It’s a challenge as old as the smartphone (or…technically… a little bit older): we want to track and know people, but, when they’re visiting us from their phone, their tablet, their work laptop, and their home desktop, we often really just know cookies and device IDs! Countless vendors tout their enabling technology…but then admit that, yes, you do have to get your customers to authenticate on all devices to provide a common key, which isn’t always easy (or even remotely reasonable). Unless you’re up for diving into the deep and murky waters of probabilistic linking. On this episode, Tablet Tim — the only one of the co-hosts who is an avid tablet user — argues that the whole topic should be pretty “meh” for many companies, while Mobile Moe smacks him down and shares her experiences (the challenges…and the wins that made them worth it!) with linking users across mobile apps, a mobile site, and a desktop site.
- Adobe Device Co-op (Overview)
- MIT Sloan Sports Analytics Conference
- Ben Gaines
- Artificial Intelligence for Marketing: Practical Applications (book by Jim Sterne)
- Google Quick Draw
00:00 Michael Helbling: Hi, everyone! Welcome to the Digital Analytics Power Hour. This is Episode 75.
00:08 Tim Wilson: 75!
00:09 MH: Yeah, is that a milestone? I was not aware, that we’d hit a milestone.
00:14 TW: Could be.
00:15 MH: We’ve triggered our milestone-aware alert beacon, [chuckle] it’s an IOT device we’ve had specially installed. Anyways, as analytics people, we often talk about having a 360-degree view of the customer, and what we sometimes don’t mention is that we mostly want it to follow people around and laugh at their online antics by name. Okay, not really. That would just be rude, but for anyone sporting more than one site in an app and some other stuff, we have a problem. And every internet user compounds the problem because they access all of our digital experiences with multiple devices. And that’s what we wanna talk about on this episode, cross-device tracking. On mobile, we have Moe Kiss, #WelcomeMoe.
01:04 Moe Kiss: Hi, team, how’s it going?
01:06 MH: Good. [laughter] Now, on tablet, we have Tim Wilson, pause your movie, Tim.
01:13 MH: And on PC, I dunno, Microsoft Michael? I don’t know. PC. I’m just hitting save on this Excel file. Okay, so let’s all set our analyticshour.io cookie to value 67Y324LEF9234LHE0907QEWRQRW and let’s get this show started.
01:36 MH: So Moe, as we were prepping for the show, you brought up a very interesting point which I think is probably best to cover right off the bat, which is cross-device, which is kind of the different devices people might use, but also cross-experience which is all the different experiences you might present to a customer. So is that what we mean, how do we all define cross-device visitor and how do we think about it when we’re trying to connect the dots?
02:05 MK: Yeah, I guess we all have different views of this and for me, in my role, I’ve always looked at it being, you have your tablet, you have… Which primarily I would group… I guess it’s more about how you’re browsing, so we have our two apps, we have an IOS app, an Android app and then I would class mobile site as our next category and then desktop. But I know Michael and Tim disagreed with me because they see it really as tablet, mobile, desktop, but I don’t, because I see a native app on your phone as being a very different experience to your mobile website.
02:44 TW: Woah, woah, yeah, nobody said… Don’t lump me in with Michael on that, I did not weigh in, I think this entire exercise is a complete waste.
02:52 MH: There you go. Tim stands alone as being completely against this whole thing.
02:58 TW: What is this mobile thing you’re speaking of? Why don’t people just have their big CRT monitors and sit and share the family computer?
03:04 MH: Yeah, isn’t everyone’s house like mine, where right in a corner of the dining room, the family computer is sitting? No. I do agree with you, Moe, around the idea of tracking people in between the different experiences, so if you have a website and you have a mobile app or you have… In the B2B world, a lot of times you have a website and you have sometimes an ERP punch-through system or something for orders. Connecting those things I think is very important but it’s not, to me, cross-device tracking, that’s all.
03:44 MK: So what would you call it?
03:46 MH: That’s a tomato and this is a tomato. [laughter] No. It is interesting ’cause I’m not sure I have… That I’d call… Would be calling more like data integration or customer alignment or customer integration.
04:03 TW: The fact is, you wanna know every possible trackable interaction that they’re having…
04:08 MH: Yeah, I wanna know who does what where…
04:12 TW: With you. To me it almost feels…
04:13 MH: I guess I separate it based on… I have different touch points that I create as the end marketer, whether that be the retail experience, the phone system, the website, the mobile app. And I do my best to connect all of those as best I can. And then the consumer is creating their own differences by leveraging different devices to interact with those various things.
