Our presentation from the NEXT10X Conference 2018 is below, in both video and PDF format. (click to download, 4.4 MB)
Can’t see anything? Watch it on YouTube here.
Just the slides:
Interested in a predictive forecast of your own? Order one now:
Want to chat about your existing analytics issues? Grab a Table for Four:
What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for watching the video.
Thank you should will have a lot of charts.
Should you want to follow along, or just want to keep this as a commemorative keepsake.
If you go to where can I get the slides dot com that will let you get the slides marketing Modern Marketing as animals saying has four fundamental problems today.
These four problems are called the four V’s.
This is an IBM construct this is not original.
The first problem we have as marketers is a volume problem and not just being loud but the amount of data, we are producing as marketers and as society as a society this year.
In 2018 we will produce about 30 zettabytes of data.
A zettabytes for those of you who are not familiar is a really, really, really big chunk of data.
The easiest way to explain it is a magic.
a gigabyte a gigabyte is about one Netflix movie right so you watch a movie on Netflix, if you were to illegally download it from the service, it would be about a gigabyte file.
Please don’t illegally downloaded from the service if a gigabyte is is that one movie if if that was a brick like it would turn that into a physical brick, we will be building 14 Great Wall of China a day there are 3.
8 billion bricks in the Great Wall of China.
So that’s how much data we create every single day.
If one gigabyte was a brick 14 Great Wall of China.
That is a massive amount of data.
We and as humans.
There’s no way we can process.
We can’t even remember what the last 20 tweets.
We saw were right, much less deal with that much data.
The second problem we have is a problem of variety, the amount of the different kinds of data we have that, frankly, we’re ill equipped to deal with.
Here’s a fun 2017 2018 perspective on just how much data there is in the world right now.
Last year, 70,000.
hours of Netflix watched per minute.
This year, 260 6000 hours maybe they all did the do your partner Netflix and chilling away 990,000 Tinder swipes per minute.
Last year, 1.
1 million this year, they don’t tell you whether the left or right.
I’m guessing mostly left thousands and thousands.
This is every 60 seconds on the internet as all these different content types we as marketers are really bad at anything that isn’t text, you’re at a conference sponsored by a company Stone Temple that does SEO right and so many people who are doing SEO are stuck in text only there’s video there’s interactive there’s all these different formats that we can’t process.
There’s the audio that you’re hearing our third problem is a velocity problem the speed at which data is coming at us is faster than ever.
If you work in, for example.
Public Relations or earned media or trying to get inbound links for your SEO program is how much competition.
Do I have there are about 300,000 news stories per day right now.
So even if you got that great you know I can’t say page rank cuz nobody uses page rank anymore but you got a great front page placement in the Wall Street Journal or the New York Times or whatever the case may be.
You are one of 300,000 stories that day.
You are less than a blink of an eye, even if it’s a fantastic placement.
This year we’re on track for 100 million news stories.
If you are born if you just graduated from college is here you at the year of your birth had about 3.
6 million new stories in in the happen that you’re now at 100.
million and not all of them are politics and finally we have a veracity problem with our data we have a reliability problem, our ability to audition.
How accurate is our data, how truthful is our data.
And not just fake news but all of the data that we’re dealing with.
If you were to go into Google Maps.
A number of years ago and ask for directions from Topeka, Kansas to Tokyo would actually have told you to kayak across the Pacific Ocean.
They have since fix this.
And it round recommends a flight, which is a much more sensible thing.
But if you were relying on this data to get you from point A to point B, it would have sent you into the ocean with a small boat right now think about that from a marketing perspective how much data are you relying on and do you even know whether the system is giving you correct directions or not, you can’t you can’t know that anymore.
So this is how Modern Marketing is broken.
We have these four problems that none of us are able to keep up with the prescription is not more cowbell but it is cognitive marketing and cognitive marketing is marketing using artificial intelligence and machine learning, it’s marketing that learns.
systems that learn at a scale far greater than we humans can do right is using AI and ml before we dig go further, let’s talk about what these terms mean what is artificial intelligence, it is getting machines to behave like humans, but using all the power and capability of these machines.
Think about human child, but this is what happens when a human child grows up in that first year of life yet be sensory inputs right they learn to be able to see to hear to bring in data and understandings and start to process it, then they develop language first sort of just lots of sounds and eventually tuning into into words that are understandable and then higher cognitive functions.
I appreciate the fact that cognitive function drops off when they become teenagers.
