 metrics, you’re not optimizing anything, you’ve got 20 people who are all just running in different directions. And there was a smart post, I thought from the folks at EPA who was like a A/B testing modern data stack, A/B testing tool. It was about how Airbnb uses metrics. They were also folks out of Airbnb and they… Essentially, were saying look, they had one, they had this… It was like bookings per night or something, it was like total bookings, I think. It was essentially the volume of…

0:28:24.8 MK: Yeah, total bookings. Total nights booked.

0:28:28.8 BS: Yeah, something like that. So the only way that they… It became important because it was like everybody could talk about it, it became… It was the thing that was in everybody’s head. Every engineer was like, “How does this improve nights booked?” Every product manager thought about it, people in the marketing team thought about it. With every other different metrics, you’re kind of all optimizing these small goals, but never really moving in the same direction. So I think, yeah, you have to have some sub-stuff and you can’t run a business on one metric, but you also can’t run a business on 1000.

0:28:53.8 TW: Yeah, which I think is… It’s kind of when you say that modern data stack it comes out… In all these platforms coming out of product time. It does seem like… It was interesting to reading up and daily active users gets kind of used as the example, and I can’t remember. Moe, you’ve talked, I don’t know if it’s monthly active users or something. Some flavor of that, which there are kind of the modern, newer, younger companies, especially when they’re kinda digital first companies like Airbnb, or Uber, or Canva. But then you turn around and you’re like, what about brick and mortar, or pharma, or healthcare, hospital systems, CPG. Some of those, there’re times where I think, yeah, they struggle a lot more because they don’t have this digital connection to the user and the conversion. And I sometimes worry that some of the problem-solving is kind of for a subset that has user-level digital data with a quick conversion, and I wonder how it maps to other ones.

0:30:11.8 TW: I feel like Moe, you and I have kind of had this debate.

0:30:15.2 MK: I’m not sure I understand your question, because what I’m trying to get my head around is like, yes, well, I look at some of these modern companies, and I include Canva in that as being willing to try things that maybe more traditional companies haven’t gone down the path of looking at or might have more traditional data teams, or data structures, or whatever the case may be. But ultimately, I still think they’re trying to solve the same problem. It’s just that there’s less friction to solve them, ’cause you’re at a company with younger people or whatever who are more willing to try different…

0:30:49.8 BS: No, no, no. I mean, I…

0:30:51.9 MK: But try to explain to me why would this metrics problem not exist at a healthcare company? Wouldn’t they have the same issues?

0:31:00.2 TW: Let me take pharma with the fucked up US system of where that’s a big money maker for companies, but if you’re in pharma, you’re like, well, there’s one metric, it’s prescriptions. How many prescriptions or maybe how many doses, or some flavor of that. That’s the metric you care about, but then you actually say, “Well, wait a minute, what is marketing doing to drive those?” Well, it’s kind of a little privacy thing, you’re not looking… Marketers aren’t saying, “I’m targeting that person who went to their doctor, and had this interaction, and then went and did this, and then… ” So you don’t have that traceability. Take an example of, I’m selling pet food where people are buying it at Walmart, or Target, or the grocery store. I’m marketing people, they all have dogs, and a lot of the CPG or FMCG in Europe will head down the path of saying, “Aha, the solution is we just need first party data, we just need to convince… Well, I don’t wanna… I’m not gonna sign up to have a one-to-one relationship with my fucking pet food provider. Who the hell cares?”

0:32:07.5 MK: It always comes back to pet food, doesn’t it Tim? It always comes back to pet food or toothpaste.

0:32:12.5 TW: I might have some trauma from a past…

0:32:14.3 BS: Tim, just download the pet food app on your phone, okay, get it over with.

0:32:19.1 TW: But there are lots of those. If you’re in financial services, sure, once somebody is a customer and you’re a big financial service is sure you have a lot of that interaction, but what I’m trying to sell to them.

0:32:30.9 MK: I still think Benn’s example fits. I worked in e-commerce and the definition of revenue is so fucking complicated because it’s like, are you including items that got returned, not return…

0:32:45.2 TW: Yes, wait a minute.