04:37 MK: Yeah, I guess… Well, I don’t know, I have a really different stance on this, and maybe it’s a little bit of the product hat, maybe it’s the environment that I’m in, but there is something about, particularly from a product team perspective, there’s this like, “Our app users are like this, our desktop website users behave like this, our mobile site users behave like this”.
05:00 MH: Yeah, I don’t buy into that at all.
05:04 MK: That’s really like a… It’s just another form of siloing. But the thing is a lot of our tools actually set it up that way because those users are split, like you have your Google Analytics account for your app users, your Google Analytics account for your website users, and it’s split that way. And the reason that I care about this issue, and I get really passionate talking about it and I’m really interested in it, is because I think it’s part of due diligence of starting out to be like, “Okay that’s not how users engage with platforms, that’s not how… ” Particularly Tim was like, “Oh, this isn’t relevant to everyone.” And I’m like, “Yes, but it’s becoming more relevant”. And it’s gonna become more relevant as everyone has the mobile in their pocket but then they are on their desktop at work and it’s about analysts starting from that point of going, “This is actually really important for me to crack, because it’s about making sure that we don’t silo those different experiences.
06:02 TW: But I think… The debate between whether it’s becoming more relevant or… That’s a industry consumer behavior shift. I think asking the question of is it relevant to me, to my organization now or will it be relevant in the next six months and that’s where… You are in a pure play online retail space that has multiple channels, and absolutely if there are people who are defining… If they are speaking about the consumer as the mobile app user versus the mobile site user, there’s a poster child case for getting that linkage there. But I am gonna continue to contend that if you’re selling toilet paper or toothpaste, and it’s an FMCG or CPG scenario and you have a website and maybe there’s the mobile, you’ve gotta stop and ask yourselves, “Are people like that into my brand that they are actually visiting on multiple devices often enough for it to matter whatsoever?” I think the media sites, absolutely. I am on bbc.com and cnn.com and slate.com on multiple devices regularly, totally makes sense. Retailers, loyal customers totally make sense, but I don’t think we can just say, “Well yeah, but that’s… Everything’s kinda heading that way so we all need to be sweating it.” That’s what I take issue with.
07:30 MK: I don’t think it’s about sweating it. I think the example you just gave about toothpaste is the exact case that I would say, this is why you should be starting off with understanding cross-device paths or journeys or whatever you wanna call it, users, because…
07:47 TW: No, ’cause I’ve never…
07:50 MK: This isn’t just my opinion, I’ve validated this with analysis, but people do their research… Like for me, I do my research on my phone. So, I open up my email, I see the toothpaste is on special. I then make a note and go, “Oh, okay, I need to buy toothpaste,” I add it to my gro… Whatever, but then I end up going and placing my order on my lunch break on my desktop at work. So, if you’re not taking into…
08:14 TW: Come on, one. But you’re not researching fucking toothpaste.
08:21 TW: I would claim that you are not having enough digital interactions for it to matter. You may pull it up on your phone and say this one time in my life that I wanna choose between Crest and Colgate and I’m gonna go read something, and then I’m gonna put my phone in and I’m gonna go buy it at the corner store or I’m gonna buy it on Amazon or somewhere else. It’s not like you’re buying toothpaste, you’re gonna be on the phone and then go to their website, and then a week later come back to your phone. I think there’s an entire class of businesses that they have less of the ongoing repeat interaction and relationship as much as they might like and that’s why I’ll throw… The CPG and FMCG is one. I think some B2B stuff falls in there too. So I think it’s very situational that there are lots of cases where there aren’t enough interactions with digital that that’s not worth chasing.
09:14 MK: Okay. I’m gonna completely flip this and I’m gonna mess up your analogy, because what I’m gonna do is talk about dog food. Because the reason is, I actually do this for dog food, and I know that we shouldn’t be basing all of our opinions. I agree that for some businesses it’s less important than it might be for others. For my company nailing this is hugely important, and for others, it might not be as important. But dog food, right, I ran out of food, well I’m going to tomorrow. Today, when I’m waiting for my lunch, will check the four people I buy dog food from on my mobile phone while I’m waiting for my salad to get made and I will see who’s got what on sale at what price. Then some stage probably late today, tomorrow when I remember, I will then go back to that site and I will buy it. Now, I’m spending $120 on dog food and I do that probably every month. If you as a business are not counting…
10:06 TW: Wait, but this is not how 98% of the dog food on the planet gets purchased, which is as a CPG, you’re saying that you’re actually shopping and buying… You’re buying from the pet food manufacturer or are you buying…
10:21 MK: No, from different online stores, from different, completely different stores.
