Business Intelligence. It’s a term that’s been around for a few decades, but that is every bit as difficult to nail down as “data science,” “big data,” or a jellyfish. Think too hard about it, and you might actually find yourself struggling to define “analytics!” With the latest generation of BI tools, though, it’s a topic that is making the rounds at cocktail parties the world over! (Cocktail parties just aren’t what they used to be.) On this episode, the crew snags Taylor Udell from Heap to join in a discussion on the subject, and Moe (unsuccessfully) attempts to end the episode after six minutes. Possibly because neither Tableau nor Superset can definitively prove where avocado toast originated (but Wikipedia backs her up). But we all know Tim can’t be shut up that quickly, right?!
- Rod Serling
- Apache Superset
- Camel case
- Snake case
- Attribution IQ
- Customer Data Platforms (CDP)
- (Video) Using Electron and node.js with Shiny to make standalone, interactive, Shiny apps (#useR2018 conference presentation)
- Manhattan Population Explorer (interactive visualization)
- (Podcast) The Tim Ferriss Show
- (Book) Essentialism: The Disciplined Pursuit of Less
- Meme-ready image of @benjamingaines called out by @stylerriggs
00:00 Michael Helbling: Hi, there. This is Michael Helbling for the Digital Analytics Power Hour. I’m excited because we are coming up on our 100th episode as a podcast. Moe, Tim and I have been brainstorming like mad, trying to figure out how to make this episode special. And we’ve come up with, I think, a great plan. We wanna make it special by including what’s most special about the podcast and that’s you, our listeners. So we want you to submit your questions we can answer on the podcast, and here’s the great thing. If your question gets selected, you win a great prize that’s unique to the podcast’s 100th episode. So, submit your questions. Here’s how you do it. You send a 30-second audio clip introducing yourself and then your question, in 30 seconds or less to firstname.lastname@example.org, again, that’s email@example.com and your soundbite or question clip, just include it in your email, introduce yourself and ask your question and then you’ll be in the running.
01:07 MH: Any question goes. If you get selected you win a great prize. We’re gonna take submissions from the time you hear this, up until September 15th. So from now until September 15th is when the submission deadline is. So submit your questions, we’re excited to hear from you, our audience, make you guys part of our special 100th episode. Alright. Let’s get it going. Oh, bonus points for anyone who includes the words “quintessential analyst” when speaking about Tim Wilson.
01:40 Announcer: Welcome to the Digital Analytics Power Hour. Tim, Michael, Moe and the occasional guest discussing digital analytics issues of the day. Find them on Facebook at Facebook.com/analyticshour. And their website, analyticshour.io. And now, the Digital Analytics Power Hour.
02:03 MH: Hi, everyone. Welcome to the Digital Analytics Power Hour. This is episode 95. You unlock this door with the key of having a lot of data. Beyond it is another dimension. A dimension of sound, a dimension of sight, a dimension of mind. You’re moving into a land of both shadow and substance, of things and ideas. You’ve just crossed over into the BI zone. That’s right, we have a show taken straight from the ’70s. Well, sort of. It’s about BI. Picture this, a quintessential analyst pursuing a podcast. His name is Tim Wilson.
02:44 Tim Wilson: Or Rod Serling…
02:46 MH: Or Rod.
02:47 TW: For this episode.
02:49 MH: No, I’m Rod Serling, sort of, but really badly. [laughter] Also, an analytics manager in Sydney, Australia whose past is shrouded in mystery. Her name is Moe Kiss.
03:00 Moe Kiss: Hey, Tim, how’s it going? [laughter]
03:01 MH: And finally, a talkative man who makes analogies and tries to stretch them way too far, but his jokes are universally accepted as amazing. He’s Michael Helbling. Okay, but seriously. We wanted to take a look over into the BI world to just sort of see where this intersection’s happening with our world of digital analytics. ‘Cause there’s a lot going on in both spaces that I think is kind of interesting. And we’ve got a great guest joining us on this show. She is the lead solutions architect at Heap, that’s an analytics company, and prior to that, she was with Teach for America. Welcome to the show, Taylor Udell.
03:42 Taylor Udell: Hi, guys. Thanks. Really happy to be here.
03:45 MH: Awesome. Alright, well, Taylor, just to first get everybody to get to know you a little bit, just tell us a little bit about yourself, what you do in your role at Heap and…
03:55 TW: And why Heap is not a BI company.
03:56 MH: Yeah, well, that too.
04:00 TU: Alright, so, yeah, I’m the lead solutions architect here at Heap. Been at Heap for almost three years now. And basically what that role means is, I’ve worked with all of our customers, both pre and post-sales, really implementing and building up analytics functions in their companies, and helping kind of create data driven orgs, especially as the analytics field has really continued to grow. I forgot your second question already, but I think it was something about BI, and how I see it emerging or why BI is not what Heap does. Heap is essentially… We focus on web and mobile analytics, so when it comes to the traditional analytics tools there’s a lot of tracking, coding, and implementation, and Heap’s main function is kind of automating that capture and implementation aspect, by auto-capturing all the user behavior. And then we collect it, organize it into events, and we pipe it downstream to data warehouses as well as having an analysis layer. So we partner with a lot of really great BI tools, but we are not a BI tool.
05:05 MH: Nice. Tim, what’s the definition of BI?
05:11 TW: There goes the whole show. I mean I… It’s funny, I’ve managed a BI team, now going back, I think, 11 years from now. Which is funny, ’cause at the time, the BI team, we were the kind of the centralized shared service for doing, I would have said reporting and analytics. And we had a data scientist on the team as well. And so that was skewed. I think my definition of what I thought BI was, is there’s been this second round generation of BI where Tableau and Qlik and Power BI and Domo have kicked in. It seems like BI is much more perceived as the data democratization platform of letting business users access dynamic, sliceable, drillable data. And so, it winds up in kind of an exploratory, but not in typically the kind of data science deep analytics modeling predictive way. It’s kinda more enabling the business. What analytics is, I have no freaking idea.
