Today, we’re pleased to feature, open to all, our opening keynote address at the NEDMA 2019 Conference. In this talk, you’ll see 5 practical applications of AI in use today, plus a never-before-seen glimpse at the use of natural language generation to create readable content at scale.
Can’t see anything? Watch it on YouTube here.
- Need help with your company’s data and analytics? Let us know!
- Join our free Slack group for marketers interested in analytics!
Note: the discount code shown in the video was only for the day of the event and does not work any longer.
What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for watching the video.
Thank you very much.
Logistically just for later if you’d like to copy the book, you have to go to this URL and I are marketers. polka.com is just kind of come by and then just bring your receipt over that way. We’re not hanging out looking at staring at each other swiping credit cards. All right, let’s talk about artificial intelligence.
We as marketers want three things. We want better, faster, cheaper, everybody wants better, faster, cheaper. How are we doing as marketers with better faster and cheaper on the topic of better marketing? Is it a tough place right now? This year? In 2019. We as a civilization will create 40 zettabytes data that’s unimaginably large number
give you a sense of how our setup is how many people watch Netflix?
If you were to start watching Netflix, without stopping to eat, to drink, to sleep, any of those things and you started 55 million years ago, you would just get to one zettabytes of data usage now, and we’re going to create 40 of these this year. marketers are not great at keeping up with this much data. And especially for those of you who work in direct mail, you know, you know, this version very well. Test, test, right? That address or bad information. as marketers, we’re not doing a great job with data quality, so marketing, not really getting better. Part of the reason why is because of the use of data or the lack thereof. And the most recent cmo survey which by the way, if you have not downloaded
is a gluten free not even a form to fill out? At cmo survey done the work is released twice a year. And it’s what’s on the minds of some marketing officers. In the most recent cmo survey, marketing analytics use for making decisions is at an all time high of 43%. Sure, what are you doing the other 56% of the time, just guessing.
And yet, when you look at what chief marketing officers care about, overwhelmingly, they said we want to prove the value of marketing, two thirds of the parking officers need to be able to prove them. So we’re not doing better as marketers,
or at least getting faster. Well, the world’s getting faster in 2019 at the variety of the speed at which data is happening, as faster than ever in 2018 there were 266,008
hours of Netflix. Within 60 seconds on the internet in 2019 that’s 694,000 I was in Netflix says a lot of Netflix and chill.
174,000 schools on Instagram last year 340,000 schools this year, Snapchat went down a little bit hundred 87,000 million email sounds last year. Hundred 80 million emails sent this year this was within 60 seconds 60 seconds on the internet. 1 million TechCrunch swipes last year 1.4 million this year. 60 seconds.
If you are looking just at something as simple as news this year, we are on track for about 122 million news stories 172 news stories from it. If your company was on the front page, your comment on the wall street journal.
You got anything seconds, right because something else has come to this place. So much
Marketing is not keeping up with the world world is getting much faster we are not. Finally, are we doing anything to get cheaper?
When you ask CMOS how you quantify the impact your marketing only about 36% can say they can prove the impact of market. 50% just kind of Yes. And 30% said, Aquavit are not even going to drop. So two thirds of marketers can’t prove, quantitatively the impact of that marketing. And we see this in the amount of money that we’re spending, we’re spending more on brand we’re spending more on CRM, we’re spending more on product we’re spending more on service. Marketing is not getting cheaper, if our marketing is getting more expensive. The combination of not being better, not being faster and not being cheaper, leads to an unpleasant outcome. Right, it leads to us getting our butts kicked. The solution or a solution to this on surprisingly, which is why you’re here is artificial intelligence which
promises three things better, faster, cheaper. Artificial Intelligence is absolutely faster because it’s computers. It’s at the end of the day, it’s all math. So it’s computers that can do things much faster than we can. Anyone who’s ever made a mistake, copying and pasting something knows that humans are much better at the accuracy than we are. And there are so many tasks that we do as marketers, cleaning lists, for example, that machines are just so much better at doing that we are so automation making things any less expensive by taking off boring tasks. So what is this stuff? You’ve heard a lot about it seen in the news. People are afraid of terminators and things like that. Artificial Intelligence is about getting machines to think like humans. If you look at how a human develops, we begin with basic sensory things, right? Our senses, basic elements touch the surface on our own, but again, we evolve to things like language, being able to communicate and then
higher cognitive function, being able to think and reason, by the way is apparently a teenager appreciated. Harvard says cognitive decline.