10:25 TW: Yes. You’re not actually talking about pet food, this is not applicable to the pet food purchaser, this is to the retailer who sells a whole bunch of stuff, right? So it’s actually not relevant to the pet food, because they don’t have any of that data. You’re going to the retailer who’s actually selling the pet food.
10:45 MK: Yeah. Okay, okay. I agree with you on that. [chuckle] If I’m the pet food company…
10:49 MH: Yeah, you’re not gonna get that retail thing though.
10:51 MK: It’s probably… Yeah, but then you’re selling to other people. Okay, yeah, yeah.
10:53 TW: I mean, you aren’t hitting on… It’s possible that somebody on this podcast has had a lot of experience with pet food and digital, but…
11:01 MK: I know, that’s why I brought it up.
11:05 TW: Thanks for bringing that one up so I can swing. But yeah, you have three retailers and are all four of those… So, even if it is the retailers, those are all four that you’ll be buying… You’d be buying online, you’re not looking for one where you will shop on mobile and then go in-store and buy or are you ordering it?
11:24 MK: I do sometimes, I do sometimes. It depends on where I am and what I’m doing. But I don’t know, I’m probably not the typical shopper, because I hate going into stores ever, which is why I work at an online retailer. [laughter] I think the point that I’m trying to make is that I feel like you miss a fundamental piece of the puzzle about your customers when things are siloed. Like this is just another form of saying like, “I’ve got this snippet and yes for some companies it’s less relevant,” but for the company that I’m working at and for a lot of companies that operate online in some capacity… And to be honest, this was actually, probably a misunderstanding the we had about our customers, where we undervalued our app because we didn’t realize that what our customers were doing were, they were using the app to find what they wanted to buy. That was where they were doing their browsing, adding to wishlist, all that sort of stuff but they were converting on desktop. Now if you have them siloed, you’re gonna be like, “Oh, the app is really great at getting people to add to wishlist but they’re shitty at converting”. Whereas the desktop, oh, it’s like killing it because people convert on desktop so we should put all about resources there.
12:35 MK: The actual takeaway from that is that you need to support your users to do what they want on the platform they want. So you need to make browsing and finding what they want easier on the app and then focus on making the transaction really seamless on desktop because that’s how people are interacting with us. And I’m not sure people get that yet. I think people just duplicate the same thing on every device. And every device isn’t good at everything and you need to figure it out for your own customers.
13:02 MH: Well you can’t be a listener of this show and believe that if you’re not exporting your analytics data into a data warehouse for that kind of integration, you’re actually doing analytics, right?
13:16 MK: Yeah.
13:16 MH: As the great Tim Wilson once said… No I’m just kidding. No but I think [chuckle] that’s… I’m gonna go quite as far as to call that ‘table stakes’ but, Moe, you’re absolutely right and I think that’s prettily well understood that that kind of stuff is happening all the time.
13:36 MK: But I don’t think the tools are not actually set up. It’s getting better.
13:42 MH: So you have to go do all the linking manually still.
13:43 MK: The reason I think… Yeah, yeah.
13:46 MH: You have to go and pull it in, find the qualifying point of, oh, they click through an email on their phone. So now I have this session and cookie and I can go take all of those sessions from there, and then connect it with all of the email authenticated or purchase authenticated sessions over here. And all of that logic can tie in.
14:07 TW: But that’s the other leap. Even if you say, “I’ve got… My consumers are interacting across devices and I have evidence and maybe it’s just because I have thought through the nature of my consumer, I’ve mapped out their consumer journey, ” whatever it is, there’s that other assumption that just happens where it’s like, “Yes! The key that you need to link those together. Just make sure they identify themselves.” Google trumpeting user ID, and like that’s freaking awesome. Except when those interactions back to researching toothpaste, why would I ever self identify myself when researching toothpaste and going over to… So I also have to have the conditions where I have permission or I have a reasonable reason to get them to self identify. I’m selling insurance, I’m a financial institution, I’m an online retailer where I’ve got a decent loyalty program. I can personalize it great.