This is how systems, learn, we do this automatically machines are getting to this point now where we have algorithms machine learning deep learning and eventually what people.
Ray Kurzweil called the singularity when we achieve a machine that become sentient and self aware.
Let’s start with algorithms, you use algorithms all day, every day, not only as a marketer, but in your life as a human being.
You probably put your clothes on the same general way each day.
You may put your pants on first for me for sure on first.
You may do the same things in the morning.
It’s an algorithm.
It’s a way of doing a set series of processes, you’re doing these all day and they’re okay they’re the building blocks of AI then you get to what’s called machine learning in traditional software.
In the old days, we would write the software and the machine would spit out the data machine learning flip that on its head machine learning takes the data and the software learns from an right its own code and then spits out a result.
So as an example, if you had a bunch of blocks on the table.
There are a bunch of different ways to make sense of this right there’s what’s called.
Will you tell the machine.
Hey, I want you to find the color red and you keep showing a red color block over and over and over again and eventually she learns that’s red and now when you show any color block it will tell you yes this is Red or No, this is not read that’s called supervised learning in machine learning.
This is how you find needles in haystacks one of the most common examples of this is IBM Watson in their oncologist software, there was a woman in Tokyo number years ago this was a CBS 60 minutes story, who was diagnosed with leukemia and she kept getting worse and worse.
The doctors are trying all these treatments and nothing was affecting the disease.
So what IBM did was in concert with the University of Tokyo took that woman’s genome, put it in a Watson and took 233,000 oncology journals about leukemia into it and put it into Watson Watson’s re sequenced the genome compared it to these what was in the journals and found that they were dying.
They had misdiagnosed or they found the.
Wrong kind of leukemia.
They’re trying to treat the wrong thing.
And so they changed the treatment and she made a full recovery.
If you may went into remission.
Now the amazing part isn’t that we found a way to fix the situation because a human could have done that the amazing part is Watson did all of it and 11 minutes.
That’s how fast machine learning works.
That’s how fast supervised learning can work.
I think about how that would apply to your marketing.
What if you could have an answer.
Instead of as an was saying, with no talking to human beings.
What if you could find the answer is so much faster.
The second category of machine learning is called unsupervised learning and this is where you say, Okay, here’s a pile blocks, what are the different ways we can categorize this pile well their blocks or read, there’s all these different colors we can sort these blocks by there’s different shapes.
There’s different sizes, things like that.
And what you end up with is a way to classify all this data to the person was asking how do we dig into what our customers are saying you would feed all the language all the transcripts all the conversations you had with all of your customers and have the machines, go to.
digging through it to figure out what was in bed what customers kept saying over and over again.
I was doing a project.
Recently we turned 71,000 paragraphs of customer data and 2621 articles into a chart and five minutes now instead of having to read through all that stuff that the humans, the customer said the machine summarized it for me so I could make decisions, much, much faster machine learning for good or ill is mostly math and statistics.
How many of you are in the marketing professional resume the entire room.
Is anyone who’s not in marketing okay you’re in the wrong conference.
I’m just getting.
How many of you got into marketing because you love math and statistics okay for the people who did not raise your hands I have bad news for you.
Thanks to the way technology is converging with marketing, marketing and mathematics are now almost synonymous machine learning, especially these are just a couple of the algorithms that you can learn in machine learning clustering decision trees Kamins all this just math their fancy.
terms for different ways to do math and statistics to assign probabilities.
Now what happens after machine learning is where things get really interesting with what is called Deep Learning and deep learning a so called because it’s essentially taking lots of machine learning and making it like big big big stacks of blocks tied together chaining things together where data flows from one block to the next and each layer constantly or finds it, you may have dozens or hundreds of thousands of layers in the deep learning system.
The outcome is better than any single machine learning technique and better than what a human can do they create machines that think like us, but faster, cheaper and better, right, it can create better results than humans.
Here’s an example.
How many of you use the app Google Translate on your phone or on the desk.
Have you been using for at least three years.
Did you notice about a year and a half ago.
It got substantially better.
The reason why is at Google switched over to what is called Deep Learning they fed 103 human.
languages to Google Translate and said we’re not going to tell you how to translate you figure it out right just it just they kept stacking layers upon layers until the software figured out what I found was, there was a common meta language underneath all human language that’s machine readable.
So, and now instead of Google having to translate from English to Dutch and having these weird idioms.
It goes English to Google to Dutch when you translate from Spanish to Korean ago Spanish to Google to Korean and the net effect is the translations are so much better because Google found this hidden layer.
