1. Home
  2. /
  3. EDGE of the Web
  4. /
  5. Transcripts
  6. /
  7. EP 314 Transcript | Automation and Machine Learning with...

EP 314 Transcript | Automation and Machine Learning with Ginny Marvin

By Site Strategics
June 5, 2019

 

Ginny Marvin Interview Transcript

 

Speaker 1: (00:01) On this episode of Edge of the Web.

Ginny Marvin: (00:04) Yeah, I mean, I think this is an area where we are all going to have to do a lot of professional development. I mean, the exciting thing is there’s ton of opportunity with this, and machine learning is going to yes, evolve what we do and how we work, but those who see, you know kind of first mover, I think will have the advantage here.

Speaker 1: (00:35) Your weekly digital marketing trends, with industry trendsetting guests. You’re listening and watching Edge of the Web, winners of Best Podcast from the Content Marketing Institute for 2017. Hear and see more at edgeofthewebradio.com. Now, here’s your host, Erin Sparks.

Erin Sparks: (00:56) We certainly want to introduce Ginny to the show, and obviously we were talking about this earlier. We’ve been reading Ginny’s articles from those respective publications for a long time. That’s made the news here at the Edge, so it’s great to be able to have you on board, Ginny, and you’re certainly doing an incredible job in those spaces.

Ginny Marvin: (01:17) Thanks so much.

Erin Sparks: (01:18) You’re more than welcome. So let’s introduce you to our audience, who may or may not know you, and if they don’t I will hear of it because they’ve been hearing from us for the longest time. Ginny is actually Third Door Media’s editor in chief, managing day to day editorial operations across all of their different publications. She also writes about paid online marketing topics, to include paid search, paid social display and retargeting for Search Engine Land, Marketing Land, and Martech Today. With more than 15 years in marketing experience, she’s actually held both in-house and agency management positions and I’ll also go to add, she is on a regular conference circuit. I think you’re speaking either keynotes or you’re part of panels on a regular basis, not only for SMX and Third Door Media conferences, but a number of other conferences outside that, right?

Ginny Marvin: (02:13) Yes. We’ve got SMX Advance coming up this coming week in Seattle.

Erin Sparks: (02:19) That’s right. That’s right. June, is that third and fourth and fifth?

Ginny Marvin: (02:21) Yes, yes. Third, fourth, and fifth.

Erin Sparks: (02:25) Excellent. We reached out to Ginny because she’s got, she’s speaking of a particular topic that we at the Edge have been focused on for a good number of months, and that’s machine learning. Before we get into that, we really want to know how you got into your job and how do you manage to keep all the publications straight as you’re pushing the content out? Just briefly give us your backstory before we get into it.

Ginny Marvin: (02:50) Yeah. So I have been working in digital and search marketing since 2005, and in 2013 started writing for Search Engine Land on paid search, pretty much exclusively. A couple of months into that, decided to hop over full time and I started writing more for Marketing Land and then we launched Martech Today as well. I have been in the editor in chief role since, I believe, August or … yeah, about August.

Erin Sparks: (03:34) Well congratulations on that promotion, too.

Ginny Marvin: (03:34) Thanks. Thanks so much.

Erin Sparks: (03:34) Excellent. So you guys are certainly putting some great information out into the ecosystem and digital marketing. We have also been watching your contributions in the space of machine learning. It is certainly rising to the top of a number of conversations in the paid marketing space. For our digital marketing audience, who may not have heard our previous shows about that, why should we be paying attention to machine learning at this point in time?

Ginny Marvin: (04:04) Because it is literally affecting everything we do. In digital marketing, specifically, but machine learning and automation are affecting a host of other industries, pretty much every other industry. In digital marketing and search, specifically, machine learning is affecting how organic search results get shown. It’s affecting every single aspect at this point, pretty much, of paid search and digital campaigns from the ways that ads get created to the way vids get implemented. So from … in a way our attribution is getting delivered as well. The reason to pay attention is because it’s an undercurrent and everything’s happening.

Erin Sparks: (05:05) It’s already there, and we’re already actually steering by a number of different models of machine learning in these ad platforms. You referenced, a while back, opportunity tabs inside of Bing ads, and on top of that, Google ad rotation is all being guided by machine learning, correct?