Machines are no different. We begin with things like algorithms, machine learning, deep learning, and they get to a whole general purpose or an artificial general intelligence, which what a machine becomes sentient. And think for itself. That’s one place all depending on who you ask, but from 550 years back, but it’s not, it’s not something we need to worry about.
All artificial intelligence begins with statistics and probability at the end of the day is, is this picture of a hot dog or is it not? Is this
is up we have several orders the set are not from stats. machines have algorithm. An algorithm is something you use every single day in market. If you’ve done an ad test, you use an algorithm. Use Albertans in your daily life, it’s just a set of repeatable processes with a guaranteed outcome.
would bet you small Asian, you tell us that
then you probably put on the same journal article clothing, everyone write something from the bottom up versus the top one verse
for their socks on first, but whatever the case is, you probably are consistent. That’s an algorithm.
Now where things change is when we get into what’s called machine learning. First seek artificial intelligence. Traditional, used to write software, write the code right from the album, and then the machine would spit out a result. Everything from your word processor to mailing label printer to your video games does exactly that. Machine learning reverse. We now feed data to machines, and they write their own software. That’s how machine learning works. And there’s two broad categories. Let’s say we had a table full of kids blocks right
The first type of machine learning is called supervised learning. We want to understand we want to teach the machine to recognize something. So we show the machine the color red over and over and over again, until it recognizes the color red, and then everything after that, they can find it all around. supervised learning as most famously known for, in this case from IBM. Back in
ibm oncology was working with the university or University of Tokyo. There was a woman who had leukemia, she was not getting better, she was not responding to three. So they fed her genome to lots and they fed 233,000 oncology journals for Watson and set figure it out. What’s wrong.
Watson, sequence the genome, read the journals and spit back you’re treating the wrong high to leukemia supervised learning history treating this kind and said they changed the treatment you may remember and I
Secret versus life. Now what’s not? What’s cool about that is a little she did that, but didn’t 11 minutes.
Okay, that’s how fast stuff works supervised learning, you will see this most often in the direct marketing. Well, it’s simple things like using AI to clean list, right? You know what proper mailing it sounds like, you teach machine that and it will clean faster and better than you can then you possibly could. The second category is called unsupervised learning. Instead of just identifying the blocks by by the call to read, what have you been invited by all colors, or by their shapes and sizes. When you had a pile of data to machine and say, do some unsupervised learning, I will try to categorize all different ways to slice and dice that data.
A simple example of this is I was getting ready for client meeting and or 2600 articles that we needed to digest how to tell the client he was
What’s what was in the news? Rob, you
wrote me suffer. If I just did it down one track was programming the individual topics. Here’s what all these articles are generally about presented to a client. Now unsupervised learning can be used in direct marketing, just to classify wonder, who are the people who are on your list? For example, what are all the different zip codes where all the different areas whether genders, the people who are on your list, being able to classify and categorize audiences, and then be able to expand that and do things like our phone analysis to figure out who your most valuable members.
The last category is called Deep Learning, which is when you combine many layers and machine learning together into deeper things like pancakes where each layer is an algorithm, which you learn and the data, not perfect announcements, like the syrup, like
deep learning lets us create machines that are faster, better and cheaper than any human machines can read. Live Seth Godin.
Get Cortana rockers an example of deep learning.
So this is the, what AI is and the reason we need to know this is not because we expect to become an artificial intelligence engineer or a data scientist, we expect you to know this because you need a detector of smash friends called experimental the total when you hear vendors talking about how we all have AI in their products, being able to ask better questions to those vendors to say, okay, explain what kind of machine learning is in your pocket. Because in the most recent In a article in The Financial Times recently, they said 35% of vendors insanely I don’t have a limit. So being able to ask these questions that your maps now what does this look like when you go to the state of the art?
this past February, IBM had a
debate with national project debater between a human debate champion onstage and machine
and they will be
Given the topic 15 minutes in advance, should we subsidize and have our machine had to go out and use research the human had us research, they both presented opening statements. They both listen to each other’s arguments. They both presented remodels and summer closing articles. Now
in theory on the human one, because the human not more votes on the audience, but the fact that the machines listened, argued, and came back with intelligible responses the entire time was groundbreaking. Imagine you will have customer service representative for example, and you truly listen to the customer, talk to them and provide them good answers. That’s the state of the art today.