15:03 TW: But I would also… That’d be another one where I’d claim there are a ton of companies out there that don’t have that. That’s an opportunity for them to say, “We’re going to create a reason where there’s value for you telling us who you are”. But that to me is that’s a big leap, ’cause now all of a sudden you’re designing content and campaigns and programs and the way that you’re interacting with your user is to say, “You are gonna want me to know that you’re the same person”. Just like when you’re buying your apparel and you wanna make sure you’re getting your personalized recommendations or you’re getting your discounts, I’m selling something else or I have another relationship or I am a service organization. And I have to give you some reason to say, “It’s worth me telling you that I’m Tim Wilson and here’s my login”. And they’re definitely…
15:56 MK: Yeah. Yeah. I agree with that.
15:58 TW: And actually that’s a good example like airlines. Airlines absolutely. I hope that Delta is linking ’cause I am on their app often on multiple… On a mobile and a tablet and I am on their website on multiple devices. And every time I start by self-identifying. So it should be a slam dunk and if it’s not, if they’re not not linking it, they’re missing…
16:24 MK: This stuff isn’t easy though. This stuff is actually really hard to do at the moment. And it’s too easy for businesses and analysts to be like, “Well it’s too hard”. And I’m going through this at the moment where I guess we need to highlight the value of doing this kind of work because it is gonna be an uphill battle. But ultimately you look at a topic like personalization, how on earth do you do personalization well if you have got different silos of data about your customers. So you need to have… This is a starting point. But getting the business to buy into… This is actually really important to understand our customers have different interactions on different devices. They use different features on different devices for different reasons and you can’t nail some of that really big ticket stuff that you wanna do unless you start from this. And it’s not the fun work, finding identifiers to link people…
17:20 MH: Yeah, but if you wanna have some level of sophistication around how you’re interacting with customers and how they interact with you and provide consistency, this is the work. And to your earlier point, Moe, when you have figured out that the browsing and the shopping is happening on the mobile device and the purchasing is happening on the desktop, that is a profound insight that impacts a lot. Because what would you have done not knowing that, you would spend countless hours and effort and money on trying to improve the conversion rate of the mobile site and countless hours and effort trying to entice more people to demonstrate quick purchase behavior on the desktop site. Like it would have gone totally crosswise your customer. And so that’s the risk, that’s why you have to do it. And there’s an expectation, the way that I interact with brands, it has to be the same thing. So when I book my hotel from Marriott on my desktop and when I open up the app, my hotel reservation should be there, and all those things have to be connected. But again, that’s another example like the Delta one.
18:37 TW: But I think, there’s a little bit of a leap saying that they’re not connected, you’re not connecting or you’ve gone through all the hard work to connect them which is a significant and can be a significant investment and can be hard but there’s this other thing you can do, is to say do we need to chase this or not? Why don’t I survey? Why don’t I do a consumer survey and ask them, are you? Which all these ways do you interact with me, of my customers or not… That can be a low cost… I think there’s one that’s just common sense and I think the iconic, you’re like absolutely we have these, let me do it manually, just to prove through whatever I can, that this is happening enough, that it makes sense to figure this out. And now, now you’re down the path of saying how do I make this where this doesn’t suck down all of my bandwidth to try to make this linkage. There are ways you can come at it simply and say it doesn’t make sense. They’re not engaging with us enough and I can’t, in cases where I’m like, I don’t have an identifier. I can chase… I can dream about all this I want but if it doesn’t exist, it is the physics of the fucking internet. Like, no amount of tying it together.
19:50 MK: I think your point on identifiers is completely legitimate and it is a debate with UX because UX are always like, you don’t want people to have to log in because it just, like, cuts out the barriers. Let people use guest check outs, all that sort of stuff and as an analytics person, I’m like I absolutely want people to log in because for me, it like, yeah, you have that identifier. It makes life easier, but I agree there has to be a value to the customer about having that identifier.
20:18 MH: So what we’ve been talking about in the terminology so far is deterministic linking, right? Linking people based on some kind of primary key that determines who they are in both places. There is now a set of emerging tools to help you link people probabilistically, right? And so, that’s where I think there’s this next layer that should make things easier and there’s things happening on both sides. The Adobe’s device Co-op and I know Google is doing something as well to basically bring you into view of the person, not identified person but across all of the devices they might be using so that you get that single understanding. And I don’t know how that’s gonna work ’cause none of that has really come to bear and so…
21:14 TW: I wouldn’t say probabilistic matching being easier, I would have said that if what Moe is…
21:16 MH: No, it’s way more dirty…
21:19 TW: If the deterministic, you say is tough, probabilistic, all of a sudden, now you’re making a ton of decisions of false positives versus false negatives and what am I gonna do with it and when is this going to bite me in the ass if I’ve got a false positive?