Ginny Marvin: (05:23) Correct. Yes, so the opportunities tabs, that’s an interesting place, because there’s obviously a lot of benefits to machine learning, and the ability to get surface insights quickly without having to dig into massive Excel sheets is one of those benefits. Obviously the caveat comes when the platform that you’re paying to serve ads is the one that’s giving your the recommendations. So, that’s no different than the recommendations and taking those with a grain of salt that we’ve always done. But, you know, what we’ve seen over the years is those recommendations from Bing and from Google and … I shouldn’t say Bing, it’s Microsoft Advertising now. Microsoft Advertising and Google ads, that that have become more sophisticated. They’re certainly more robust in the variations and breadth of recommendations that they’re able to make within these platforms. That’s just one example of how we’re seeing automation kind of free up our time from a tactical and having to be … not only just analyze, but take hours to compile the data that we need to analyze.

Erin Sparks: (06:57) That’s right. And it gives us more free time to be able to create more and more experiments with that data and kind of freeing us up of that mundane execution, right?

Ginny Marvin: (07:07) Right. Yeah, I mean that’s the great promise here, is that we can move away from mundane tasks and have more opportunity to be strategic, look up and see the big picture and execute based on the kind of bigger picture and more strategic frame points.

Erin Sparks: (07:31) From a kind of a basis of understanding for our marketing folk here, artificial intelligence is not what we’re talking about, everybody. Artificial intelligence is this kind of sweeping concept of if software mimics human behavior. That’s not in the equation.

Erin Sparks: (07:51) Machine learning is a set of technologies that make it possible for computes to solve problems autonomously based on prior inferences from the data. So you got to be able to infer from that data, so that means you have to put together models in which the computer can actually infer. I mean, it’s all about building those models to be able to guide the computer, right?

Ginny Marvin: (08:14) Yes, exactly. So we are … in building the models, obviously, is a highly desirable sought after skill and these companies are all fighting for talent. The thing with machine learning is the lack of transparency for us in the end user chair. So that, there’s so much trust that has to go into using these algorithms, taking advantage of these algorithms, and that’s where there’s tension. Between Google and Bing, or Facebook and LinkedIn are saying, “Use our bidding algorithms. We have smart campaigns. We’re going to run these for you,” and in the meantime, the advertisers in many respects are kind of the dated guinea pigs, feeding the machines to help them learn and train and become better at what they’re doing.

Ginny Marvin: (09:24) So we’ve seen this evolution from really, really terrible machine learning where, you know, a lot of these algorithms, when they first came out were not good, and they’ve gotten better. They’ve been able to work with much less data, so you know, things like Google store visits, you don’t need nearly as much data now to start getting conversions showing for store visits. Things like that, but at the same time, there really is still this black box nature of machine learning that is going to be this enduring source of tension between the users and the platforms.

Erin Sparks: (10:13) You know, it’d be powerful if we could get, pull back the curtain a little bit and be able to see some of the models and some of the behavior analysis, but that’s the IP. That’s the difference between one platform and another platform. We certainly would appreciate if we could actually see how that’s being interpreted because at the end of the day, we’re, again, a bit chafed by trusting these applications. I mean, a lot of times, being steered straight into a mountain by the predictions that didn’t understand context.

Ginny Marvin: (10:54) Yeah, and so much of this depends on the data that gets put in.

Erin Sparks: (10:59) Yup.

Ginny Marvin: (11:00) And so, you know, we don’t really have visibility into … and not the actual data, but the … what is informing these data sets and we, everyone knows that bad data makes bad models, so we are kind of beholden to if there are ten advertisers that are really optimizing their campaigns well and they’re doing all this great job and then there are 20 not doing a great job. All that data’s getting fed into the same model, theoretically. I don’t know for sure, obviously, but how does that affect those who are optimizing really well? Those are the kinds of questions that make me really interested in how all of this is going to play out and those are the kind of questions that I think we all need to be asking.

Erin Sparks: (12:01) No, you’re absolutely right. I mean, so the intent of this machine learning is to be able to create, pick up pattern recognition, and I mean, those patterns come from the models that are created. So would it be a huge stretch if the advertising platforms started to open the doors to these advertisers to be able to help create the modeling sandboxes that they can actually contribute into their understanding of their marketing target, not just being beholden to how the platforms are interpreting data and modeling at a particular date? If there is a way to be able to have even a factor of control, or I should say contribution, into how to make sense of these things, right, would that be a place that would make a heck of a lot people more comfortable?

