In this episode of In-Ear Insights, we’re featuring some interviews that Trust Insights team members have done recently on podcasts, live video, and broadcasts. Today’s show features a half hour conversation with Christopher Penn and the B Squared Media team about the jobs AI will and won’t take, as part of the run up to his keynote at the HELLO Conference in late March. Enjoy the episode!
Listen to the audio here:
- Need help with your company’s data and analytics? Let us know!
- Join our free Slack group for marketers interested in analytics!
What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for listening to the episode.
This is in your insights the trust insights podcast.
In the next couple of episodes of inner insights we are featuring some audio interviews from broadcast podcasts and shows we’ve appeared on recently. Today’s episode is an interview I did with booksellers and our team from B squared media other live show be to TV catches all that books Hello conference in late March in New Jersey enjoyed this talk
everybody welcome to be to TV. And you as you notice that broken I are not alone today. But I’m going to mute before my dog has a complete panic attack and the other room and I will let her take it away.
Hi everybody. So we are here today with someone who I absolutely adore and had the pleasure of meeting life. I’ve actually stopped you for quite some time but Chris and I had the pleasure of meeting last year at Social Media Marketing World over tacos and Margarita I think when we first came to
To have our first conversation, but if you don’t know who Christopher pin is, he is the co founder of trust insights. And they are a data and analytics consulting company. And essentially what they do is really simple. They help people like me. Marketers understand their data and analytics better and have better results through that data. So, yes, exactly. Chris carries making the heart symbol because she’s our resident data nerd, everyone. Thank you so much for joining us today. We’re really, really, really excited to have you and ready to like, you know, have you explode our brains? Well, thank you for having me. You bet. So, I think we want to start off with like, some softball sized questions, if that’s okay with you. And then we’ll we’ll try to get into like some of the scarier stuff as we go along. So I think first and foremost, right. Some people even marketers when they hear artificial intelligence or AI today, they still don’t know what
We were talking about. So can you explain in layman’s terms, like so easy that someone like me could understand what you describe is AI.
So the formal definition of artificial intelligence is getting software, getting computers to be able to perform tasks which don’t typically require human intelligence. So if you can understand the sounds coming out of my mouth and parsing that into words, that’s called natural language processing, right? We do it, we learned it as babies, we have to teach machines explicitly how to process like, if you can see the image on screen and and differentiate me from the background you using vision and there’s, you know, there’s a whole field of computer vision. So it’s getting machines to emulate human intelligence capabilities, things that humans can do now, AI is a super big, super broad term that can literally mean almost anything that fits that category of trying to get a machine to do what a human can do. So everything from robots
Next to process management, to speech to composing music, all these things would fit under the umbrella of AI, where marketers are you paying attention is in a subset of AI. And that subset is called machine learning, which is when we teach machines to learn. So traditional software, like your word processor, like an app on your phone,
humans write the code and then the machine spits out data, right? Could be, you know,
silly awards and Candy Crush, it could be word processing documents could be video from your phone. All these things are data that the app produces. Machine learning is the opposite. You feed a whole bunch of data to machine learning software and it writes its own software, it design software for itself that creates a certain type of outcome. And very broadly speaking, there’s sort of like there’s two big categories of machine like this. supervised and unsupervised. supervised means I want to know
Something I want you to explain a result better. So let’s say you’re doing, for example, social media analytics and you want and you have the customers revenue numbers and you have all your social data because you’re going to use supervised learning to say, Okay, I want to know what data points, what factors, what dimensions lead to have a relationship to that revenue number. And so you would use supervised learning techniques to find that the second type is called unsupervised learning where you know the outcome, you don’t know what’s happening. So a really good example that is social conversation monitoring. When you put in like a hashtag, you don’t know what people are saying. So you download 10s or hundreds of thousands of conversations. And the machine kind of arranges them says, Hey, here’s a group that seems to cluster together. He is a group that seems to cluster together and this was talking about tacos and this was talking about, you know, building walls and things and this one over here is talking about you know, fish that that you know, evolved into great
And whatever, whatever all these different conversations are, you didn’t, you didn’t know that was going to be the outcome. But you need the machine disorder. So very broadly, those are the two big categories. And what we’re talking about AI for marketers, we’re talking about marketing, trying to do three things faster, better, cheaper, right? So AI should help us do our work faster. When there’s no way any of us could read 1000 tweets a second, right? There’s no way this could be 1000, we probably most of us would not want to be as tweets a day.