21:32 MH: Well, obviously there’s things you can and you can’t do with the probabilistic match versus a deterministic one, but if you have a competent probabilistic or a combined match, based on say Adobe’s device co-op, you know that whoever’s browsing right now anonymously has also browsed via their mobile device because that’s what the device co-op is gonna do, you at least know those two things about that person at that point in time. And then once they authenticate, then you can create a deterministic match for that visitor.
22:13 TW: So, do we know what kind of traction that whole… That was two summits ago and when they…
22:17 MH: No, it’s been growing. So there’s actually a website, I wanna say it’s a cross-device-privacy.adobe.com that actually shows you the companies that are participating in it and also allows you to see what devices of yours have actually been connected via the device co-op and I wanna say they’ve…
22:40 TW: But even the participating there’s… I’m gonna sign up because I have the promise that I’m gonna get great data out of it, it doesn’t necessarily mean they’ve actually figured out how to use it. No, that’s actually good to know that it is growing.
22:52 MH: I wanna say from March to July, very anecdotally, it’s probably more than doubled, of this year. So, yeah, there’s quite a few companies that are getting involved. And so… And again, this is probably more a US thing, and this whole episode goes up in smoke [chuckle] with GDPR or the new ePrivacy Draft Regulation, so [laughter] yeah, we’ll just delete this episode. This whole episode is in violation of the EU, basically… Well, I don’t know, but probably. But that’s…
23:28 TW: We’ll be sending a targeted email to everyone of our listeners. They think we don’t know who they are, letting them know.
23:35 MH: Yeah. I will say this about doing the deterministic linking yourself, by doing that work of finding those cookie values or that authenticated visit, that work has gotten much simpler over the last 10 years, just because Maya was working on this 10 years ago and we do this work now. And the very fact of the cloud-based capabilities of these data combinations in BigQuery and being basically as cheap as free, that makes the barrier to doing this so much lower. It used to be you’d have to plan out hundreds of thousands of dollars in net appliances to house and process all of this data in your own data centers. And now you basically can chuck it into the AWS or into BigQuery or whatever and then just apply the logic to it, and… Now you still have to do a lot of work, you gotta create all your business logic, you have to create sessionization unless you have the BigQuery GA connector. Oh, and you don’t have to pay for data feeds from Adobe anymore. So it used to be a long time ago you had to actually pay extra for data feeds. And now that’s no longer…
24:54 TW: And configure it yourself too.
24:55 MH: And so now that’s just part of the admin console.
24:58 TW: I was having the same thought when you… That it’s hard, but I was like, “Oh, this is why we strategically plan the Google cloud platform episode right before this one. Because I agree that’s like the other case. So Moe, I was thinking the same thing, but you actually have been doing this and you’re saying it’s hard. So there are challenges I assume with actually getting the data linked up, but is it… Do you feel like it’s relatively easy to actually at least get the data that you’re trying to link all into an environment where you can join it together? I see you asking questions about BigQuery and joining stuff. So where’s the challenge? Is it getting the data? Is it finding the keys? Is it actually figuring out where you can get the data that you’re trying to join into one place?
25:50 MK: I actually… I’ve actually… What else is Web Analytics Wednesday for other than analytics therapy session? And I was talking about this a few weeks ago at one of our Web Analytics Wednesdays. Because for me the challenge has been, when you look at cross-device behavior and stitching… Ultimately right, your fist step is you have to stitch users. And the stitching user bit is getting easier, the bit that I am struggling with is that the dataset is then so incredibly big you need to find a way to aggregate it up and then visualize it. And the problem is, is that when you talk to people in the business about having this, a stitched version, and despite all of the privacy stuff, you don’t need to know someone’s name, their date of birth, you need to know that this is visitor x and they’ve visited us three times and they came via email and they looked at these 10 products.
26:46 MK: The problem is though, that some of the challenges that I’ve faced and if anyone has advice suggestions, please tweet, is getting it from that raw data point to an aggregated view that you can then use really well for reporting. Because, I mean, the dataset that I spent the last week working with is, it’s so big I can’t run anything on my local machine, you could never open it. I’m doing it all in BigQuery.
27:14 TW: But in a way, that’s interesting ’cause what you’re pointing out is that if we look at every one of the analytics platforms that are collecting data, that there is the data collection and then there’s tools they give you to interact with it. Well, once you’re pulling data from those multiple systems and stitching it together, you’ve completely negated the ability to use…
27:34 MH: Yeah, you’ve walked away from all those tools.