0:32:45.9 MK: How is this not the same for pet food?

0:32:47.9 TW: You just, you literally just did a… Completely shifted and said, “I have the same problem in an e-commerce.”

0:32:55.0 MK: Well, I must because Helbs is smirking. Yes.

0:33:00.2 MH: I am laughing.

0:33:00.3 TW: But I was saying I think there are places where having that finite set of metrics is hard to have the finite set and you turned it around to a pure play e-commerce company, still falls in the world of pure play e-commerce what about when you’re selling to in-store and lots of anonymous… I’m not saying the problem, doesn’t exist, I’m saying…

0:33:21.8 MK: But I am not talking about… Oh my God, I feel like we need a whole another podcast to see so you and I can get this out of our system.

0:33:28.1 TW: We’ve gone down this path before and it hasn’t worked, so now Benn’s just gonna make some popcorn and be like “What the… ”

0:33:33.5 MK: I know. I feel really bad for Benn. But I don’t think I understand… I still don’t entirely understand the problem, like I get that, yes, it’s not mapping one-to-one with users, but the problem we’re trying to solve is that everyone’s using the same definition, counting the same way so that reporting is consistent.

0:33:52.8 TW: I was responding to the finite set of metrics, that we need to kind of lock in on a limited number of… And looking at many… Or I think… I do think it’s easy, if you’re a pure play e-commerce guess what, you care revenue, profit… I mean there are three or four… If you’re a subscription company, yeah, you’re looking at active users, new subscriptions, whatever. There are, I think many industries where marketing is in digital and the data they’re collecting is definitionally, there’s a big gap in what you can see, so you wind up going to these what you hope are leading indicator metrics, you hope that you can do modeling to get to them, you hope you can figure out the connections between what you’re doing with TV, linear or connected or paid social or whatever, but you have a… It’s a lot harder to get to the narrow set of metrics that you care about, I think.

0:34:49.3 MK: Okay, I feel like we’re not in disagreement anymore, but… So now I’m gonna turn it back to our actual guest. Benn one of the things I keep thinking about… I wanted to ask this basically since we first jumped on the show together. I have been through so many migrations of data viz tools, warehouses like you name it, I’ve had to navigate the team through one of those, and one of the things that I’m sitting here thinking about is like, “Cool, so I’m gonna tell my business that we’re gonna spend three months building a metrics layer. What do you think?” And all I can imagine is like eyes glazed over… I somehow got people really excited about a data warehouse migration, I think ’cause we were moving to Snowflake, but I just feel like getting the buy-in for this would be really tough for a lot… And I know lots of data projects fall into this category, right, but have you seen companies that are doing it well or have done a good job of getting that buy-in and getting the business to understand why it’s important?

0:36:00.7 BS: No.

0:36:02.2 MK: Ooh. Okay.

[laughter]

0:36:06.0 BS: Half. So there are a handful of companies that I know that have done this with metrics tools. Now, get it. Metrics layer tools have been around for, I don’t know, six months. DBT’s is not out yet. Transform’s is the only one that’s actually on the market. They had a few beta customers. The beta customers liked it, how beta customers go, you basically sit next to them and you implement it with them, and so yeah, I guess that works. I think that people did it with Looker and Looker is kind of the same thing, so if that can get people to do Looker, you can probably get people to do this. I think to your baller proper point of like… Say that you have it, say you’re using Looker, so you’re using Tableau and you’ve got DBT and Tableau and Looker, and you’re like, “We need to move all the this stuff from Looker into metrics layer tools.” Yeah, it is probably a hard thing to do. And I think this is one of the reasons that one of the hardest things that Looker always had to do, in my understanding from what we hear about them is it’s exactly this. It’s like there was the same thing at Looker and they invested a ton in customer support and success and enablement to help people write the LookML because it was expensive upfront thing to do.