So acceleration. Second is accuracy, better quality data, again, humans are prone to mistakes. And if you’re living your life inside of spreadsheets, there’s a good chance that you or someone in your chain of command is introducing errors along the way, if we can automate and if we can, can improve the accuracy by taking that away from the humans and giving it to machines to do the same repetitive stuff will get better accuracy and the third is is cheaper or more efficient, which means
Getting stuff that is repetitive and low value away from humans and getting it to the machine so that you, the human can do more valuable things. So I remember at a company I used to work at, there was this one person on this one team. And they literally spent eight hours a day, a full workday, copying and pasting stuff from one spreadsheet to another. And I like I’m pretty sure that entire appreciate we can replace this task with machine the personal be happier, the work will be better and so on and so forth. So those are the three things that marketers should be looking for when they’re when they’re looking to embark on using machine learning to make life better see I love that you explained it that way because the way you’re explaining it you know the articles that we’ve been seeing right or like take your jobs the robots are coming you know it’s it’s like made into this like horrible scary thing and it’s not right it’s the three examples you just gave are really excellent, but it’s also not completely removing the
Human, which basically, we, we know you have that mindset, because that’s where you’re going to be attending and speaking at the conference, right? We’re not removing the human from me intelligent tools loop, but we are helping the human make better decisions faster and taking out maybe some of those low level tasks so that we can be cheaper and more efficient maybe.
And I say maybe because there’s a fascinating piece from a reporter who was at Davos last week. And they were, they were privy to, and talking to a number of, you know, CEOs and, you know, major movers and shakers and the behind the scenes conversation for corporations is we want AI to remove as many jobs as possible because humans are expensive this one representative from a major bank said Our goal is to take thousands of jobs and turn them into dozens because we can’t we will afford will have you know, much better results and we’ll make our margins Will you know, our quarterly numbers for the stock market will be so
What’s better a lot of CEOs at that event ci is the golden ticket to hitting their quarterly numbers and, and shedding a lot of headcount. So one of the things that’s important for all of us to understand is, there are a number of things that you should be doing in your career to make your job more resistant to being taken by a machine. The more repetitive your your job is, the more repetitive every task is, the easier it is to automate away the more creative or collaborative or
broad in scope, your job is, the more difficult it is to automate. You can automate pieces of it away, but you can’t automate the whole thing away. Right? machines are really bad at things like general life experience, they’re really bad empathy, a really bad broad judgment.
And so things that make you a great human are things that you will double down on machines have machine algorithms have a very difficult time crossing.
disciplines. So if you are a a Twitter ads manager, right, that’s a super easy task.to automate, right? It’s almost automated as it is. But if you’re a Twitter ads manager who is also good at data visualization and project management, you’re super hard to replace. Because you you have multiple things you bring to the table. So the more we as people can
build up our skills base across disciplines and network within our companies to to, you know, have that data to have a you know, what is a finger and every pie or some
making analogy that makes it harder for you to be replaced? I love it. Okay, so good. So there it so it’s, it’s not as scary as for me, it’s scary to be scary if you’re smart about it, right? And truly, I think, you know, this is just my viewpoint but as someone who owns the company, you know, for me that’s, I love that answer because it means the people who should
should be on the team will be making those smart decisions on how they can remain on the team and the people who you know don’t want to go the extra mile to get a weed themselves out because I’m going to be talking with smart people like Chris on how we can automate some of these tasks that right now we’re paying humans to do right exactly and at the very least what you want to do is you want to be able to even if you never change the amount of headcount you have you want to use machine learning to scale yeah so we’ve had a conversation in the past but I think it’s worth you know bringing here the amount of time people spend on curating content can be you know hours per week at trust insights one of the things that we do is we built our own software to gather data score it analyze it and then you know essentially pre curated and then it just goes to one final you know passive humanize go okay that even though that was a popular articles on topics, it’s still not something we feel comfortable publishing, but instead of spending 10 or 15 hours
We curating content for the week for you know our individual social accounts for the company account it’s five to 10 minutes a week so it’s is a great time saver delivers quality and frees us up to be able to do more valuable work yes and that is like Case in point why social media marketer should be looking to AI right because there are a lot of things that we have to touch right there’s a lot of high touch tasks that we do they could be replaced by machines not removing the human completely because you know we still want Carrie to go in once that created and put our own spin on it create our own creative leads maybe you know Carrie is is a is a gift queen. So she can, you know, be in there with the with the content that’s created. But yeah, I mean, I get it, I can see it already. So I’m hoping that through this talk, we can help you know, other people maybe understand what we’re talking about. So without giving away the kitchen sink or giving it away. I’m gonna kind of dive in a little bit too.