27:37 TW: Yeah, so now you’re building your own analysis interface.
27:42 MH: Right.
27:43 TW: And that is…
27:44 MK: Yeah.
27:45 TW: That’s, okay that’s interesting.
27:46 MH: Well, and that’s where things like Tableau, Google Data Studio, Domo, that’s where they fit into this paradigm, right?
27:55 MK: Yeah, provided that you can aggregate the data to a level that can be ingested into one of those tools. Because if it’s in its raw format you can’t, it’s too big.
28:05 MH: But typically it’s this, you’re creating segments that you then apply to your data, that you then report on and those segments or cohorts or however you wanna define them, are indicative of some set of values or actions that mean something important to the business. And that’s why you’re reporting on them.
28:25 TW: But that’s the gap between you have… You’ve stitched this together and you have 35 million rows of data before you bring it into Tableau, whether it’s a segment or a filter or maybe it’s just sampling. You say, “You know what, let me just sample 10 per… Let me sample and get a smaller group, but there is that step.
28:43 MH: Yeah.
28:44 TW: That is that aggregation or sub setting or filtering or segmenting step that is a query, right? And you’ve got to have… That’s where Moe was saying you’ve gotta have the horse power to do that.
28:55 MH: And I’m not good with databases, isn’t that just the creation of a view?
29:00 MK: Yeah.
29:00 MH: At a certain level. You’re just creating a view at a…
29:02 MK: Well, I mean basically that’s how I’ve decided to tackle it, is that I’m… I built it in R first and the dataset was too big to use, so now I’m building it in BigQuery and creating it into a new table and then I’m gonna query that table for what I want.
29:16 TW: But that’s where historically when we were building data warehouses and feeding stuff in, that was the questions around what are the views, how are we indexing. So it’s interesting because that was an IT development role, where they were… There was ability to have performance and speed, and how do we design. Now the infrastructure has gotten way, way, way more powerful but not so powerful to the point that we say screw it, we don’t care about that, just keep the millions and millions and millions of rows and I’ll hit it directly. So it’s Moe is kind of stumbled across the part like whose role is that?
29:53 MH: Right, right.
29:53 TW: Because Moe did all this clever stitching as an analyst and said, “I sort of know how this should all work but now shit… Now I’ve gotta be the person who actually builds usable, queryable, optimized views that now I can actually go and do the actual work that I wanna do.”
30:09 MH: The actual reporting, the actual dissemination of the data and that’s where…
30:13 MK: Exactly and ultimately you can do all the stitching you want, but if you can’t get it into a format that it can be digested by people that are relevant to their KPIs like there’s no point, yeah. So that’s why I say that it’s really hard, it’s about… And I’m learning as I do all of this stuff which has been awesome but it also means it’s taking me a shit ton of time which is frustrating.
30:36 MH: Moe that’s what you… What you need is a data scientist.
30:39 MK: We just hired three but I heard that analysts and data scientist are like kind of doing a bit of overlap with the skill sets.
30:49 MH: Yes.
30:51 MK: May be I don’t need one.
30:52 TW: Then I will say that I wound up over the last few months have stumbled across one of those sort of scenarios… Financial institution like banking where there was a application process and very early in the process you get assigned this like application ID and this is all happening going into the same web analytics platform but we also are capturing the Google Analytics cookie in a custom dimension. And as I started digging into it, what I started to find was the cross-device behaviour was happening because I’d see the same app ID show up with different Google Analytics cookies which may be they cleared them or may be they started the application at work and then went home but then I’d see the reverse, where I’m like “Well this is the same Google Analytics cookie ID of multiple applications. They’ve started it multiple times.” So I wound up spending a lot of time saying “What’s the logic where I can kinda narrow this down to basically a user ID?” And that took a lot of work, once I got that dataset… But I’m talking half a million rows of data so I can work with that. Now all these cool questions are coming up and it’s super easy to get to that but it took a lot more thought and…
32:07 MH: Right.
32:07 TW: It did take some scripting in R and putting in the logic and handling all these corner cases, and I do look at that as saying “Wow, there’re ton of analysts which I can’t see how they would do that. And oh, by the way instead of half a million rows, what if it was 50 million rows?”, now that little plan goes out the window. But that was one where I wasn’t consciously coming at it from a cross-device perspective, I was just seeing wacky stuff in the data. And I had two keys and could kind of time stitch those together and said, “Wow, this is all the same person.” So I guess that’s actually back to your original point where… But sorta I wasn’t consciously looking for it but I wound up being able to go back to the stakeholder and say, “You realize, I can actually tell you how many people seem to be starting this process multiple times before they complete it? And I can tell you how many people seem to be going through the process… Either at least multiple cookies which could be multiple browsers, but there’s a good chance looking at the time stamp this is day time versus evening.” But again that was a banking scenario.