Now, there are some things as hapa
generally speaking, AI is not good at acting. machines can understand sentiment and emotion and text, but it can’t act on empathy is being able to understand somebody built in the tape actually, she’s
but she has a very bad complex job and
Go outside the walls for example
many business for a while your friend or that Hollywood that person you like that, you know what we’ll just copy a break on this thing and what we’re going to do an extra that’s human judgment that overrides the rules machines by definition cannot override the rules they made those
machines are very bad a general life experience they can cross domains, they can solve problems very narrow, not taking half a picture.
And for the most part, humans, machines cannot substitute.
I say for the most part because there are some places and some companies where you prefer for example, the registry of motor vehicles. I would gladly go from a few any day away.
So let’s look at some practical examples. How this How does this stuff which sounds like science fiction how’s being used in reality today, solving five kinds of problems, untapped data unknown influence
There’s unclear actions to take.
unfocused, marketing and unprepared for planning, marketing planning.
Let’s take a look at untapped data. First, you are sitting on a tremendous, tremendous amount of data, you’re not moving from who’s on your mailing list to what they create on social media to your CRM. We didn’t example, recently for International Women’s Day, there were 1.6 million social media posts published on that day about that topic. That is the equivalent of $8,960, you cannot read that much.
But a machine can. So we’re able to digest down to one chart what all the main topics for that day and then provide that as guidance to social media managers and players and say if you want to participate in this conversation, these are the things that you should talk about. And by the way, there’s some things here you should not talk about unless you have to be female. But being able to digest down that kind of insight is tremendous.
Industry valuable if you want to be able to act within the moment machines can help you do this. And another example, we looked at, we did a project for a recruiting company looking at all the listings that they put up on their in the job boards, and all the words they put in their job. And then we took 17,000 calls in their call center, how does she transcribed them all?
The words that people talked about on the phone with their recruiters.
There’s almost no overlap. And so this company is wondering why we have such a hard time recruiting well, because in your job as is nothing new there that people actually want to talk to you about. People care these these truck drivers care about things like region and job type and will they be home for Christmas and none of that is in the job. So being able to take that untapped data, the all those calls in the call center and turn them into useful actionable information is the value.
The second type is being able to identify you
Pulitzers people who could help your business by representing an event.
This is an example from Adobe summit, which is a major mega mega tech conference and about a month ago.
120,000 social posts that occurred on the first day. If you want it to talk to people who everyone else talked about, using machine learning and a technique called network graphing, you can take all that conversation and find the most influential people. Some of them are speakers, but a whole bunch are folks, ordinary folks that you could talk to, to get a hold of and presents your products and services. And that comes in handy form a spreadsheet that you can use to say, Here’s who’s worth talking about this show. Now imagine being able to do that at a major conference or major event. If you’re in b2c, imagine to do something like Coachella and figure out who are the B or C lyst. Folks that you might want that you’re not going to be as the A listers, but you might just be a seamless
Folks, you get that represented around.
The third category is being able to clear up what’s in your data. We have a lot of data. You have a lot of data right now, but a lot of it is very difficult to manage because there’s Frank are so large.
And an example recently for an insurance company, we looked at their search engine optimization, the keywords that this company rank for, they had
80,000 keywords, because after the SEO manager trying to figure out what words the phrases should we focus on, that’s a very daunting task to redo all 80,000 and figured out what to focus on. But if you would have put it into a machine and having some technical cluster, you can find the things that the pages on someone’s website that they rank for past and help them shore up their listings show up for their their search engine optimization. Conversely, you can then use the exact same technique to figure out what their competitors rank for. They don’t
go after those those topics and create content around those topics.