What you’re going to be talking about and how this conversation is going to come about at the helo conference, which is March 28, for those of you who are listening in, and we would love for you to come will post links down in the in the show notes. But Chris, what are you going to be telling everybody like, give us some of the knowledge bombs are going to be dropping, we’re going to look at five practical applications of AI specific to social and digital marketers,
mining, clustering, forecasting, driving
in within these within these applications. We want to look at the practicality of it. One of the things that’s very challenging about the way people talk about AI is that the assume it’s either a magic wand which is not as math
or they just assume it’s a big black box and and it just kind of does his thing and then you come back and stuff and none of those things true.
The first area we’ll look at is text mining. So what given a large pile of unstructured data
text like what’s in your emails or the people who tweeted to your all your Facebook Messenger bot conversations, whatever the the text, the body of text is
looking at ways that you can take out ideas from that or topics or concepts very, very quickly and be able to use that to guide your decisions to guide your content. Last week at Davos, we were not there, I was not there. I was sitting in this chair in my basement,
but we’re able to pull in an exam and
approximately 150,000 conversations a day and digest them down and understand, okay, this is what is you know, this is what’s being talked about at Davos, these are the major themes these and and do that, you know, a single run of that data. Well, let me ask you, how long would it take you to go through 150,000 conversations even with a full team days, days days, I don’t even want to think about it. Quite honestly,
the software can crunch through
That and come up with a single page summary in about six and a half minutes. So
right blowing up emoji of the conversation right here. But also Christmas kind of plays into what you guys have been doing it trust insights, which the conferences that you guys are involved up, which is like all of them. So PS if you again, if you don’t know Katie and Chris have Trusted Sites, you will because they speak at all of the major conferences, but you guys have been running these you, you remove the word cloud cloud words, and I forget what the word you use and replace was. But essentially you’re making all these conversations around these events. And you’re kind of telling the event organizers like, hey, here are the main topics of conversation happening around your conference, which is like Okay, number two brain blowing up emoji because like, That’s huge. It is huge. So text mining is it really is one of the principal uses of artificial intelligence for marketers to be able to digest down things.
quickly and get a sense of what’s in the box. Again, unsupervised learning. What’s the thing when you’re when you’re dealing with a massive amount of data, and you need to get it relatively quickly. If you are, if you have a booth, for example, at IBM think which is coming up in in two weeks, and you want to,
and you have a social media manager and you want to participate in the conversation the meaningful way and hopefully drive traffic to that booth. You want to be able to run this analysis while you’re sitting, you know, while you’re standing at the booth. So, you know, as soon as people leave the area and go to session, you start running your analysis, you start participating in conversations, you identify the conversation topics and you publish any relevant content you have so that it’s an alignment with what the Zeitgeist of the moment is, there’s no point in running a social analysis days after the event is over. You want to run it in the moment. Second application, it’s called network, graphing network. Graphing is where you digest down data and turn it into and you identify relationships between things that the most common and most powerful application of this is
Of course with influencer identification so at an event, yeah, one of the the algorithms we use looks at who is most talked about? So not who does the most talking not who’s got the biggest number of followers, but who is everyone else talking about, and looking to and referencing. And that gives you again, a very good look at a conference, for example, or a major world events to see what who is talking about things where things coming from.