33:12 MK: Tim that almost sounds like you’re agreeing with me just quietly.
33:17 TW: It’s the corner case where there’s an ID created so yes then…
33:21 MH: Moe, we all agree…
33:23 TW: I was pretty excited when I found it.
33:24 MH: We all agree, we just don’t call it ‘cross-device tracking’.
33:31 MK: Yeah, potentially. But okay, I was just gonna say… This is a point that Michael made earlier which I agree with, it’s… This work is really… Like I said, I found this work really hard, really challenging but ultimately very rewarding and it’s rewarding in terms of…
33:47 S?: Yeah.
33:47 MK: The impact those insights can have on the business and you’ve just given a really great example there, Tim. And I have a similar one where, like I said, we found that our app was really used for a lot of browsing and all that sort of stuff. We also found that our mobile site was typically where people had their first ever visit with us. Now when you take that away as a product team and you go “Oh.” So basically, to summarize we found the mobile site is where people have their first ever visit, the app is where people are really, really deeply engaged with us, and the desktop is where people do their admin tasks; it’s where they place their orders, it’s where they update their subscriber preferences all that sort of stuff. And as a product team, to be able to go away and be like, “Okay this is actually the purpose of our mobile site, is to be really clear for new visitors” or “The purpose of our desktop is to make those admin tasks easier.”
34:43 S?: Yeah.
34:44 MK: The value that brings to a business is huge. And the message that I’m trying to give to our listeners particularly analysts in a similar position to me, is that this stuff is really hard but it’s ultimately really worth it. And yeah, I just think… That’s the take away…
35:01 MH: Alright, great show everybody, this is perfect.
35:08 MH: I’m gonna stop, I feel like that was an awesome bow that you just put on it, Moe, that’s exactly right.
35:13 TW: Yeah.
35:15 MH: Very difficult, very worthwhile and some really amazing examples from your experience of how you guys are actually doing that. What I’ve also found is that as you do this kind of work, it levels up the conversation in the way you’re… Because typically businesses look at customers, not visits or visitors. And when you’re able to take the digital metrics you’ve got and connect them to how the business is talking about customers, you make so much more sense to other people in the business and suddenly you’re having a lot of traction with the data analysis and insights that you’re providing. And that’s something I’ve experienced for sure.
35:57 MH: But we do need to do last calls, who wants to go first with their last call? Actually you know what, forget both of you I’m going first ’cause I think Moe has a really good one and I’m afraid Tim might be…
36:09 TW: Dibs on second.
36:10 MK: I feel like there’s hype now.
36:12 MH: There’s hype yeah.
36:14 TW: I think Moe thinks she has the best…
36:15 MH: Ever.
36:15 TW: Last call that’s ever in the history of mankind.
36:18 MK: No I think it’s the funnest I never said best.
36:20 MH: Funnest, best, same thing. Okay, I am aspirationally last calling a conference that I would like to attend, maybe even next year, if I can make the appropriate case. It is the MIT Sloan Sports Analytics Conference, and I don’t have any… I mean, I like sports as a fan, I’m not doing sports analytics. I am an analytics person and I like analytics, and so the combination of those two things I think I would find endlessly interesting. And you might remember if you’ve been a longtime listener Ben Gaines was our guest on our first year of the podcast and was talking to us about sports analytics and things like that. And he’s attended this event and actually…
37:03 TW: Multiple times.
37:04 MH: Yeah many times. And so I’m really intrigued and so I’m kind of like starting to scheme right now at this time of year. So that’s February 2018 in Boston. So if you are a sports fan and you love analytics, definitely something to take a look at. What about you Tim?
37:25 TW: [chuckle] So my last call is because… Let’s just pretend that I read it very quickly and there’s a lag in when the actual show comes out, but I recently finished Jim Sterne’s latest book, the Artificial Intelligence for Marketing Practical Applications, and I really enjoyed it. So I’ve started using it when I’m talking to people about machine learning and AI and data science in R, of saying, “This is a good way to get inspired.” He has gone through an insane number of interviews and papers and books and so there are tons of examples and quotes of people. So I actually found it to be really interesting. And he puts it in the book pretty early on he makes the comment he says, “This stuff may seem like it’s over the horizon, and it may be, but it’s just over the horizon.” So I found it a little bit of kinda validating that yeah a lot of the stuff we’ve been talking about around the data science and reinforcement learning and AI and agents is all stuff that really is coming.