0:37:18.0 BS: Metrics layer tools probably have the same fight and there is a back door-ish way to me, to make it where it kinda can be like the DBT thing where it just sort of slips underneath where it already always exists and doesn’t have to be disruptive. The way that DBT is doing this, for instance, you can create metrics that are entirely additive, you can continue using DBT exactly as it is, and then these metrics can just be things that are now accessible going forward and you can kind of migrate stuff to them if you want to and all that. Okay, that might work. My suspicion is it something that takes a longer time than that to get to where we… If this becomes a standard, it will take some time for people to bake that in. It’ll take some time for people to migrate. It’s not something that was like, “Oh, of course, that’s easy. We just layer this on.” I am somewhat of an observability skeptic, but if that is a thing that everybody wants, that doesn’t require you to migrate anything, it’s just like tack on another tool and presumably push some buttons and you get all your things. This is very much not that, and so A, you’re right, there is a migration problem.

0:38:19.8 BS: I don’t know if anybody is sure how to solve that yet, other than sort of the promise of the value and hope that it’s not too much of a pain. And then if sort of the market proves that it’s worth it, then we’ll probably all do it.

0:38:30.5 MH: But I think there is a window of opportunity because a lot of companies are going into version 2 of whatever this stack was, and they’re noticing that they’ve got a ton of stuff that doesn’t match anything, nobody knows what it is. And nobody can do anything with. And so they’re having to encounter the cleanup of all this data debt or tech debt if you will. And I think that’s the insertion point for tools like this potentially is like, “Okay, as you move forward from here, reassert a new structure, a better governance, let’s put this tool in as we redo this, so that we’ve got clean all the way up. And I think you have a much more receptive audience at that point, because of the amount of work to go back through everything and delete all the junk is significant. It’s a lot. It’s a lot of work. And I don’t think it’s understood very well by most businesses today. Because they’re like, “Well, I need these Looker reports. And at the business level, they have no idea that man-hours are going to be required in engineering work and analytics engineering work, Tim.

0:39:35.0 MK: We didn’t tell them. We didn’t tell them about how long… We didn’t even tell them LookML or that we had to do that piece of work.

0:39:41.8 BS: But I think, Moe, you have to tell them, you have to tell them a little bit. Only because it’s gonna require a significant investment. That’s the only reason why, is you just don’t want people coming after you for stuff on the wrong timelines.

0:39:55.2 MK: I think they got that. But, Benn, one of the things I also wanna understand is as we keep talking about the metrics side we keep talking about tools, and I harp on about this all the time, but I don’t know where the industry is at at the moment with buy versus build in-house. Do you think this is something that people, like companies can build in-house? Or do you really think that you need a tool to solve this?

0:40:18.8 BS: I think it’s pretty tough to build like, it’s… Yeah, I think it’d be pretty tough because it is… I think, to me, it is akin to building a visualization tool. So Node is a visualization. Node has a fair bit of visualization technique. If you’re thinking about building a visualization tool, my God, do not do it. [chuckle] It is the sort of thing that… There’s this famous bit about when people started Dropbox, like somebody on Hacker News, or something was like, “Isn’t it just an SFTP server with this thing. I could build this in a weekend. Hup a bup” And like, maybe you could. Visualization tools are very much that. And I think the metrics thing is very much that where it’s like, “Oh, how hard can it be? It’d be like, I choose a dimension, I choose a couple of dimensions. I could use a metric, I could use a couple of dimensions, it’s just a simple formula. Just put, select dimension, dimension, computed formula from table, spit out a metric.” Yeah, it is. That’s all it is, for a day. And then you’re like, “Oh, but where do we add these other derived metrics? What if we wanna add stuff where we’re starting to have more complicated filters? What if we have something that has any sort of joins to it?”

0:41:22.4 BS: You immediately tumbled down this very, very deep rabbit hole, that is basically the entire tech that made Looker worth $2.5 billion dollars. I think the same is true for visualizations that what I know from experience, where it’s like, “Oh, isn’t it just choosing X-axis and a Y-axis and draw it on a chart, there’s a bunch of charting software out there.” It’s like, yeah, but now people wanna be able to change this, they want to change that. And they want to aggregated in different ways, they want to have tooltips that show one thing, and they want to show like, “Don’t show me the percent… Differences of these bars shown me compared to these other bars.” The amount of computation that requires to go into it is endless. And it’s not really an 80/20 thing, it’s not really something where everybody wants to do this other thing, there’s this long tail of weird stuff that most normally people don’t care about. It’s a much gradual, more gradual curve than that, where you’re never really off the steep part of the slope on the long tail, you’re just kind of gradually descending into things that you got to keep building.