So by the way, this technique is incredibly valuable for direct marketing because if you use this information about what page is the most valuable on someone’s website, guess what that can inform the paper collateral, the real world prints the bus routes and things based on what you know people care about. So you can take digital data and bring it into the real world.
For example, understanding what drives performance.
We have more data than ever, and we have a lot of it. If you were to take every metric that you have in marketing, from locations of billboards, number pieces of direct mail sent out search keywords social media posts, email center, point your marketing automation system on your email list all the way from your CRM to Hey pay us money.
Apologize, Reggie, you as human would find it very, very difficult figure out which comes
combination of those variables leads to revenue leads to business technique, operating procedures and machine learning. You can do that the machines can mix and match and understand what is the most impactful things that you can be doing. This is an example. We had a surprising for an automotive. We wanted to understand what things got them feet in showrooms. Oddly enough, it was actually the size of their social media postings, we did not expect that we expected to be something else.
But for them, it was about getting to audiences. Right? If you want to prove the value of direct mail, for example, there’s no better way to add all your data into all of your other marketing data, run it through this have a process and show. Here’s how direct mail contributes to the business outcomes people care about. And you can do this across multiple different types of technology. We looked at Instagram stuff, we looked at b2b content, being able to understand that
What drives the outcomes we care about is so important.
And finally, parent preparing in advance for me marketers are so reactive because it’s like juggling playing chainsaws mostly is the office.
How do you how do you get out of that says the answer is blue. Hi, Lucy and Nicole predictive analytics using the same keywords
and a technique called all the rest of integrated moving averages.
stats, you can take the amount of time someone to search for something in the past and forecast is over to predict for when somebody’s going to do something. So for customer viewers, for example, that example sold life insurance. You could show them in the next couple of days or weeks or months when certain terms would be more searched for indicated interested by the audience. You want to help out somebody sent it out send out the most impactful malware for example. Thank you
When the audience is going to be searching for and make sure that it’s in their mailbox that week, when they’re thinking about predictive analytics, and it’s not something that is actually this complex, you could literally turn it into a sheet that you would enhance your marketing, or, hey, this is going to happen the next four weeks, get ready, let’s plan ahead.
We’ve been doing this for ourselves with our podcasts, trying to make sure that we’re tying the topics we talked about, with what the audience is interested in. And we’ve seen faster than average, we’re about 44% faster growth than before we were doing that.
Predicting and understanding what people are interested in for most of our businesses is a pretty easy thing to do. Because most of our businesses are fairly cyclical in nature, especially b2b. b2b is incredibly cyclical.
So those are five examples of using machine learning technology today in market. This is not fictional stuff. This is not navel gazing. These are real things done and
The technology is so powerful, and so mature that you can run on a laptop, you don’t need to have a billion dollar Supercomputer Center. Most of these techniques run on today’s laptops. The important thing is to start now,
remember that machine learning is based on thing. So if you’re not collecting the data, you’re not cleaning the data not getting data ready. You’re at a disadvantage very food pattern has. Conversely, if you get started as a competitor weights, not only does it have to catch up in general, but they have a data deficit because you’ve got rumblings How do you get started?
Getting started is a seven step process.
The first step and the hardest step by far is that strong data foundation. Does your company know where to say it is? Is it clean? Is it well prepared is optimized for machine learning, artificial intelligence.
This will take at
90% of your time when you get started, because it is such a difficult thing to do. We worked with one company last year that
their sales department prohibited their marketing department from seeing sales data. Like how does that work? Not very well, by the way.
It turns out sales was not selling nearly as much as they said they went.
From there, you must become measurement oriented, become a data driven culture be able to set meaningful key KPIs key performance indicators. My favorite definition of a KPI is this is a number with which you will either get promoted or fired.
If your numbers that you’re working with and not going to get promoted, or five, not APS, but building that data driven culture is the second step in this journey, the third step is to be able to do qualitative research, which actually the direct model has done very, very well and over the years and let me
For this is that most of our data is really good at telling us what happened. Very little, our data is good at telling us why something happened. You have to ask somebody, why did you make that decision? Why did you purchase this brand? Why do you not go to this store even though it’s slightly closer to your house, building the qualitative research capabilities is important.