And again, you can digest down a tremendous amount of data and boil it down to say, Okay, these are 10 or 15 or 20 people that we need to make sure our on our radar
if you change algorithms, and you use something like instead of going from the you know who’s most talked about to who is the the, the hub at the at the center of most of as many networks as possible who’s like the mayor, that person who everybody knows everybody you know is
that you can do that to identify different
Types of influences the type of influencers, like hey, you’re at Social Media Marketing World and you talk to this influx and they and they can get you a meeting with that person you really wanted to chat with a CEO of that company, whatever. So you use this type of algorithm. So So network graphics, a second application, it can be done. It’s it’s mostly we use a lot for social but it can be used for any type of interaction where there are discrete entities. So there have been applications using academic papers, who is the most public reference in publications, for example, in news events and things? So lots of different applications.
The third technique, tech technology is called clustering and clustering is it’s a form of unsupervised learning. And essentially, when you take two two numbers that are distinguished by data points, so for example, keywords in SEO, like you have a word you can look how often it’s search and how difficult it is to rank for that term, or what the cost per click is that term Okay, I’m tracking
Yep, you could put it on a, you put on like a piece of graph paper a plot, right? And start plotting where each of these these things are, you want to find out how these terms cluster together, which terms group up cluster naturally and figure out here, how do we identify this. So if I were to take a two by two plot of the number of searches that a term gets, and then the How difficult is to rank for the term I would want as a, as a marketer, I would want the terms that I rank for to get more and more difficult, right? I want them to be high volume, but I want I want to make it hard for competitors to rank for those terms. So I’m going to try and get as many links to those those keywords as possible. On the other hand, when I look at my competitors, I want identify the terms that are high volume but low difficulty, I want to take market share away from them and use this and so I would use for example, I was giving this example earlier today, I would use those terms to craft so my social media content and the content then that resonated well on Instagram or LinkedIn or whatever. Guess what, that’s
It’s ported to my website and now I can start taking away market share on the search side from people but using social media as a proving point like up that really is the popular thing. So clustering is very powerful technique and I talked about this right I think I won’t mention the client but it was a wine clients and my on the right track here. We talked after Social Media Marketing World and you guys you guys had showed me some powerful things. Um, yeah, towards some advertising, which didn’t come to fruition, but but I am. Is that is that what we’re talking about here? So that was a very primitive look at it. A lot has changed in the last year,
the techniques now are much more refined, much more actionable. So that’s a super important thing. So clustering. So third technique, I will dig into more of that at at the teleconference. The fourth one is driver analysis or understanding is supervised learning, it’s it’s understanding what leads to something else. So we talked about this early on in our conversation about supervised learning when given a set of numbers.
and a bunch of other numbers that may or may not lead to that what combination of variables or how the highest mathematical relationship to the outcome you care about. So imagine taking all of your social media data from, you know, sprinkler, a Sprout Social, or talk Walker, whoever you have, like 100 columns of a spreadsheet, you know, likes on Facebook, and wow, and
all the reactions and they have that from Twitter, LinkedIn, and Pinterest and YouTube and have all these columns of a spreadsheet. And then you have your web analytics time on page bounce rate, exit rate, number of visits, and so on. And then you have your marketing automation software, lead score day, you know, date number of touches things and they have your CRM data, you know, sales rep, assigned deal stage, all this stuff. Now imagine you have your spreadsheets about what 10,000 columns wide right?
If a closed one deal or the value of the closed deal is the target for driver analysis. A machine can go through and mix and match every column with
Every other column and understand after several million tries these five or 10 or 12 columns are the ones that matter the most. So it may be Facebook, post reach, plus email, touches, plus lead score leads to revenue.
And doing that type of analysis gets you at what are your actual KPIs. So one of the, my, my minor pet peeves in life is that people use the term KPI without having any idea what what it means its key performance indicator, which means that if if you’re looking at a number of that and and you’re not even going to get promoted or fired, it’s not a KPI. Right.