0:42:10.3 BS: And so I think that’s what this will become as people will be, “Yeah, probably I can build that.” and then you’ll build it in a weekend, and it’ll be okay and then a month later you are like, “Oh, my God, I have… I have an endless roadmap of things to create.

[chuckle]

0:42:20.7 BS: So… There’s one other point I would make about the difficulty of adopting this too. And this is a challenge for I think, the transforms of the world, perhaps less so for DVTs is, the tool itself doesn’t have any immediate value. So, if you have a metrics layer tool, you can’t sell it alone, you gotta sell it with something else. And so DVT has the advantage of they already have, it’s already packaged with something else that people are using. So they don’t actually have to convince people to buy a front end to it. But if you are just building, say you’re building Malloy which is this Looker standalone thing. You got to build a client for it or something. You gotta be able to show people, Once you do this, what do you get? And the first thing to check was can I just have a simple dashboarding tool that lets me do it? And all a sudden, you’re just a BI tool. And so that’s apparently where Looker actually came from, is the founders built LookML, and then they built the BI tool out of customer need because they had to show why LookML existed.

0:43:18.5 MK: Wow.

0:43:18.7 BS: Yeah.

0:43:19.2 MK: Mind blown.

[laughter]

0:43:22.4 BS: So that was a while ago, so there may be… It doesn’t mean that will happened again, but certainly, it’s difficult to build a piece of middleware that you have to build… You have to have something before you have something after it for it to be at all useful.

0:43:35.1 MH: Okay, everybody, it’s for the Conductrics quiz, the quizzical query that is a conundrum for my two co-hosts as they go toe to toe on behalf of our listeners. Before we get started, let me tell you a little bit about Conductrics. The Analytics Power Hour and the Conductrics Quiz are sponsored by Conductrics. They build industry-leading experimentation software for AB testing, adaptive optimization, and predictive targeting. To find out how Conductrics could help you in those areas of your business go to Conductrics.com. Alright, let’s get the quiz started. Tim, are you ready to know who you are? Competing on behalf…

0:44:19.0 TW: Tell me tell me, tell me, who, who.

0:44:22.5 MH: Okay. And I wanna try to pronounce her name correctly. I believe her name is Shu Yen Shen. And Moe you are competing for a listener named Michael. So that one is easy. Michael Krakow, Shu Yen Shen Xi and Michael Krakow. Alright, let’s get into the quiz. Alright, Moe and Tim, I seem to be waiting like ages for Michael to finally join the podcast recording session. When he finally does login and apologizes. “I’m so sorry for being late again. But I was waiting forever for a SQL query to finish so I could get some analysis out to a client. They have a massive database I have to hit each time I run this report.” Moe slightly irritated replies. “This seems to happen a lot. Have you thought about spending some upfront time to set up an OLAP cube or some other denormalized data structure so your analytics queries can run faster?” “Speaking of OLAP”, Tim chimed “Did you know our old friend Edgar Codd coined the term for OLAP Online Analytical Processing.

0:45:35.4 MH: “He’s the relational database guy, right?” I replied somewhat surprised and exceptionally pleased with myself for making the connection… I’m not sure why I’m reading, all these editorial.

0:45:46.1 MH: “That’s right,” replied Tim. “The one who came up with the very first database query language Alpha”. Which you might remember from another conductors quiz long ago, and also showed what the relational calculus is equivalent in expressivity to the relational algebra, which SQL is based from. “And,” added Moe, “He came up with the 12 rules for relational databases, and this is now our question, which one of the following is not, one of Codd’s 12 rules, is it A. Comprehensive data sub-language role, B. Guaranteed access rule, C. The information rule, D. The systematic treatment of no values or, E. Third normal form rule”

0:46:35.1 MK: Well, that’s not the direction I thought that was gonna go in.