06:21 MH: Yeah, I think I agree with that. I slice it into three layers, personally. Like that bottom layer is sort of all of your data gathering, the second layer would be all your data transformation or integration, and then the top layer is how you present it out. And there’s lots of, and I feel like a lot of BI tools focus on one of those three, or try to kind of interact with all three of those layers in different ways, and with varying degrees of success in a certain sense.
06:52 TW: But when you say gathering, though, they’re gathering it from… It’s not the actual… It’s not the raw collection.
06:56 MH: Whatever system.
06:58 TW: Some other system like a Heap or Snowplow. They would be…
07:02 MH: Yeah.
07:03 TW: Whatever is actually doing the collection. I guess that’s collection versus gathering.
07:06 MH: Yeah, that’s fair. Data storage or data… I don’t know, where data’s got to sit.
07:16 MK: Okay. Alright.
07:17 TW: Moe can’t take any more.
07:19 MK: I’m losing it. I’m sitting here like, I just… Don’t you think it’s just all the same shit? It’s like when web analytics became, started to be called digital analytics, like we used to have a BI team and an analytics team, they did a lot of the same stuff. The only difference was that the poor BI team ended up building way more dashboards. But is it really actually that different? Do you really need to delineate between the two in your team?
07:42 MH: No, Moe, you’re absolutely right.
07:44 MH: And it’s been a great show. Thanks everybody.
07:54 TW: I don’t know, well… Taylor.
07:56 MH: At the risk of not letting Taylor say anything. I wanna hear Taylor, maybe what your thoughts are on what Moe just said.
08:04 TU: I kinda have to take Moe’s side on this. I feel like maybe when I think about BI, I really just think about the tools, but primarily the users, just like the analytics team, you have the head of analytics taking on these major initiatives to consolidate all of the customer data and have that BI tool be that visualization layer to help democratize data, but at the end of the day, the function is bringing analysis and analytics to all of the varying teams at your company. So, I don’t know if we have to draw the line.
08:33 TW: Well, but I… To me, a lot of times BI does get defined as the tool. So I would say that’s where I came from, BI was the label, and it was kind of all of that, and that seemed to make sense. We had… Market research was in BI, ’cause that made sense, that it was qualitative research data and that logically made sense. But when, in the market now, it seems like BI winds up being… And you even kind of headed that way, Moe, that part of it is the instrumentation in the tool to enable end-users, which is more of a engineering function, ’cause it winds up being sort of this is a BI tool and how do we implement, and enable, and train people to use the tool. And it is very… BI tools have some very, very strong hard line limitations within them. And in some analysts that’s where I think, I don’t think there are a ton of hardcore data scientists who are saying, “I’m getting everything I can out of my BI tool.” So I agree, they could live in a BI department. I just when I… I feel like when I talk to companies that have BI departments, what they’re really doing tends to be a lot more on the tool enablement and maybe even the building of dashboards, that’s not necessarily where the R, the Python, the SPSS, the SaaS, the predictive modeling is necessarily living.
10:07 MH: Yeah, I think that’s somewhat fair. I think the other thing that we probably need to create a zone for, is, businesses create a lot of data that is not related to marketing. So, there is finance data, there’s sales data, there’s HR data, all of those different areas, or functional areas of businesses need somewhere for their data to live too. And so, for me, the world of BI is more encompassing of sort of, well, where do you… What is all of your data? How do we put it all in a place where it can be, quote/unquote, used effectively? I’m not a fan of data democratization, because I exist in an oligarchy of understanding data, and I just wanna keep it that way. Yes, I’m a white man.
10:57 MH: No. We can get… It’s a whole other show topic. But that’s one thing, like digital…
11:00 TW: It takes it more to the enterprise… It’s more… It inherently becomes a more enterprise-oriented thing than a marketing or not?
11:08 MH: Maybe, but maybe not. Because think about it this way. Even in a start-up probably the least useful metrics are your digital metrics from the standpoint of actually understanding whether or not you’re surviving or making your goals, and those kinds of things. It’s, did I get customers? Am I… I mean, some of those metrics are important but not like the digital metrics we measure for digital analytics. Like conversion rate isn’t as important as, “Did I just get the number of customers I was trying to get for this quarter?” Or those kinds of things. And so, depending on what kind of company you are, you could be a small company or a very large company, but you need to use data in a lot of different ways, and I think that’s where digital analytics lives in the marketing world mostly, and there’s a lot of other worlds that I think BI touches that digital analytics doesn’t.
12:02 TU: So I feel like there’s this emerging trend, though, to take, I guess, all of that data that previously was siloed, like the operational data, the server side data, the performance data and kind of marry that with customer experience data, whether that’s digital from your website or from your app or from your email marketing tool and kind of marry them all together to create this very holistic view of what is your business actually. So how do these performance metrics actually impact things like conversion rate. And because storage is so cheap and there’s all these great ETL tools now, and you really have the ability to transform and customize the way your data appears in a warehouse or have that modeling layer on top of the BI tool or whatever you’re actually doing the analysis on. It’s a whole lot easier to kind of drive your company for it, or there’s a lot more pushes to drive your company for it based off the standardized all-encompassing view of the business versus just having like, “Oh, my operational data only affects the finance team, and my customer data only affects the marketing team or the product team.” Like all of that data is very intertwineable and the companies that I feel like are growing fast or growing well have done a really good job of centralizing that data to see that full story of their business.