The only numbers that are KPIs, the ones that you get fired for,
right, well, we try to tell her place like where we became guys that we’re documenting our tracking our life tied to your business goals, like this social media activity, you must be tied to the business goal, which is the KPI. Yeah, but then a lot of people I like that analogy and so much it’s simpler as
You’re going to get a higher fired Exactly. Now, when you do this type of driver analysis that may surface either individually or more likely groups of metrics that function as a KPI like if this group of numbers together goes down, you’re in trouble. And by being able to monitor that all the way up the marketing operations funnel you can detect trouble sooner rather than later intercepted and hopefully fix the problem. You know, if you’re fixing stuff at the top of the funnel operationally or early in the customer journey, you can keep that ripple effect and spreading down to you know the your sales team stuff. So driver analysis is super important.
And the last example I’m the one that I think is probably the easiest for marketers to grasp right out of the box is time series forecasting, which is what’s likely to happen. So given we use a lot of search data for this because search data is really reliable, it’s very clean and doesn’t require access to the clients data which is super
Helpful given a whole bunch of search terms, what is what when over the next few weeks will a term spike. So the example that we give often is when two people search What are people doing when they search for the phrase out of office template or out of office out What are you looking for? They’re looking for a setting in the email classes they can go on vacation, right
which means that if you can forecast when bye week that term is going to be the most used you know, as a marketer don’t send email that week. Don’t blow your ad budget that week. Don’t do a major campaign launch that week. Nobody’s home nobody’s reading. So by using cancers forecast, you can say okay, I instead I’m going to time my launch campaigns for when people are searching the least for that phrase. Because that’s would be a logical time to do that.
Yeah, I love that. I mean, that’s a great idea. Wait, so conversely, really quick if I were if I were on vacation company if I was like VR, Bo, would I be looking for those times people
with my people are on vacation.
You want to be thinking when people searching for cheap flights to Orlando, or cheap flights to Vegas or whatever, because they’re there at the early out there. But you know the sort of the consideration and evaluation phase of the customer journey by the time so we’re looking for out of office, they’re already got it running in the planets. Okay this is again number three.
So one of the things that we’ve changed recently is digesting down this data into one page forecasts that are four weeks long and says that use our four P’s content Guide, which is plan
prep, publish, and promote the fourth stage journey for content that’s four weeks long. And what we’ve done is we’ve we’ve isolated sort of the top three or four terms that are going to occur each week for the next four weeks. So you can start now starting to plan what your content and four weeks is going to be. You write the content then you know the contents that
Three weeks out, you start actually draft it and build it, and so on and so forth. stuff that’s two weeks out, you publish it, give Google and the other search engines the chance to index it. And then the week of you promote it, you put it on social you run ads on and things like that. And by having this repeated one page forecast that gets emailed to you every single week, like, Oh, now I don’t have to guess what try be blogging about right, I know have to and your editorial team and your and your management team can all be literally on the same page. Okay, this is our plan over the next four weeks. And it’s just like a treadmill. You know, the next week forecast comes out, stuff moves down, there’s new stuff to start planning and you can your content marketing can be a lot less stressful and it’s timed with what people are searching for. So you’re in the market when people are thinking about what it is that you’re you’re publishing. So time series forecasting is super valuable and very, very powerful. I love it. I my wheels are already spinning Carrie Did you wanna, I feel like you’ve been like, just so good.
I just been like that, that give and I don’t know if it’s the, the given the heart eyes or it’s the the
the giant number five give me more input or it’s the feed me see more gifts I don’t know which one of those I am right now but I was literally watching the whole thing like this I was so glad to be on camera because it was just it’s just too exciting
any any like questions that popped up before we yeah
and it’s kind of my my favorite sort of weird conversation to have with clients because they get to swear in it but my my my big analytics question is pulling out sentiment because there is such a difference between like, if I’m reading a tweet, I know it’s what this is shit and this is the shit I know the difference because I’m a human and I can read that it’s getting machines to read that that’s what I want to see. Like that’s that’s the record I want to pull in the legitimate sentiment.
Want to be able to say people are saying this about you? And here’s why. And so
so the reason why a lot of sentiment analysis automated is terrible is because the vendors have many social media marketing tools. Not all, but many of them are using a very simplistic approach that it’s called a bag of words approach, meaning like, Hey, here’s a bundle of words. Now this word appears and it must be bad, right? There’s a very famous senate library called eighth and that is
basically that it’s a series of 3000 words that were tagged you know, minus five to positive five and much of the the correct criticism about the bag of words approaches that it is context independent. So to your point, you know, shit versus the shit it are two different concepts that share the same stem word and as a result, they’ll be treated identically in a bag of words approach what’s happening now in machine learning, and particularly in deep learning is that we’re introducing more and
Better models for more advanced sentiment analysis to be able to see not just a word but a word in relationship to 12345 words around it. So these larger scales and scope things so that machines can understand this is a these are how these terms of semantic related technical vector ization.