0:46:37.9 TW: This is for SQL? This is 12 rules for SQL?

0:46:42.9 MH: 12 rules for relational databases.

0:46:44.5 TW: Oh, for relation… Oh okay.

0:46:47.2 MH: And I know this would be particularly difficult for you Tim, because I’m sure you’re now only working in, what? Star cluster… What’s the new latest database structure?

0:46:57.9 MK: Data Mesh.

0:47:00.3 MH: Oh, Data Mesh. There you go. No, anyways. So those are our options. What thinketh you?

0:47:06.0 MK: I’m gonna take a stab and try and rule out C.

0:47:13.2 MH: Rule out C, the information rule. So when we’re asking which one is not one of Codd’s rules, you are correct, that is one of the 12 rules, the information role, so C is out of there. So we have A, B, D and E.

0:47:30.9 TW: So I have… If I was gonna guess which one it was, I think I know what I would guess, but I wanna go for the elimination instead ’cause that’s better odds. So I am going to…

0:47:38.3 MH: Okay.

0:47:40.2 TW: Say A, is not one of the 12 rules.

0:47:43.9 MH: No, you’re supposed to be guessing which of the following is not…

0:47:45.4 MK: Yes.

0:47:46.7 MH: One of the 12 rules…

0:47:48.9 MK: That what he said.

0:47:49.1 TW: Right. Oh, I’m sorry, I’m saying A is one of the… No, I think A is one of the 12 rules.

0:47:55.2 MH: Okay.

0:47:55.7 TW: Yeah, I think A is one of the 12 rules. I wanna eliminate A.

0:47:58.3 MH: Yeah, that would change your…

0:48:00.8 TW: Yeah. Okay.

0:48:02.2 MH: Okay.

0:48:02.8 TW: Yeah.

0:48:03.4 MH: So, I’m glad we clarified that because A is one of the 12 rules, and as you spoke it, you would have been eliminated at that point, but trying to help you out.

0:48:13.0 TW: Thanks for the leniency on the interpretation.

0:48:16.5 MH: So comprehensive data sub-lang… I knew what you meant. [chuckle] Anyways, alright, so that means we have, B. Guaranteed access rule, D. Systematic treatment of no values and E. Third normal form rule.

0:48:28.8 MK: I’m going to try and eliminate B.

0:48:32.5 MH: B. Guaranteed access rule. You are correct. That is one of the 12 rules. You do know a lot about these 12 rules. So nicely done.

0:48:42.9 TW: So well let’s see If I can screw it up now. I think the one that is not is the E, the third normal form rule, ’cause you can have first or second or third normal form, so I think, E. Is not one of the 12 rules.

0:48:55.1 MH: So you’re saying that’s one of the… That is the answer.

0:48:58.1 TW: So I am selecting E as the answer.

0:49:04.2 MH: Okay, that is correct. Third, normal form rule is not one of Codd’s 12 rules, while the degree or form of normalization is an important design component for a database, the normalization form was not one of Codd’s 12 rules, which actually is more like 13 rules. Alright, so I guess Tim, that makes you the winner, so Shu Yen Shen, you are a winner. So congratulations, and you’ll probably be receiving a little surprise from our sponsors, Conductrics, who are the sponsors of the Conductrics quiz here on the Analytics Power Hour. Alright, let’s get back to the show.

0:49:42.1 TW: But if you look at organizations that have… That are basically have acknowledged that they have an enormous amount of tech debt-based, they’ve got Teradata and Azure and Google Cloud Platform and stuff, duct tape together and bailing wire, and they’re going through a big reset. Is there… Knowing that the metrics layer tools are in their Nicene.

0:50:06.7 TW: But do you… Is your sense that there would be the potential to say, We can do this gradually that as we get metrics from being in a hodgepodge of competing definitions, we’re now gonna move this metric into the metrics layer paradigm, and it’s now gonna be a golden metric, and maybe it’s not one at a time, maybe it’s five at a time, or where it could be a gradual roll out, or would it be like no, probably to do a metrics layer, you’re probably gonna have to do some degree of a monolithic cut over to where you’re getting the bulk of this. I guess maybe it’s an agile versus waterfall question is to get to the adoption.