13:21 TW: Are there two ways to… I like that, and it’s taking me back to some of the things… Like I remember when Adobe first rolled out Adobe Social and they said, “Oh, look, now, you can have a chart that shows your Facebook engagement on the same chart as your web traffic, and you can see if they’re trending together, you can visually look for correlations, that sort of aggregated data, which I feel like the Tableaus of the world kind of go to that level of saying, “Are… ” Let’s take time as a common factor or some common dimensions, and let you kind of pull data from different areas of the business, from marketing and operations, and sort of put them together, but separately, if you look at Google Analytics raw data in BigQuery or Adobe’s data feed, that feels like it winds up needing one. That’s where that’s kind of the more granular level data, that you’re really gonna work with if you’re doing machine learning type stuff. And how often does that sort of data get rolled directly into the BI tool? ‘Cause it’s more raw, it’s tougher to work with, and it’s larger, which means those tools don’t necessarily have great performance. For Heap, for instance, you’ve got all this raw collection you said you do kind of… Can integrate with various BI tools, are they… Are they pulling in your raw data or is there some aggregation that goes on before they pull it in, so that they can actually use it? And do you know, that might be an unfair question to ask?
15:06 TU: Yeah, so basically, because we have a retroactive ETL sync to BigQuery or Redshift, and a lot of our customers will actually put the BI tool on top and actually just query the raw data, it’s really easy to get counts or a basic funnel. But a lot of the time, what they’re doing is kind of two-fold, so they’ll have the raw data, maybe that’s more ad-hoc, maybe they’re actually doing that kind of analysis in the UI where you can actually drill down and do this ad-hoc investigative stuff a little bit more flexibly than with a BI tool. But then they’re also creating these aggregate user summary tables, kind of transforming the data into an aggregate roll up to better look at it in something like a BI tool. A lot of the time. So a little bit of both.
15:57 MK: It sounds like the lynch-pin, though, to all of this is a tool, it’s about visualizing something to the business. Do you think a BI team can be successful without necessarily dependency on a specific tool? Is part of the key to what a BI team does is that getting into a tool so people can visualize stuff?
16:19 TW: Them getting into it, or their users getting into the tool?
16:23 MK: Or giving stakeholders access.
16:24 MH: It’s the modern definition. Because I think BI back in the old days was just kind of whatever anybody wanted it to be. And it was basically a big honkin’ system that everybody had, but nobody used, like Cognos or something like that. And I think the… I don’t really know the answer, honestly. I thought I was going somewhere but…
16:45 TW: So Moe, what do you got, I know like you as an analyst will write SQL against BigQuery, maybe ultimately you take something and you drop it into Tableau and then you deliver it, and maybe there’s some interactivity or not, I don’t know. Within the iconic, what is the… Outside of the analysts and data scientists and the people who kind of live and breathe the data, what is the data access mechanism for the others, for the merchandisers?
17:12 MK: So, at the moment we’re going through a process, we are building out all of our reports in Superset. I was actually explaining this to a new person on the team yesterday. And the thing that’s actually been really good about it, is that it’s forced all of us to really think about what we’re pumping into the tool, because you can’t do a join in there. Every analyst or data scientist has to build an aggregated table to sit behind the report they wanna show. And what that’s done is, it’s actually forced us to do a lot of the heavy lifting in our code rather than getting the tool to do it for us, which, to be honest, is in my view, how it should be. That’s kind of been forced by limitations of the tool that we’re using, which has had an upside almost. But I…
17:58 MH: Moe, a long time ago I used a web analytics tool called Webtrends. [laughter] And it had a backend database that wasn’t related. And so to create any relationship between any dimension of measure, you had to pre-create them ahead of time, and then it would deliver your report with that information.
18:16 TW: ‘Cause it had an analysis table and a reporting table, that it generated once you created them.
18:21 MH: Yeah. And I feel like you just described that to me again.
18:27 MK: See, you could go back and be an analyst now, Helbling.
18:30 MH: Oh, no, I prospered in that world. I did a great job.
18:34 TW: Well, although in that world, it was interesting, because it had inherent limitations, right? You’d think, “Oh, I can just jack these limits up. I’ll make this… I’ll set this up, but I don’t know, Moe. That doesn’t sound like a… That sounds like…
18:49 MK: Okay.
18:50 TW: The BI team, this analytics team, is spending their time doing a bunch of data engineering to support dashboards.
18:57 MK: Okay, so let me explain it to you this way. BI tools like, love the shit out of them. But we had people dragging our entire Snowplow atomic events table into Tableau. Now, our atomic events table is freaking massive. That is raw, hit-level data… Like years of data and people were putting it… Dragging that into Tableau, connecting it in Tableau to our data warehouse, which has all of our transactional order data. And then being like, “Oh, this is taking 15 minutes to load,” and it’s like, “No shit. Of course, any tool in the world is gonna take 15 minutes to load.” It’s not like a specific tool thing. So what I’m trying to say is I understand that maybe it feels like we’re going back in time, but I actually think it’s forcing analysts to be smarter, and be like… I think we got lazy because we’re like, “Oh, the tool can just do all this stuff for us.” But the truth is we should be thinking more about how are we joining our different data sources? How do we aggregate to the specific level that we need for this piece of work that we’re doing? And… Because otherwise everyone just… It’s now all the trend. You wanna go back to the raw data at hit-level. And it’s too big. It’s too messy. I don’t know.