Ginny Marvin: (12:57) Uh, perhaps. You know, perhaps. I would say, though, that is probably not that simple. I wrote about Zillow’s efforts to build an AI, machine learning, in-house in order to run their bandits testing for user experience on their sites. Based on a talk I saw from one of the AI engineering leads at Zillow, he gave a talk at our Martech conference this fall, and I mean, it’s really interesting to talk to somebody who is building these models and the work and fine tuning that has to go into them and testing to make sure that they’re actually working right. So, yes, I think a sandbox could be that kind of environment and would be great for some, but for the vast majority of us, it’s not going to be something that we can go in and play with and benefit from. It’s going to be the select few who have the skills to be able to understand how [inaudible 00:14:19].

Erin Sparks: (14:21) Yeah, and that’s the thing. The skills are highly coveted right now for be able to, people that can actually do data modeling for machine learning algorithms. You don’t have certification for that. This is some deep channel stuff here. So, now I was just … it just dawned on me, because of such a gray area and so how it is so nebulous and we have to literally trust on it’s face what it’s giving us as guideposts, or at least feedback for optimization and campaign steerage, having at least an area in which you can create some sort of contextual mapping because you, the marketer, should know more about your audience that you’re trying to target and be able to give some sort of insights into these platforms. Maybe I’m just thinking from a utopian standpoint, but it would be, I think, of values to these platforms because is the one thing. They’re trying to interpret context, decision making context, infer different patterns with the modeling that they’re doing, right?

Ginny Marvin: (15:28) Yes, and you know, I mean they’ll say that that’s what they’re giving when they’re showing you projections if you increase your budget or if you increase your bids and those based on historical data from, you know, auction data, that those are going to be … you’re going to be able to see those kinds of projections.

Erin Sparks: (15:54) Oddly enough … go ahead, keep on going.

Ginny Marvin: (15:55) No, no, no. You go.

Erin Sparks: (15:59) Oddly enough, I’ve never seen Google ads tell me that I should decrease my budget.

Ginny Marvin: (16:05) Well, you know, that is actually … I have found that the new recommendations, they are starting to be more nuanced with that, so that is a positive sign.

Erin Sparks: (16:18) Absolutely.

Ginny Marvin: (16:18) Not a lot, but you will start to see some recommendations for if you lower your budget or you know, projections score you would get the same number of conversions, fewer clicks because they’re able … whatever biz strategy that they’re trying to push your towards would be able to give you the same results. But again, then we go into the biz strategies.

Erin Sparks: (16:44) Yup. And my next question was just kind of in that same space. Defining these models, even from a high level view, the marketer, the digital marketer, is not about to take their steps into a whole new world where they were first, they were finally realizing all of the different measurements that they can pay attention to, and there’s a heck of a lot of attribution, heck of a lot of mutli-touch concepts that are now in the eco-system, right? But now they’re going into this predictive behavior and I guess my question is are we disciplined enough as a digital marketing world to be able to take on this type of modeling and be able to contribute back into it?

Ginny Marvin: (17:34) Yeah, I mean, I think this is an area where we are all going to have to do a lot of professional development. I mean, the exciting thing is there’s tons of opportunity with this, and the machine learning is to, yes, evolve what we do and how we work, but those who see, kind of first mover, I think will have the advantage here because we’re not going back. There is … we’re not going to not have machine learning. We’re just going to have more of it and so those marketers that are willing to say, “I’m going to keep asking questions. I’m going to keep my skeptics hat on, but I’m going to start figuring out how, for my business and for my career, start leveraging this to our advantage.” Those are going to be the winners.

Erin Sparks: (18:38) Yeah. Yeah. You’re absolutely right. We’re not going back, and I think my analogy’s changed a little bit. It’s no longer SkyNet. I think it’s the Borg, you know. Resistance is futile. We are going to be absolutely be in that space. But there’s also this hesitation that I don’t have enough data, that I don’t have enough to be able to make these types of decisions, or I don’t have enough information. A quote from Pedro Alves, the CEO of Opal, he says, “You don’t need to have perfect data sets to get predictions. You can have holes and inconsistencies and still manage to find solutions you’re looking for.” Basically if you can continuously tweak the models, you can overcome not-so-clean data challenges. You have to just let the machines do the work and learn from it.