0:50:52.3 BS: I don’t know, I would guess that you could actually do it kind of piecemeal, Moe probably has a better answer to this than I do, ’cause it sounds like Moe, you’ve got some more experience with things that are pretty related to this. I think you can’t do that with a BI tool, you can’t do that when you’re like, go to this… Log into booker.com for these metrics and log into Tableau for these metrics, and we’ve got a giant directory of what tool has which numbers that doesn’t work.

0:51:19.5 TW: But if you’re swapping on your BI Tool and saying to get this metric, use the metrics layer now instead of using…

0:51:26.3 BS: So it shouldn’t be. No. It depends how all the these shakes out, who knows, But take mode for instant. So if you’re using mode for dashboards and stuff, you could create some of those dashboards on top of a metrics layers and some of those dashboards on top of just regular tables and to the user, it’s possible that’s all transfer… They may not even know if you’re just looking at the dashboard, you don’t need to know what’s under… In the same way, you don’t know if a DVT table is underneath it or not, you may assume that it is, but it could also just be a bunch of raw stuff, you don’t have to replace everything at once, you can just replace the things as they need it. So I think the same could happen here, it depends again on how it actually gets implemented, if you’re using transform for instance, and transform’s a little bit, more of a metrics directory type of thing. Then you’re in the place where it’s like, Oh, some metrics live here, look there first, but if they’re not there then go here, that feels harder to roll out, not all at once.

0:52:17.1 MK: We actually did just roll at a new tool that does something like this, I’m not gonna lie, I’ve opened it three times, so I haven’t looked at it in great depth, but it’s called caster, and the really nice thing about it is it tells us.

0:52:31.6 MK: Like what every table… Let’s say, so I type marketing like a 1000 times a day. If I wanna look at anything that has marketing in the title, it tells me not only what tables in the data warehouse have that word in it, but also it connects to Looker and tells us what dashboards do, like what uses those tables, which is kinda nice, but that’s also, I guess, a symptom of us having 50,000 million tables and no one knowing where stuff is.

0:53:00.1 BS: Yeah, and there are certainly things like that too, that are the… That there’s like the all of the broad definition of observability, discoverability it’s the that lens, the casters, the stimulation, Collibra. That whole collection of things here.

0:53:20.1 TW: Well, this is getting bigger by the day.

0:53:23.0 BS: Yeah, it is tons of these things that do this kind of tracking of, “Okay, these are the assets that you use these things.” That kind of stuff, so.

0:53:32.3 MH: Speaking of tracking things, we’re headed out of time, but this has been excellent for a brief moment there, I felt like I was part of the new data movement going on, and that’s new for me, ’cause I feel very much in the outside of this world, but no. Benn, thank you so much for coming on the show, sharing that, really pleasure having you.

0:53:51.7 BS: Yeah. Good to be here, I appreciate it.

0:53:53.6 MH: All right, well, one thing we do is we like to go around the horn and share something that we think is interesting, called The Last Call, and so let’s wind up our last calls right now. Benn you’re a guest. Do you have a last call you’d like to share?

0:54:09.1 BS: I have two things, One, I’ll like plug a… Well, I’ll plug three things, I’ll plug one, then I’ll do the shameless plug for the stuff, that the very specific we’re talking about CyRAACS stuff it’s on a sub-stack ’cause it’s 2022, so it’s benn.substack.com. It’s Benn, two n’s, I will plug someone else in Substack. So there’s a bunch of data-people out there having various conversations. One of the ones that I think is very good, that is sort of under-shared is called compiler queen. So it’s from Ashley Sherwood, she’s an analytics engineer, I believe at HubSpot. But it’s good, it’s good stuff would recommend checking that up. The other thing I’ll plug, which is just an article from like five years ago that I found, that I send to people all the time, and any opportunity I can. It’s an article on Vulture, and it’s called, what is the name of this, “Why do corporations speak the way they do?” And it’s essentially about phrases like Synergy and who’s gonna run point on this and all those kind of stuff, it’s like these very difficult to participate…

0:55:12.6 TW: Is that in the metrics layer?