20:05 TW: Well, there’s the big… But you’re making a bunch of decisions when you’re doing that. And I will claim that the BI tool vendors marketing is touting… Not going to the raw data. They don’t talk about that. I agreed with you on that, earlier. Things break down. They’re not good at handling that, but the promise… What gets touted out there is, “Hey, now, you can go and slice, and the user can discover these things… All these things on their own.” And what you’re saying, which I don’t think I’d disagree with, is saying, “Well, no. We still need to draw some sort of box around it.” So they still can slice things an infinite number of ways, but it’s not an infinite to the fourth power number of ways. It’s like building PowerPlay Cubes back in the day. That was like a Cognos thing, was you picked which dimensions you could slice by.
21:01 MK: Yeah.
21:01 MH: Did they have a cool glove that you would wear? Is that… That’s a Nintendo thing. Never mind. [chuckle] Yeah, I… That’s sort of why, Moe, I don’t believe in data democracy, is stuff like that. When people take the entire database or the entire atomic event table and try to stick it in Tableau.
21:20 MK: Yeah, but that is what some of BI tools have done. They’ve made analysts lazy.
21:26 TW: I… I don’t…
21:27 TU: I feel like there’s also something to be said for this kind of middle ground that I feel like we’re talking in two extremes: All the raw data, or all of these very organized tables that are pre-aggregated. There’s definitely something to be said to having all of the raw data. There’s always questions that come up, where you really just haven’t thought of it, or you haven’t needed to track it, or it hasn’t been that KPI in the metric that people needed to slice and dice by. Making sure all of that raw data, or having… I think the real interesting part is how do you actually have that very flexible modeling layer that allows you to improve or change those kind of aggregated tables on the fly or with relative ease, without huge data engineering constraints to make sure that as these new questions emerge, or your new initiatives in the business change, or all of a sudden there’s a new leader and all of a sudden, you have to have this thing that you never had to have before. How do you actually build that scalable system, so that the raw data you actually have can be flexibly input into these tables to make your systems efficient.
22:36 MH: Yeah, Moe, you gotta work on your data virtualization. [chuckle]
22:41 S6: Exactly.
22:42 TW: Just to see if I’m understanding right… I think for years and years… And still a shortcoming of Google Analytics, when it comes to goals, which I feel like you move to a certain point as an analyst and stop using goals, because they’re only from the point that you set up the goal going forward. And generally, if you set up a goal you could say, “Well, I could set up a segment, and I could see this stuff retroactively,” but this thing called a goal just has this big limitation. If somebody comes in and says, “Oh, this thing actually matters, we’ve had this thing on our site for… ” This is gonna wind up sounding, I think like maybe I’m plugging Heap, which was not exactly where I was heading, but I think it’s a difference that, from my understanding. [chuckle]
23:25 MH: So a goal was something… It is an inherently aggregated thing. As BigQuery’s integration with Google analytics has come along, a lot of analysts will say, “Well, I just need to have a query. I just define whatever those criteria is.” It’s the same thing as building a segment. I think, from what limited looking at it that I’ve done, that Michael was kind of excited [chuckle] about. With Heap, you could go in at some point and say, “Oh, this action that was being tracked anyway. Now, I wanna give it a label.” It’s not necessarily called a goal, but I’m gonna kind of tag it as saying… This thing, no matter which page it happens on, is a thing and then that becomes retroactively available. Is that a… I don’t know if that’s… It seemed a lot clearer of a distinction in my mind.
24:04 TU: It’s a pretty… That’s a pretty great description. I think if you think of data virtualization, what Heap really does is try and take it to the micro-event level. So you have all of the raw data, all of that clickstream data, change on events, and you’re basically doing virtualization at an event level, so you’re wrapping all of the HTML classes, in this semantic label of an event and that actual raw data is never really altered and you just have like, “Oh, all of a sudden, I care about this button click or… Actually, this old button click and this new button click mean the same thing. So, I need to update kind of that layer of semantic meaning on top of them.” And you have that complete retroactive dataset, because that raw data is there.
24:48 TW: Yeah, okay.
24:50 MK: I’m curious to hear how often that comes up with clients. It sounds like that would be something… Is that used a lot?
25:00 TU: Like for us? Yeah, it’s like the… It’s pretty much the number one, I would say, reason why people really turn to a solution like Heap is because there had been a situation where actually just implementing all of the tracking code has resulted in skewed data, or missing data, or even downstream, if you think of all the joints you actually have to do because someone had camel casing and someone used snake casing and all of a sudden, it’s just like a mess, and it breaks everything. Just being able to control that virtualization there, that semantic layer at the event level is really a core value prop for Heap.
25:39 TW: I’m now thinking through, I just had a case with an Adobe client, where they had changed some tagging so they had basically two different values for an eVar that basically covered the same area of the site, and so we were working with the data feed, and at some point, they said, “Well, these two things really should be collapsed.” Well, the analyst who was working on it could say, “Okay, I can go back into Python and change things and basically collapse those.” Well, okay, he’s got a specialized skill. He could do it. A way a classic analyst might do it is say, “Oh, I’m gonna classify that eVar and I’ll use that to sort of collapse it” Which I think, Moe, is actually analogous to your point that it’s like, okay, the analyst needs to think, “Yeah, I need to be able to make these happen continuously over time. I would, in this tool, I’d use this classification now in Report Builder, or whatever I’m using to surface the data, it makes more sense to the user.” So it seems like every tool that it goes back to the raw data in some form, it’s got some mechanism to say stuff’s changing all the time, and part of the role of the analyst is to say, “Well, how do we remove some of that messiness because it really is just the nature of things changing over time,” and then by the time that gets surfaced to the end user, be it through Superset, or Power BI, or Tableau, hopefully they don’t have to worry about that.