The fourth step is process automation.
We are busy as marketers, so we have to figure out ways to automate as much as possible free up our time. So we can focus on things like data science, they
are really simple example.
I used to do social media scheduling for my company. And it took a while to put together you know, 100 and some odd posts, leave it at that and open up the software and built software to do that for me. So now instead of taking three hours a week, it takes three minutes a week and we’re gonna move on from more guy to work. So finding ways to automate every process. You can
Practically automate is really important.
Once you become data driven, you find the data cannot explain your data you freed up your time. That’s when you start getting into the actual machine. With data scientists people can understand your data and build capabilities for analyzing experimented with data, implementing machine learning, and eventually becoming an AI first organization. There are very few of these little feel I only have so far behind. There’s like 10 companies that truly are AI first and you can name most of them facebook, facebook, and Google and Apple and so on and so forth.
So this is the journey you need to take in order to become an average any one of these sessions bringing you returns but taking as many of them as possible will change the way you do business.
Now there are a lot of vendors out there should you buy these capabilities are built them to interesting question.
The same question time and money, time, no money. We built it into that money, no time What? But it turns out there’s a bigger thing to think about here. And it’s a strategy. This is concept that people love to talk about called digital transformation, which promises puppies and unicorns, rainbows.
And what’s funny about digital transformation is once you clear we all have there is some legitimacy to it.
When you look at how a company is governing its marketing technology properly, it looks kind of like this. Let me highlight this row here, the thing that is worth paying attention to is this box.
Part of your marketing is creating value for your company, but in a digitally transform how many? So the data analyst the models, the machine that we built could be valuable as its own product. For example, let’s say you were popping up and you built a machine learning model, that effect
Chris, how well certain type of coffee tree grows, you can pick up that
coffee tree growing model and apply it to soybeans, to rice to stupid crops and video games, right? You can even do any of those things. And that model itself has value above and beyond the fact that your home is copied on your direct mail house and you build a model that is predictive, based on the data you have of which customers are going to respond to a piece of collateral. You can pick that up and sell that itself.
So the question of buyer build is not so much time and money, it’s really using artificial intelligence to change the core promise you to comment is the answer is yes, you have to go live because that’s the secret sauce. And the answer is now by save yourself some time.
Some examples of just buying them off the shelf. They’re real simple one that I love a lot is a tech
Oracle honor, honor takes audio files, like those holes, or even like this schema we’re doing right now. We’re just recording the audio from it and transcribes it.
How much do you talk every day? How’s your customers talk physically talk to you. Being able to unlock that data freedom is incredibly valuable. This cost like 80 bucks a year.
For those of you who are hybrids and working digital, Google Analytics, adding a GSK to his top work, your gnosis box in the upper right hand corner called insights. In Google Analytics, click on the box, it will tell you, hey, our AI was looking at your data, we found some things you might want pay attention to, it’s doing the work for you. You can even ask it. Ask Google questions about your data. So in a lot of cases, your vendors will be adding this stuff into your technology already.
If you want to build it, if you want to have a core competency
approved by AI. couple different things to look at. One would be IBM Watson studio, which is a very nice drag and drop environment to build AI without having to write code. And if you or somebody at your company wants to write code, the two languages that are the most popular are an Irish little bar, another language called Python.
So how you prepared your company for the stuff, you’re going to need three kinds of people.
There’s an expression data is the new wild 2006. I like this expression a lot. Because if you ever work with criminals, nasty discussing stuff doesn’t do much. It’s like saying I think it’s actually you need to extract it or find it and bring it to market some kind of usable product.
This is the same for your data. Need developers to extract your data from we’re all places that lives in your company, for your vendors, your third parties from your customers event and bringing into one place.
We need data scientists, people who can help you take that data and look at all the traffic and turn it into those models into those those pieces of software and you can use to draw useful conclusions. And then you need marketing technologists, people who can take those insights and models and apply them to the business.
If you’re a smaller company, you start marketing technologies work backwards, you’re bigger company, you started it all for some workloads. The reason was the marketing technologists did their good and help you incrementally move things like revenue to afford to hire better, folks.
Now after the big question, what about you?