Ginny Marvin: (19:25) Yes, and I think that’s correct, and I think the clean data … the gaps may not be your downfall. If you have bad data, though, that’s not good. I mean, you’re still going to need put a focus on cleaning your data or whatever term you want to use for making sure that this is good stuff going in, but I think that’s probably completely right about the gaps. The more these models train, the less they need. And that’s certainly what we’re seeing, and have seen over the past few years as the platforms have evolved their own machine learning. They need much less data now to make predictions and to have the models kind of to stop learning and start acting.

Erin Sparks: (20:34) But even that, that’s a scary point right there that you’re mentioning, that they need much less data to make judgements on that. You’re just hoping that it’s not preening through your data that you put in there with potential gaps or what have you, but it’s also … it’s got the bigger model from all the other data that it’s pulling in from all these other campaigns that are maybe more rich context so it can actually borrow from those type of models to be able to apply that. That’s the only way that I can see this working. Otherwise, it’s kind of like an echo chamber. You’re making decisions, if you’ve got an audience of ten and one goes this direction, you’re making a ten percent direction from a terrible data set, right?

Ginny Marvin: (21:24) Yeah, I think so … thinking about Google, obviously there’s probably no company on the planet that has more data, and so it makes sense that their models need less and less of it because they’ve had these models training forever. But, yes, to your point and what I was kind of getting at earlier is that the integrity of that data is not apples to apples across these various implementations.

Ginny Marvin: (22:00) So when you’re talking about campaign data and you’ve got, say, ten thousand campaigns and you’ve got some advertisers, we’re talking about [inaudible 00:22:18], you’ve got multi-national chains that are bidding against three or four outlet regional chains. The budgets are different and Google said, “We control for all of this, and we’ve learned and our models have been training against all of this,” but you know, I really do think that it’s worth asking the questions about how the models are actually working to your benefit and really taking a critical eye to your campaign results.

Erin Sparks: (23:03) So, where do the digital marketers stand in all this? Their participation? They’re just not observers, they are certainly executing the campaigns, but is there … obviously there’s common dialogue right now, but marketers far and wide need to be able to contribute back, maybe not to the platforms, but to at least share with each other is what I’m seeing, without giving away the farm of the strategies that they’re executing. Is there an environment happening right now where top PPC and ad spin marketers in the digital space are sharing their thoughts and their experiences as it comes to machine learning and these platforms?

Ginny Marvin: (23:49) Yeah. I think we’re seeing conversations like this happening on Twitter and Facebook groups and in columns contributed on Search Engine Land, and certainly at conferences and events next week, I think machine learning will be a topic that is kind of a thread throughout pretty much every single session, if not conversation that attendees and speakers have with each other over those few days.

Ginny Marvin: (24:22) I think it’s something that is going to be just a key part of the fabric of everything that we do, and there are really big questions to be asking, and there are really big opportunities here. I am not anti-machine learning in any way. I think there’s huge promise here. We just need to be not just-

Erin Sparks: (24:57) Passive.

Ginny Marvin: (24:57) … bystanders. Yes. Right. Not passive. You know, there’s just a lot of really interesting things to be thinking of for our business and our careers as all of this continues to evolve.

Erin Sparks: (25:16) No, well said. We got to get used to it. It’s here, and we also have to participate in the evolution of this as well. You know that the platforms are listening in one way, shape, or form, right?

Ginny Marvin: (25:31) Yeah, and I think advocating for transparency is really important and that means both internally within our own organizations when you’re talking with clients, when you’re talking to stakeholders about what is happening in this realm. What can we see, what can we not see? What, if we’re using this attribution platform, what does that mean in terms of … are there any biases that we could be thinking about from the attribution platform itself? Would it behoove them to … can be thinking about it, what the waiting actually looks like, even if it’s a black box. Things like that.

Ginny Marvin: (26:12) Externally, advocating for transparency for more transparency from the platforms, and why is it important, what should we be able to see, what kind of ways can we have dialogue and feedback through all of this?