27:07 MH: That’s what I think is like the emerging difference, is sort of like, if you take a digital analytics tool from a while ago, you have data collection, some kind of aggregation and analysis, it dumps out into a set of reports that are somewhat manipulatable and growing in their ability to be manipulated, and addressed in different ways. What I think is sort of happening is we’re sort of splitting these component parts up a little bit, in that we’re taking and saying, “Okay, listen. I don’t need my analytics tool to do all of this aggregation and stuff. What I really need it to do is just make sure I’ve got everything I need to have, and give me a vehicle to capture or virtualize around things that I might need that I don’t know that I need today in some kind of graceful way.” But then, now, we’ll use different tools like Redshift or whatever to go do the actual running of all that stuff, and then, leverage some other tool to then visualize that all out in some kind of way to all of our users, whether that be a BI tool or something like that. And so it’s almost like maybe that’s the intersection. Oh, I solved it, way to go, Michael. [laughter] Analytics is becoming BI, at least according to my definition at the beginning of the show. Congratulations.
28:35 MK: It’s funny, actually. I was at an event a couple of weeks ago, and someone asked me, they’re like, “Oh, I’m a business analyst. I really wanna move over to analytics.” And I was like, “Huh? It’s the same shit. You’ve got the same skill set. It’s not that different.”
28:51 TW: Should we talk to the definition of what business analysts wind up doing?
28:55 MK: I don’t know. I was talking to her about the work she was doing. Well, business analyst is a really generic term. But I was just like… I don’t know, I feel like people sometimes get really stressed about the definitions, and I think maybe we should just worry more about the skill set.
29:11 TW: But I think that’s actually a…
29:13 MH: Don’t you need a definition to be called it, Moe?
29:16 TW: I have worked at multiple companies where if you take this idea of a business analyst analyzes the business or the process and comes up with a solution. And I’ve never worked at a company where that’s actually what they do. I’ve… Business analysts, no offense, I mean, there’s some fantastic ones, but they seem to drift into IT, being the checklist requirements gatherers, scopers for what IT’s gonna do. They’re analyzing the request, which I see as being very different from analyzing customer behavior, or channel performance, or content effectiveness, or applying statistical methods at some point.
30:00 MK: No, I have actually met some business analysts who share what type of work you were just describing, but I wouldn’t really call that a business analyst. I think that… I don’t know why the hell they’re doing that stuff. I don’t know. That’s my own personal opinion.
30:12 TW: Well, because that’s what 95% of them are doing is the industry…
30:18 MK: But why? Anyway, I just…
30:19 TW: We don’t need to worry about definitions. The definitions are consistent, but dead wrong.
30:26 MK: It is! It’s dead wrong!
30:27 TW: It is, I agree, but…
30:28 MH: Uh-huh.
30:28 TW: Okay. Wow.
30:30 MK: I feel like a business analyst…
30:30 TW: Do you have business analysts at Heap is the question. And have we just insulted them?
30:36 TU: That’s alright, you didn’t insult anyone. Yeah, we’re an analytics team, and an operations team, and no business analysts.
30:42 MK: Oh, great.
30:45 TW: Do you guys have business analysts, Moe?
30:47 MK: No. But we… No.
30:49 MH: Wait. I was a business analyst. Was that bad?
30:52 MK: We used to have people that sat in BI, which I don’t actually remember what their job title was, because BI is now just all part of analytics and data.
31:01 TW: Wait, what was BI? What was the old BI? What was the scope of your old BI’s world?
31:06 MK: They did some… There were some that were a few more technical that did kind of a little bit more ETL work, but it was a lot of report building, dashboard building. And they’re still doing really similar work, but now they work in analytics and data. So, that’s why I just feel like the line’s so fuzzy. I feel like BI and analytics, and when you actually look online and see how different places define the difference between them, which is crazy, and it talks about BI looking at the past and analytics being more predictive. I feel like that’s just now the difference between analytics and data science and what everyone’s going on about now. We’re just using different terms for the same old shit.
31:46 TW: And it goes back to when we said analytics. It was a fancy word and we were just pulling reports. That was like… It’s bothered me for like, “Man, that’s a really… We’re putting on airs.”
31:56 MH: Well, yeah, and I think this show will not be a venue to promote bullshit around labels and titles that make people get paid more money for no obvious reason. [chuckle] So, let’s just lay that down right now. [chuckle] We’ve had a lot of topics, Taylor, over the years, where we’ve maybe covered some of those things. So, [chuckle] data science being basically an analyst from California. [chuckle]
32:23 TU: Yeah, we’re important out here.
32:25 TW: Specifically the Bay Area, I think.
32:28 MH: Yes. San Francisco, Bay Area, Silicon Valley, data scientist. Absolutely.
32:33 TU: You also have all of the people who joined as analysts who were then kind of force-functioned into data engineering capabilities, because the companies are too small. So, even that line, it shouldn’t be blurred, because they’re pretty different jobs, but it’s starting to even get simpler as well. I know several companies that are up and coming, who hired somebody to do data, and all of a sudden, they’re in charge of managing all the ETL pipelines, transforming all the data. And so, we’re just gonna see a continual line of blurring and a bit of bullshit.
33:05 TW: Well, that’s… I saw, at one point, data scientist versus data engineering. I feel like there’s… That’s another one where a lot of people wind up as a data scientist, and Tableau did their Ten Trends in BI thing I went to. It actually sort of presented data engineering as being a subset of data science, because there’s a world of value, kind of back to your original point of just like… Just the data collection and housing of it, which is very operational and it can be done terribly and cause all sorts of problems, or it can be done really elegantly and set up a ton of value, but that is the tactical. Those are people who really know the… They type pseudo, like they’re in terminal and do things, and they say Docker, and they debate Redshift versus cloud platform.