38:28 MH: It’s all happening.
38:30 TW: That all stands alone yet I also just from podcast interest we are actually called out on page 263 basically because Christopher Barry was on and there’s like a three paragraph quote from him so it’s not that we actually did anything it was just a scrolling through the acknowledgements there’re multiple of our past guests who are credited there. So I’m like “hey I think we’re getting some good people” that was all kinda cool.
38:58 MH: Validation that we pick good guests, good looking out.
39:02 MK: Sometimes make guests into hosts.
39:05 MH: Very rarely, Moe, very rarely.
39:12 TW: I don’t know that we have enough data from the…
39:13 MH: That’s a qualitative dataset not a quantitative one.
39:19 TW: I think it’s a binary one ’cause we’re dealing with ones and zeros…
39:22 MH: Oh, there you go. Okay Moe what is your last call? We’re all…
39:27 TW: Knock us out of our chair…
39:29 MH: Let’s interrupt Moe a couple more times so the audience is just like, “Let us know what it is already.”
39:36 TW: “Let her say what it is.”
39:37 MH: Okay, sorry, okay Moe.
39:40 MK: I’m concerned because I’m worried everyone will have heard of this and I’ve only just discovered it, so for me it’s like this the coolest thing ever, but…
39:49 MH: It’s not Rick and Morty, is it?
39:51 TW: You’ve already talked about Domo.
39:52 MK: Has anyone heard of…
39:54 TW: Sorry.
39:55 MK: Jeez, no this is way more fun. So this is using a neural network to recognize doodling so that’s hint number one have you got it yet guys?
40:07 MH: Yes, yes it’s the Google…
40:09 MK: See… The Google Quick Draw, I’m obsessed with it.
40:13 MH: It’s awesome.
40:14 MK: See, now I feel like it’s a let down. Now actually, Rod, our next guest is the one that introduced me to this, and I have spent countless hours on my phone sitting on the couch drawing a sheep and… Anyway it’s super, super fun. So if you need a little bit of distraction I highly recommend checking it out.
40:33 TW: See when I sit and try to draw on my desktop and realize that that is like… I am not a great drawer as it is but trying to do it with a touch pad it really… But I’ve actually gotten a couple of my kids kind of hooked on it as well, but then I try to talk to them about how it’s AI and work and one of them has continued to be kind of intrigued by what the…
40:53 MH: Yeah. I messed with it for a little while and then I realized I’m literally training a computer to take over the world and I stopped.
41:03 MK: So you gave up, okay.
41:07 MH: Also I’m… Like Tim, I’m an even worse artist, anyway.
41:12 TW: Which is still what’s impressive that there are many, many other shitty artists out there, ’cause it can tell that this is a corn cob pipe from… What I drew looked like oak tree.
41:20 MH: It’s pretty impressive.
41:22 MK: So the [41:22] ____ plain is is that we’re never gonna build up any of my last calls ever again ’cause I feel that was like a [41:27] ____.
41:28 MH: No what no, that’s a good last call.
41:30 TW: No. There’re a ton of listeners who have not heard of that and this will make them… It is both addictive and cool and it’s AI that’s the thing is that it’s… That if you read on kind of what it’s doing and how it’s doing it’s some of the stuff we’ve touched on on past episodes it’s mind blowing not just that it’s cool but the actual under the hood stuff that’s going on is pretty slick.
41:56 MH: We’ll set up a vote in the Measure Slack, Moe, about who has the best last calls and we’ll try to stack the votes in your favor. No problem. [laughter] Anyway, as you’ve been listening I hope you found something worth chiming in about, because I think tracking visitors across devices, you know of course as you listen to the podcast, if you should start listening on your computer and then switch to your phone in the car, we are tracking that. No we’re not. We just get two downloads and we thank you. But we would love to hear from you because that kind of qualitative data is all the data we need to keep our little podcast going. So reach out to us, let us know what you thought. And you can do that on the Measure Slack. It’s a great place to do it, our Facebook page, our website, analyticshour.io, or on Twitter. And I have two co-hosts on the show. You may have met them. Moe Kiss, only slightly a relation, but not Michelle Kiss, just Moe Kiss, and Tim Wilson. And on their behalf, I’m telling all of you analysts out there to keep on analyzing.