Erin Sparks: (26:30) Absolutely. Well, that’s our course and we certainly want to contribute into it. What are, to kind of wrap up these concepts here, what are the risks in using machine learning for marketing right now? I mean, we’ve covered a number of them, from the bad data side of things, but what do you see as the risks?

Ginny Marvin: (26:52) So I think the risks are jumping in without having a plan for analyzing the outcomes, and kind of taking it for granted. So I think, you know, there are a number of different avenues that we’re actually in many ways no longer having any choice over using some of these automated mechanisms. When you adopt them, you start really paying attention to all the various factors, and there may be some metrics that look terrific, and then you turn around and three months later realize your farther downstream metrics don’t look so great, or vice verse. I mean, the opposite happens too, where it looks like this is not working. This is, you know, oh boy, I better turn off this automation and then it turns out downstream oh, actually, it did perform pretty well. So you just need to really think ahead of what do I need to be looking at and with a lot of this is have some patience because things take time to start to work. So that’s the hardest part for a lot of us. We want instant results and if you start seeing things go badly quickly, the knee jerk reaction is-

Erin Sparks: (28:24) Yank the plug.

Ginny Marvin: (28:25) … pull the plug. Yes. Yes. And that’s often not the best course of action, particularly with machine learning.

Erin Sparks: (28:30) No, no, and I’ve even contributed into that, as another two cents is create an audit log, okay? A journal of everything that you’re doing, everything that’s being suggested to you. Of course, you can go into the platforms and look at their change history, but it’s creating arguments of how to change your delivery in one way, shape, or form. This is the experiment being presented to you. You need to set up your journal to be able to track everything that’s being executed on that, so you do have to reversion control but you’ll start learning these experiments as this system’s giving you some feedback, right?

Ginny Marvin: (29:13) Yeah. I think that’s a really great piece of advice. The documentation is key. Also, you know, kind of watching to see how the results change and evolve during your learning process is really important to do.

Erin Sparks: (29:33) Got to pay attention. You can’t trust it on its merits, but you, digital marketer, have a responsibility to truck along with these suggestions and watch very intently because it’s making decisions off of how you’re also presenting the data back to the system. Make sure you have all your conversions mapped, please.

Erin Sparks: (29:57) So, kind of wrap it up as we’re looking at machine learning. Top three benefits that a digital marketer gets from these pattern automation platforms inside of the digital marketing. Give us the top three things that we, as marketers, benefit from.

Ginny Marvin: (30:16) I would say the ability to be more strategic. To look up, take a big picture view and we’re freed up from a lot of the mundane tasks. That’s kind of the big one, right, and the big promise here. Another benefit, obviously, is the real promise is better results. Can we get better results with less effort and that’s the … many are saying yes absolutely, some are saying TBD, so that I think is the number two with a caveat. And number three, I think it is a real opportunity to start thinking outside of our silos and being able to look across channels and departments and teams and work much more collaboratively together.

Erin Sparks: (31:18) Very good. Well, like you said, it’s not going away and we have a role in this evolution, so all digital marketers should have that experimenter’s mindset going into it. It’s just set … well, you’d never want to set it and forget it. And the platforms are not going to let you any more. So, there’s that.

Erin Sparks: (31:41) It’s a pleasure speaking with you, Ginny. Any final thoughts on the … you’re about to make the presentation next week. Any final thoughts on machine learning?

Ginny Marvin: (31:51) I think that, obviously, the underpinning here is data and I think that is another big opportunity for digital marketers to think about how to be the data leaders in their organizations and we’re seeing roles be developed for chief data officers. But there is … obviously digital marketers, data is core to what we do. We’ve been working with audience data and certainly performance metrics data and thinking cross-organizationally about how to be a steward in this area I think is a great big opportunity as well.

Erin Sparks: (32:45) Yeah, it’s beginning to be a bit of a [inaudible 00:32:48] industry, and the data specialists that are out there, and they’re going to be sought after. So it’s a heck of a career path, guys and gals, if you’re in that space and you have a particular proficiency. Keep your data clean as well.

Erin Sparks: (33:07) Again, it’s been a pleasure speaking about this with you, Ginny. We usually wrap up our shows asking a couple of questions. What really bugs you in your industry right now?

Ginny Marvin: (33:19) Oh-

Erin Sparks: (33:19) And you can name names, too. Go for it. Go for it.