33:56 MH: It used to be SAS and UNIX and DB2. But that’s… Yeah, you’re right. And actually, so that brings up an interesting point, because I actually think as you move to kind of the framework that’s valuing these things more individually, your need for strong data engineering actually becomes more. I don’t know if I have well-formed thoughts around this yet, but I feel like I observed sort of this little bit of a trend, if you will, where software companies and fast-growth companies and tech start-ups, a lot of them will be like, “I don’t need Adobe Analytics. I don’t want Google Analytics. We’ll build our own analytics platform. We’ve got… ” And it’s because they have all of this engineering capability, and over on the enterprise side, it’s kind of like, “Well, we don’t have all the skills to go engineer our own proprietary thing. So, let’s go buy the off-the-shelf products from Adobe, whoever, because it’s built into the way that product works.” And so, I think that’s interesting, because I think if there is a shift or a trend that’s happening over the last few years and continuing on, it is that maybe that data engineering capability is migrating upward into larger and larger companies over time. And that’s kind of interesting to me. And it will be interesting to see, kind of, maybe how that plays out.
35:24 MK: So, Taylor, can I just ask, with your team at Heap, is there a really clear delineation between people that sit on the analytics side of the house and the data warehouse, ETL-type people, or is everyone kind of expected to know a bit of everything?
35:39 TU: Yeah. So, right now we do have a very lean operations analysts team, it’s three or four people. We’re just breaking that 100-ish employees mark as a company. So, right now, it’s kind of everyone is expected to be able to do a little bit of everything, or at least be able to find vendors that can help relieve some of the burden of having a lean team. So not all our time is sucked up with ETL-ing the data, and it’s more focused on transforming the data because we use a tool like Stitch to pump the data from Salesforce so we can do our operational reporting on our pipeline or close rates or pull in attribution data from Heap and join it. We have a lot of managed ETL services so that the focus can be on analysis, but it’s kind of on our team to really, if we need to go bother the engineers to actually build our own infrastructure which is really good, but for that kind of like warehousing, ETL analytics system altogether.
37:57 MK: I think it’s because, I think they wanna be in control of the decision making. You start taking on more and more of that and going back to more and more granular data because you wanna be able to control it. You wanna make the decision about how it’s aggregated, not the vendor that you paid. And I think that’s… That’s kind of why we’re going back in time. [chuckle]
38:18 TW: But I’ll counter with, I think it’d be really hard to build some of the things that are there in some of these tools.
38:25 MK: Yeah.
38:26 TW: So Adobe just launched Attribution IQ and Analysis Work Space, and it would be fairly complex for you to go kind of create your own version of that, I think. I’m not good at building tools so I don’t know. [chuckle] But it just sort of seems like I think there’s still a world where both of those kind of can live.
38:47 TU: Yeah, I feel like a lot of… There’s so many tools out there right now that are best of breed versus full suite that are really emerging. Adobe’s tool might be great, but also I could pay for a tool that integrates with my entire stack that focuses solely on attribution and that is their one thing that they do and they allow me to customize everything, and I can adjust weights and I actually understand the predictive or if they do predictive modeling for their attribution models, I can get a lot more granular. And the people I talk to, if I’m talking to the support team, for example, understand the entire thing because that thing is just doing one specific thing. So I feel like there is a lot of people buying that best of breed kind of like… I’m gonna choose optimizely for my AB testing tool instead of test and target. I’m gonna choose… I don’t know… I just feel like I see that quite frequently, people kind of moving off this kind of all-encompassing suite, especially with digital analytics like [chuckle] all-encompassing suite that does everything, and instead, kind of assembling their own, so they have a little bit more control, but don’t feel the same amount of vendor walk-in.
39:54 TW: And then they need a BI tool to actually pull that stuff together into one spot, so people can actually look at it.
40:02 TU: Exactly. They just wanna spread their analytics budget around a lot, too.
40:04 MK: I was about to say like all that I’m hearing as you’re talking about that is like budget, budget, budget, budget, because if you go to five different companies for five different specific niche things like, you’re gonna be spending a lot of money. I don’t think I’d get it over the line.
40:22 MH: There is some, this is massively prognosticatory, but there is somewhat of a reckoning, I think, coming, right? Like Google just went down this path of becoming the whole thing, right? The whole suite kind of thing, with digital marketing cloud. Is that right, Tim?
40:41 TW: Google marketing platform. Google marketing platform.
40:41 MH: That’s what I thought it was. I knew that, I was just making sure you were.
40:45 TW: We’ll fix it in post.
40:45 MH: Yeah, we’ll fix it in post. [chuckle] But I mean, so, Taylor, there is a huge company that disagrees with you, right? They just literally went for the cloud, right? The big platform play.
40:58 TW: No, those companies, that’s… That’s PHPing, but with…
41:01 MH: I get it.
41:02 TW: I’m gonna let Taylor finish.
41:03 TU: You can put words in my mouth, but…
41:05 TW: Let me put words in your mouth. You’re saying like… The new companies are wanting to go the… Not the companies that are selling, it’s the buy side that is actually doing that.
41:11 MH: Yeah.
41:13 TW: Yeah.
41:13 TU: Yeah, the buy side.
41:14 MK: Not the sell side.
41:14 MH: Right. And that’s where I think this reckoning is coming, right? Is where will this happen, and how will that all weak… I’ll put it this way. Tim, how many MarTech companies are on Scott Brinker’s latest slide?
41:32 TW: Oh, it’s in a blog post, several thousand.
41:32 MH: It’s like 8,000 or something. We can’t keep going like that.
41:35 TU: Sure.
41:36 MH: It’s too much.