0:55:16.1 BS: No. It is this…

0:55:16.2 MK: Like half his word is always context. It’s like, I’m just gonna share some context with you all. Let me get some context on this, it’s like, context is out.

0:55:25.0 MH: Let’s put a pin in that Moe, we’ll circle back.

[laughter]

0:55:32.3 MK: Oh, wow.

0:55:32.4 BS: Exactly.

0:55:32.8 TW: Yeah, yeah.

0:55:32.9 BS: But there is like a sort of data point in this, which is… The point of the article is based like we use this kind of language to kind of hide what we don’t know, that like simple language means you can’t hide and complex language sort of gives you impression, you know it’s time.

0:55:45.6 TW: Yup.

0:55:46.2 BS: Data-people have a tendency to do this a lot where it’s sort of like over-complicate stuff, with here’s a bunch of chart like… Throw a bunch of numbers and charts and stuff at people to the point…

0:55:53.3 TW: Blind them with science.

0:55:57.8 BS: Where it’s like, I’m overwhelmed by this and I can’t possibly understand what any of this means. And I don’t think it’s intentional. I don’t think it’s trying to hide, but I think there’s the same thing applies, where it’s just like simple stories are better, simple charts are better. Tell people clear stories. Use small words. I don’t know, I’m too long winded when I talk on the internet, so I probably should follow that advice, but…

[laughter]

0:56:15.3 MK: I love this article though, I actually can’t wait to share this with the team. I feel like we’re gonna get some good laughs about it ’cause we always like… We’re like, “Oh, we’re gonna take things async, so we’re gonna talk about over Slack List it and get off soon. Alright.

0:56:29.1 MH: Yeah, exactly. Well, Moe in that light, What’s your last call?

0:56:33.9 MK: So I watched a really cool video the other day by the, I think is it Fox Media or just Fox?

0:56:43.4 MH: Sure.

0:56:43.5 MK: Fox media. Anyway…

0:56:44.7 MH: Yeah, one of them.

0:56:44.8 TW: It’s Fox Media Group, yeah.

0:56:45.9 MK: It’s called how American conservatives turned against the vaccine, and I don’t know, I just really loved the reporter, the way she used data visualizations through that video just like blew my mind. And it’s one of the times, I don’t know, she starts using clear slides and she writes on them, and I just watched it in complete awe of the way she was able to communicate the story and really get people to absorb data without even necessarily feeling like you are. Like you’re just, I don’t know, I thought it was beautiful.

0:57:20.2 TW: And it completely turned the tide about vaccine hesitancy in the US, it’s amazing.

0:57:25.3 MH: We’re crushing it over here.

0:57:27.1 TW: It’s a real uplifting topic.

0:57:32.2 MH: Alright Tim, I’m sure you can do better with your last call, sir.

0:57:36.5 TW: Oh, dear. Well, since we’re talking about the modern data stack a little bit, or a piece of it, Ibrahim Elawadi, has been… He’s from Greenpeace, and he’s kinda hitting the road with a talk about the modern data stack. So we saw him at SUPERWEEK. I did not see him, but I believe he presented very similar information at MeasureCamp North America, and he will be at Marketing Analytics Summit in June in Vegas talking about it. Which as it so happens, so are we. So, if you would like to see the first opportunity for Moe to take out her pent-up aggression towards me in person, this will be the first time we’ve seen each other since SUPERWEEK 2020.

0:58:21.9 MK: Yes.

0:58:22.2 TW: But we will be… The whole podcast crew minus Josh, alas will be at the Marketing Analytics Summit in Vegas, June 20th through 23rd. We do have a discount code if you’re interested and ready to get back out in-person, so hit us up, but also there is a modern data stack session at that very in-person conference. And I have like four other last calls, but I’ll leave it at that. What about you, Michael?