Deep Learning is is advancing a lot of this especially at very large volumes, the one area which is still
really eluding machines, because there’s no tone there’s, there’s no mathematical representation of tone is sarcasm machines still cannot effectively do sarcasm they cannot tell the difference between I love this and Oh, I love this right. We I think we as humans, so sometimes depending on the person right and how how well we know them. It’s hard for us to detect sarcasm as well right. Exactly in an email for me You are no no it’s like a 75% chance what I said was sarcastic but it
I’ve never met you before. And I, I just typed that out and you can hear the inflection in my voice you might not know exactly. So the reason why you’re running some trouble is because companies are using very old techniques. And the reason they’re using those techniques is because they are computationally very cheap. Vector ization and deep learning are very computation expensive operations they take lots of big hardware a long time to run comparatively and I mean a long time in that
it took it may take a deep learning model two or three minutes to process a text Well guess what, when you everyone in the world is used to tapping a button on their phone and gain instant result and so two minutes is not acceptable but the trade off is we get terrible data instead so the the solution to that is
well the solution is that is either you find a vendor that uses the more advanced technology within understand this probably batch
Rather than real time or if you have the aptitude the attitude and the technical capability build build your own software on top of a social monitoring tool to do the processing yourself that’s what we do we use a company called talk Walker which is still there a great company and we export their data we we very rarely use much of the tool itself we use them as a data source and then we build run our own machine learning software that we’ve customized on top of it to do topic modeling and influencer mapping stuff because we like we like to be I personally especially like to be able to control what the algorithm is doing yeah what I’m hearing you say Chris’s call Chris he helped
so I know we have to wrap it up we run a time and you know if you haven’t already figured out why you need to come see Chris Hello conference Chris in like you’re really short synopsis Why should people come to the Hello conference and at all but also to see you
You can rub elbows with someone of your caliber. It’s going to be an interesting contrast. So Mark Schaefer and I have been friends for a really long time, and I love him. He’s a wonderful human being. And he is going to present on sort of, you know, making marketing more human, the marketing rebellion, his new book, and I’m going to present the opposite approach of resistance. few times, you will be assimilated
the Borg approach, but
the the two approaches use different methods to get to the same thing we all want. Marketing is better, faster and cheaper. The question is, what’s the machine going to handle? What’s the human going to handle? I think the two of us together will have offer complimentary occasionally contrasting approaches towards these are what you know, these are the things that you need to double down on as a human and then these are the things that you’d better have a machine on your side handling because if you don’t, you’re gonna be out of a job, huh? Yes. And, you know, for business owners like
yourself and myself scaling, right? Aren’t we always talking about scaling? And this is how I think we answer that question of how to scale is using machines to help get there. So anyhow, I know we’re out of time. But Chris, thank you so much. You know you and we just I’m just if people don’t watch this and figure out why they should come and see you in person I, you know, then then fine.
So thanks again for for being here. We’ll tweet out all the links and where to find Chris where to find trust insights. All that will be down below drink.
Just check it out down there and just reach out to us if you need to get ahold of Christopher will put you guys in touch. Thank you for having me. You bet. Thanks, guys. Bye everybody.
Thank you for listening to me your insights, the trustee insights podcast, please ask a co worker or colleague to follow our show on Google podcasts Apple podcast.
Yes, wherever you listen to your shows got a question like us to answer Watson help solving your data and analytics challenges visit us at www dot trust insights.ai today
Get unique data, analysis, and perspectives on analytics, insights, machine learning, marketing, and AI in the weekly Trust Insights newsletter, Data in the Headlights. Subscribe now for free; new issues every Wednesday!
Want to learn more about data, analytics, and insights? Subscribe to In-Ear Insights, the Trust Insights podcast, with new 10-minute or less episodes every week.