41:38 TU: I feel like very, it’s very… The historical pendulum is a bunch of diversification, specialization and then consolidation.
41:47 MH: Yes.
41:47 TU: And right now I still feel like we’re on that specialization kind of like swing. I feel like there’s still a new product every day that’s like, “I am this. I am that.”
42:00 MH: This just in, Heap acquired by Salesforce.
42:02 TW: Yeah. [laughter]
42:05 TU: I’ll take it, but… [chuckle] No, but I don’t know if we’re quite there on the consolidation of all the tools. I mean, maybe Google’s new marketing cloud platform is an indicator that the market is maybe starting to be ready for a one-stop shop. But I feel like there’s been a lot of push for a bunch of things that are the best thing at what they do.
42:30 MK: I also think client-side, companies, less and less, and I don’t know if it’s the same in the States. But here, people don’t wanna be locked into one company, or one vendor. It actually starts to become like a security and IP issue where you’re like, “I don’t want everything to sit with this one company, because what happens if something does happen to them, or they change their policy?” I feel like people are looking for that diversity or that choice, which is why we’re seeing a bit of a shift towards open source.
43:00 MH: But I can show you, Moe, a counter example, a number of counter examples of people who desire to have everything in one place. I honestly think it’s starting to break down by type of person and personality.
43:10 MK: That could be.
43:10 MH: Surprise, surprise, Michael thinks it comes down to people. [laughter] Because some people want the security of it all kind of being in one place, and me having one vendor who’s gonna do all of this stuff for me. Other people, kind of on that more entrepreneurial side or more risk taking, are looking out there and being like, “I wanna get out onto the edge further. Where that’s happening, is not in the big consolidated places, because they don’t move as quickly as what’s happening out on the cutting edge of technology, and I wanna take advantage of that.” And then as those other companies progress, the current trend is, right? They… Somebody goes out there, distinguishes themselves and they become the acquisition target of one of those big companies.
44:00 TW: Salesforce, Oracle, or Adobe…
44:03 MH: Adobe, Google.
44:04 TW: Or whoever.
44:05 MH: Yeah. There are lots, and lots, and lots of acquisitions. I have no experience with that.
44:09 TW: But [chuckle] the consolidated both Adobe and Google, kind of their platform, like their play is really fundamentally about a single view of the customer. Kind of what we can do is, we can do this joining at a very granular level, that would be very expensive for you to try to build. Although now, you’ve got the rise of customer data platforms that are saying they’re there if you’re building your best of breed from all these different things, use them to kinda key it together. Oh, man, I can’t retire soon enough. [laughter] So…
44:45 MH: It’s not gonna be our world anymore, Tim. [laughter] The next generation, it’s on you guys.
44:53 MK: You guys do okay, still… You do okay.
44:55 MH: It’s on you guys, you gotta take it from here. Okay, one thing we can still do, Tim, while we’re still alive, is start to wrap up. ‘Cause this has been an amazing conversation, but we do have to start. Well, one thing we love to do on the show is go around the horn and do a last call. Something we’re interested in, something we think might be of interest to our listeners. Taylor, you’re our guest. Do you have a last call you wanna share?
45:22 TU: Putting me on the spot. Actually, I just got access to one of Stitch’s betas, called Scripts. And so, like I said, I do some of the transformation of the data on our own data internally. And basically this allows us, or should allow us to kill some of our air flow processes to get the data in the right format to create those R created tables. So, super excited to see how it ends up working out and what’s going on with that.
45:50 MH: Alright, some of our listeners understood what you said. So I’ll just… [laughter] No, I heard air flow and I was like, “I’ve heard that word before, air flow, nice.”
46:01 TW: I’ve heard Stitch. What is Stitch?
46:04 TU: Stitch is… It’s an ETL…
46:04 MH: ETL. ETL, okay.
46:05 TW: Okay.
46:06 TU: Tool, yeah. So basically, instead of me having to control all of my ETL processes, it’s just adding a little bit more that they’ll help with. So I can have some things in one system instead of all over the place.
46:17 MH: Does it have a drag-and-drop interface?
46:20 TU: It has toggles.
46:22 MH: Toggles. Okay, well, that’s… Well, maybe they’ll evolve to drag-and-drop, at some point. [laughter] Interesting. Alright, Tim Wilson.
46:29 TW: Well, so, continuing the theme of things that maybe Michael hasn’t heard of, I hadn’t heard of some of these, this will be the same. This was a tweet popped up from Mark Edmonton, and what was kinda cool, so in Brisbane. Brisbane? Is that how you say it?
46:45 MK: Nailed it.
46:45 TW: So user, user 2018, the big R conference, was not too long ago. And all the videos are posted on…
46:51 MK: Is it actually user or use-R?
46:55 TW: Use-R, maybe it’s use-R. The R is capitalized.
46:58 MK: I don’t know. I’ve been seeing all the tweets and stuff about it.
47:01 TW: I don’t know. Yeah, I don’t know. But the big R conference…
47:04 MH: Tim, I have heard of R before.
47:05 TW: Yeah, the big R conference, and they post all their videos, and they’re really well done videos. They’ve got a nice… The speaker over in one little corner, but the main slide’s elsewhere. But he tweeted this video of Katie Sasso’s session, and that’s kind of awesome, ’cause she’s speaking at an upcoming Columbus Web Analytics Wednesday, and she is the co-founder of the Women in Analytics conference in Columbus, and she’s awesome. But her session was about using Electron and NOGIS with Shiny to make stand-alone interactive Shiny apps. I did not realize that like Slack and WordPress were built with Electron, so I didn’t really know what Electron was, so somebody had to point that out to me. NOGIS I’ve heard of, but vaguely, only vaguely aware of it. But basically, if you’re an R user, and you do stuff with Shiny, you know the challenge is having a Shiny server. So what they basically had done in her organization, the Columbus Collaboratory, was used Electron and NOGIS to be able to literally make stand-alone, executable, Shiny applications that they can just send to a client to run on a Windows or Mac, and it’s a fully, wholly contained Shiny application. And, it was cool, ’cause two people I think are pretty damn sharp and cool, Katie and Mark, they kinda came together in my Twitter feed, so that made it kind of cool too.