0:58:53.9 MH: Alright, that’s probably what you’ve been wondering is what my last call is. So, I recently ran across something on Twitter and then as I dug a little deeper, I was just super fascinated by it, and I actually already posted this on the Measure Slack, but a guy with the name of Danny Richmond took GPT-3 and made it into a regex converter. So you write plain English and GPT-3 changes it into a real regex. And I was just like, “That’s really cool.” Because we’ve all been there having to write some regex, and this is a little tool he built that basically allows you to write what you want the regex to come out as, and it’ll just pump out the regex code for you.

0:59:36.4 TW: But did you see the follow-on that Smith, Joe Hanson wrote, “The GPT-3 thing to come back and with auto-generate critiques of… ” Well, choices were made for the regex?

0:59:46.6 MH: Yeah, I think.

0:59:47.7 TW: No that’s totally fictitious.

0:59:51.6 MH: But it’s part of the whole process. Anyways, I don’t know if you find yourself having to write regex too often anymore, but it always pops back up every once in a while, so good to have some tools. Alright, those are some great last calls, thank you everyone, and I’m sure as you’ve been listening, you’ve been thinking to yourself, I would like to know more about this topic, or I have comments I’d like to share. We would love to hear from you. The best way to do that is through the Measure Slack group or through Twitter, or you can also send us an email at contact at analyticshour.io. And if you haven’t already, or whatever platform your listening to the show on, we’d love to get you to give us a rating and a review because we’d love to see how we are helping or what other things we should be covering on the podcast, as always, no show would be complete without a huge and hearty thank you to our producer, Josh Crowhurst. Thank you, Josh. You help us out immensely, and we appreciate it, so, there we go. That’s the show. Benn, once again, thank you so much for coming on. We really appreciate it.

1:01:03.2 BS: Yeah, thanks again for having me. This was fun.

1:01:06.4 MH: That was very cool, very awesome and current topic, and I know I speak for my two co-hosts, Moe and Tim. When I say no matter what metrics layer you use today or tomorrow, remember, keep analysing.

1:01:24.1 Announcer: Thanks for listening. Let’s keep the conversation going with your comments, suggestions and questions on Twitter at Analytics Hour, on the web, at analyticshour.io, our LinkedIn group and the Measure chat Slack group. Music for the podcast by Josh Crowhurst.

1:01:40.7 Charles Barkley: Chose smart guys want to fit in, so they made up a term called analytics. How analytics don’t work?

1:01:48.9 Tom Hammerschmidt: Analytics, oh my God, what the fuck does that even mean?

1:02:02.1 MH: Positivity energy. It’s for myself not for anyone else. [chuckle]

1:02:06.4 MH: Alright, here we go, in five, four… There’s a bunch of stuff that happens on the podcast that has become ingrained in its lifetime, and so there’s mergers that… The rituals must be observed.

[chuckle]

1:02:24.3 TW: And they all done in 200 episodes this?

1:02:28.1 MK: 190.

1:02:28.9 MH: Yeah, we’ll do 200 sometime middle of this year. Yeah, we started the podcast in 2015. Is that right, Tim?

1:02:36.0 TW: 16, 15.

1:02:37.1 MH: 15, 16 somewhere. Yeah, one of those numbers out of Pre-covid, yeah so…

1:02:44.4 BS: Different times. Somewhere between two and 50 years ago.

1:02:48.8 TW: Yeah, and we’ve moved on from immediate family to extended family its…

1:02:51.4 MH: That’s right, that’s right. I’ll listen to basis.

1:02:58.6 TW: Rock flag and one metrics there to rule them all.

Leave a Reply



This site uses Akismet to reduce spam. Learn how your comment data is processed.

Have an Idea for an Upcoming Episode?

Recent Episodes

#199: Media Measurement Revisited: Matched Markets, Media Mix, and More with John Wallace

#199: Media Measurement Revisited: Matched Markets, Media Mix, and More with John Wallace

https://media.blubrry.com/the_digital_analytics_power/p/traffic.libsyn.com/analyticshour/APH_-_Episode_199_-_Media_Measurement_Revisited__Matched_Markets_Media_Mix_and_More_with_John_Wallace.mp3.mp3Podcast: Download | EmbedSubscribe: Google Podcasts | RSSTweetShareShareEmail0 Shares