Who’s gonna lose their jobs die? I want to change or eliminate 70 to 75% thousand seniors. The easiest way to tell whether your jobs at best is to open up your calendar and look at the first row the second Monday in July.
Know down to the minute what’s going to happen that day at work you Jonathan nation, because it is so routine so rope so free program that what you do can be automated. On the other hand, if you open up your calendar for this coming Monday and you have no idea what’s going to happen, but you know it’s gonna be a service jobs probably okay
to further insulate yourself
focus on multidisciplinary sales. Remember the machines are not good at Gen X rates. If you look at the top 10 most in demand skills on LinkedIn.
Any one of these categories has things tasks in it, that can be automated. audio production has things that can be automated, analytical reasoning, artificial intelligence itself, but if you are good at multiple things, if you’re good at audio and you’re good at sales leadership and you’re good at people management, you are very difficult to replace
extremely different lyst good people management and good at cloud computing, add your good UX, you are extremely difficult to replace, because you’re bringing in insights and things that work from multiple disciplines into your work. That’s tough program.
Second thing is learning how to think like a machine. So algorithmic thinking, it says we need to write code. But one of the things that distinguishes a great Modern Marketing practitioner, from the old days is to look at a problem and say, How can I programmatically or systematically solve this problem? So it never happens again, as opposed to how can I solve this problem today? And oh, by the way, next Wednesday, but I have exactly the same problem. Right? Being able to think like a machine is so important.
Being able to oversee the machines is going to be a profession unto itself in 2016.
In Florida, which is where all real things happen.
Somebody a police officer tried to write an algorithm to predict recidivism, the likelihood that
Someone would commit another crime. The outcome was 20%. Correct, which is far less than the focal point. But it predicted African Americans would we offend five times more than they actually put their thumb on the scale said, I want this machine to spit out this result.
Last year, Amazon got a tremendous amount of
because they wrote an album to try and predict which candidates to hire. They trained the algorithm incorrectly no stop hiring women.
This ability to oversee machines to understand the audience and say, Hey, machine that’s not okay. is a vital part of our work in the next five years.
We have to be outcome focused as marketers worry less about process and more about the outcomes, the questions you need to ask to get to the outcomes you want.
Three years ago, I was trying to learn this technique called keras. framework, and it got to about first column on here and what the idea was thing conference and they said Why are you doing that is a drag and drop invite
That will do it all for you. You don’t promote it was like well,
this year I went to the bank and they said, Oh, we don’t do that anymore now, but she just does for you is the Lord your data and we’ll figure it out. And then you judge whether the outcome is what you want or not. So instead of worry about the coding, we focus on the outcomes. Your job is to become the chief questions Officer of your company to ask what outcome we after, Is this fair? Is this reasonable? Is this doesn’t will this improve the business?
To give you a sense of where things are going? How many of you somewhat regularly use social media? Okay, pretty much everybody.
And natural language generation is going to be the next thing that will drastically change.
This is something I was writing last night in my basement because that’s what I do for fun.
All the social posts from from them and some of the folks here from
The most overdrive from a couple of publications. I have this deep learning model, start writing new social posts and it came up with some really good ones in about 15 minutes time I wrote 20,000 tweets.
That’s something that was silly. Some of them are not RMB I mean, some of these are really, really good.
That’s if some goofball in his basement they do this in an hour at zero dollar costs.
This is what’s going to change all of your marketing in the future, they’re going to be two kinds of jobs. How many people are locals is regions in a month? Okay. This is the marketing robot driving around. Marty didn’t used to exist. These be a kid was cleans up last night up and down the aisles. And they brought in Margo to detect spills that kids Job Change. Now that kid is being managed by machine. Marty goes up and down the aisles and says hey, there’s a still an aisle six. Come over and clean up by the way he laid out in front of Barney, thanks. You’re still
This is the future. There’s two kinds of jobs, you human eyes machines. The machines got to manage you. That’s it. So the question for you is who you want to be? Do you want to be the person manage by the machines do you want to be?
Thank you very much.
What helps solving your company’s data analytics and digital marketing problems? Visit trust insights.ai today and let us know how we can help you
Need help with your marketing data and analytics?
You might also enjoy:
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.