48:29 MK: Cool.
48:30 MH: What version of the Java Runtime engine do I have to have installed for that, Tim?
48:33 TW: No idea.
48:36 MH: Alright, Moe, what’s your last call?
48:38 MK: I’ve got some old ones today. So, firstly, there’s this visualization of Manhattan’s population going around. Http/manpopx.us, and it’s just one of those weird things that you kinda need to check out and be like, “Whoa, whoa, this is like cool shit that people in our industry do.” The other thing that I’ve recently discovered that I’ve been listening to, which is a little less analytics-y, I’ve started listening to the Tim Ferriss Show, which is a podcast, and I don’t love him. Pause, but what I have liked in some of his really late episodes, this is like podcasting, I don’t know, verging on podcasting laziness, but they read out a couple of chapters from different books. So, the one that I listened to the other day was Essentialism: The Disciplined Pursuit of Less and it talks about how to say no, gracefully, maybe Tim, you should listen to that one.
49:35 MK: Sorry.
49:38 TW: Oh, my God.
49:39 MK: It’s been great, because I actually have kind of listened to a couple of books, but they’re just picking out the chapters that are actually really useful, so you don’t have to read the whole book. So, that’s been kind of cool.
49:52 MH: That’s awesome.
49:53 TW: Okay, as long as it’s not a 24-hour work-week plug, we’re good. Michael, what do you have?
49:55 MH: Alright, I’ve got a two-fer. Alright, so first off, in full transparency to our listeners, we’ve been messing around with the Heap platform on the analyticshour.io website. And so that actually is what got us in touch with the folks at Heap and connected us to our guests, so I wanted to let people know. But, what’s cool is, I think they have a free version of their product and so, if you’re curious about Heap, that’s one idea. We don’t really support or promote products on the show, but when somebody’s got a free version, we like to tell people, so they can go kick the tires, that’s a great way for people to learn, and stuff. Alright, my next one is, there’s a Twitter account owned by a guy named Tyler Riggs, and you specifically need to go search through his timeline to find a template he created of Benjamin Gaines, that you can put any words you want on, and we need to all work together to create embarrassing Ben Gaines memes on Twitter.
50:52 MH: Okay, so that’s the other one…
50:57 TW: That’s good.
50:57 MH: Search for Tyler Riggs, and let’s all put together some really cool Ben Gaines quotes. Okay, listen, this has been a really interesting conversation and it’s actually kind of like, we’ve really covered a lot of ground and speculated wildly, and I think that’s a great place for our listeners to chime in. And, by speculated wildly, I feel like I speculated wildly, Taylor, you didn’t, you did a great job not speculating wildly.
51:23 MH: But, we would love to hear from you. The best ways to do that is through the Measure Slack, or on our Facebook page, or on our website, or on Twitter. So, please, love to hear from you, and we hear what you have to say about this show. Alright, Taylor, thank you so much for being our guest, we really appreciated having you on, great to have you on the show.
51:47 TU: Thanks for having me, I really enjoyed it.
51:49 MH: Awesome, and from my two co-hosts, Tim Wilson and Moe Kiss, and all of you analysts out there, no matter what your title is, do this for us, keep analyzing.
52:06 Announcer: Thanks for listening, and don’t forget to join the conversation on Facebook, Twitter, or Measure Slack group. We welcome your comments and questions, visit us on the web, at analyticshour.io, Facebook.com/analyticshour, or at analyticshour on Twitter.
52:26 MH: So, smart guys want to fit in, so they made up a term called analytics, analytics don’t work.
52:32 TW: Oh, look, we’re recording now, so I have to watch what I say. We should all gang up on Tim, ’cause I think he’s feeling just a little too sure of himself these days.
52:49 MK: I didn’t know that was an Australian, I mean an American… I thought that just was in Australia where young people get criticized for eating avocado on toast.
52:57 TU: Oh, no, no, no, no. It is.
52:57 MH: Where do you think it started? It all started in San Fran…
53:01 MK: No, it started in Australia, I’m gonna find the original story.
53:05 TU: I’m pretty sure California has all of the avocados.
53:08 MH: I’m with Taylor on this one, Moe.
53:10 TW: Well, that took a turn off topic.
53:12 MH: Well, it is important to establish where avocado toast came from.
53:18 MK: It came from Australia, I swear to God, this is not one of those things…
53:20 MH: Moe…
53:20 MK: I’m gonna find the original article.
53:23 MH: Find the article, and make it your last call, then.
53:27 TW: So, I might have ranted about that in my own brain, a few times. I google Google Marketing Cloud. Google Marketing Cloud.
53:36 MH: Google Marketing Cloud, platform?
53:38 TW: No, it’s the Adobe Marketing Cloud.
53:40 MH: Google Cloud…
53:40 TW: No.
53:41 TW: What is their new one?
53:43 Announcer: It just literally, it’s Google Marketing Platform, I think.
53:47 MH: I think that’s right.
53:47 TW: Ahh, geez, I should know that.
53:49 MK: I said, “shit” a lot.
53:51 TW: I said “frick it”, once, I’m like, “Why am I saying frick?”
53:57 MH: Rock flag and what the fuck’s BI?