Ginny Marvin: (33:24) No, I mean, I get a … nothing bugs me other than complaining without solutions. You know, we’ve got … obviously there are always challenges and things are always changing and so have something to say beyond the complaint. That’s a little bit nebulous, but …

Erin Sparks: (33:54) Oh, we know exactly what you’re saying. Very good. Well, conversely, what excites you about the industry you’re in, because you’ve certainly seen a lot of change in the digital marketing space.

Ginny Marvin: (34:06) Yes, and I … what excites me so much is the, and it always has, is the openness that this industry has to share and learn from each other and grow together. Everyone kind of wants to see this industry grow and help people out and so that, you know, has just been one of the big inspirations for me to be working in this industry.

Erin Sparks: (34:37) It is really neat to see that, because decades ago, everybody was closed. Their cards were close to their chest. They weren’t really divulging, or they were touting themselves more than contributing into education. Now, you’ve got a good set of people that are actively wanting to share, to make sure that people don’t make the same mistakes that they made, but it’s also this rolling desire to go to that next level, right?

Ginny Marvin: (35:06) Yeah, yeah, yeah. I mean, the word community can start to sound kind of trite, but it really is that, the community aspect of this industry that is inspiring and really exciting.

Erin Sparks: (35:21) I dig it. I dig it. You gave us a fun fact before we went on. You’re an avid sailor. That’s too cool. That’s too cool. How many sailing and boat references have you made to digital industry? The analogies? How many have you done over the years?

Ginny Marvin: (35:40) You know, I kind of reserve those because I feel like probably most people have no idea what I’m talking about when I say we’re going to head upwind and tack and you know, so I don’t throw those around very often, but perhaps I should start to.

Erin Sparks: (36:00) Yeah. Sow them in there, and there needs to be more sailors out there, and if you can get more digital marketers understanding … we drop analogies. I’ll string together metaphors and analogies like four or five long in a sentence, which is really terrible. You need to speak in analogies, because people get it that way, and if they don’t follow you in geek-ese, they’ll certainly follow you with boat references.

Ginny Marvin: (36:24) Right. Yes.

Erin Sparks: (36:24) Absolutely do that. Is there anything that we can promote for you on the show today?

Ginny Marvin: (36:29) No, I just … we’re really looking forward to SMX Advanced in Seattle next week, and hope to see many of your viewers there. That’s it. And if anyone is also a sailor and wants to chat boat references, you can find me on Twitter.

Erin Sparks: (36:48) There you are. We’re certainly going to … we’ve been at SMX conference before. We’re certainly going to swing back around and see if there’s an opportunity to broadcast at one of these in the near future because it’s great to be able to see so many people that are, again, thought leaders in the space. It’s a fun thing and there’s so much you learn. We’d love to be part of that environment as well.

Ginny Marvin: (37:12) That’d be great.

Erin Sparks: (37:13) All right. So you certainly want to track Ginny down on her Twitter handle, ginnymarvin and on LinkedIn at Ginny Marvin as well. We certainly appreciate your time. I hope you don’t mind that we’re going to keep on quoting your articles each and every week. You’re doing a bang-up job and we really appreciate what you’re doing for the digital marketing industry.

Ginny Marvin: (37:38) Thanks so much.

Erin Sparks: (37:39) You’re more than welcome. More than welcome.

Erin Sparks: (37:40) All right. That’s a wrap. Let’s thank everybody here. We certainly appreciate our listeners and our audience. Please make sure that you like and review and give us some feedback on our show, because we are always listening and that’s how we do our own marketing as well. So let us know, not only how we’re doing but also who you would want us to interview next. Be sure to check out all of the information that we have over at edgeofthewebradio.com. That’s edgeofthewebradio.com. Thanks to our colleagues here at SiteStrategics forever continuing to create some good production on the Edge, as well as especially our guest, Ginny Marvin, this week.

Erin Sparks: (38:20) We’re going to be dropping in here very soon with another interview, so keep on checking back every Thursday at 3pm. However, we may be changing our course a little bit along the way. See, I did it. I literally did a boat reference right there. So check it out, because we may actually be breaking some of the new segments out to you as quick parcels information from our guests along the way.

Erin Sparks: (38:43) So from all of us over at Edge of the Web, thanks so much. Don’t be a piece of cyber driftwood. We’ll talk to you next week. Bye bye.