You can even now feed it fragments.
You could write a sentence.
It was a third and Dutch a third and Spanish.
The third in Swahili and it will come out cling on on the other side does work right and if you speak cling on you’ll notice is pretty good cooling on if you don’t speak going on.
You’re not a nerd.
But that’s the power of deep learning because it was able to find stuff that humans simply can’t do we, our brains are just not strong enough to do that.
So this is sort of the hierarchy.
When we talk about.
Types of machine learning and artificial intelligence.
This is the landscape.
So when we talk about AI, a lot of the times we are talking about machine learning.
Sometimes we’re talking about deep learning the distinctions will be important when you evaluate vendors, because a lot of the time.
That’s how you’re going to access it when somebody says we have artificial intelligence in our product.
What kind go back to that picture the robot with all those algorithms.
Tell me about which algorithms, you use most of the time.
That’s where the marketer who have that vendor usually needs a change of underwear.
Here’s what machine learning and AI are not going the outcome of the future is probably not judgment day for a while, define a while Kurt’s while believes that will reach a singularity when we get to essentially machine in about 12 to 15 years.
And when that happens, that there will be it will be interesting to see what happens.
We’ll get into that later.
Let’s talk about now is we call this conference now.
Next, what’s happening right now in artificial intelligence for interesting areas.
One is foundational uses of machine learning.
So attribution analysis.
If any of you have ever enjoyed Google’s attribution 360 software.
It’s a fantastic piece of software.
It’s also hideously expensive.
It’s something $50,000, a month with machine learning.
Once you understand what algorithm.
They use you can build your own version of it.
This is one version I built ran it on a friend’s website and say okay let’s figure out what the contribution of every marketing channel is not by the broad channel but by individual source and medium.
So now I can go back to my friends say you need to send a lot more email because it’s really working well for you and you need to focus, especially on your Google organic search results.
Don’t worry too much about being just yet.
And by the way that that whole social network you thought was dead Twitter as actually the fourth biggest refer of traffic conversions to your website.
The reason why this is better than what’s in the box is that it takes into account all those hidden interactions that we can’t see because every.
single path to conversion can have one 510 or 100 steps and with machine learning, you start pulling little steps in and out to see when conversions fall apart.
So this is one very practical use today to take your analytics data and turn it into real attribution data.
Another one was a lot of fun.
Well, I guess a lot of fun for me was LinkedIn stopped giving you shares of a URL on February 7 of this year, they said no more.
You can’t know how much a URL was shared on LinkedIn.
So for those you who use services like buzz Sumo and stuff knows that column is now all zeros all the time.
I don’t like that.
So we had a data training data set about a quarter million shares from the past before that date and taught on machine learning algorithm to fill in the blanks say okay build an algorithm based on all the previous data that will help us infer or impute what those shares would have been in testing it came out about 19 8.
Remove the outliers 99.
So now we can go with.
and fix that problem that we have in our analytics.
So that’s an example of how we’re using today foundational Lee, there’s a way to use machine learning for connection for connecting the dots, quite literally.
This is the marketing profs b2b form from last year, it’s a terrific event.
If you’re not going to it, you are probably not a very good marketer you welcome man.
But in all seriousness, one of the toughest things to do in marketing is influencer identification, even though influence identification and influencer marketing the literally the hottest thing we did a predictive forecast recently influencer marketing is the most interesting the most searched for type of marketing and next year will be 79% more search for so everybody in their friend is going to be asking you about this the tough part is how do you find who’s influential it isn’t just the loudest person and isn’t the person with the biggest number of followers.
It’s the person who everybody else talks about or the person who connects people.
So if you we look at the hashtag for next 10 x.
At the end of the day, who are the people in the room who are helping Connect others to others by using machine learning, we can figure that out in near real time you can figure this out in near real time and so if you were to do this right now with the next 10 x hashtag, you can figure out at the break, who you should talk to in order to make this conference as valuable as possible.
So this is a this is called network graphing software.
The third area where we are using machine learning a lot these days is understanding.
One of the most interesting things about Google and about search engines in general, these days is how much the AI runs everything.
There’s no way to know what is in Google’s ranking.
There’s over 200 dimensions even Google doesn’t know because it’s deep learning.
It’s a big black box to Google has no idea what’s in the box.
But if you were to take, say your company’s name or your industry’s top search term and extract the text from the top 10 or 20 results in Google and feed it to text my.
You could figure out what are the top terms that co locate and all those results using machine learning and do it very, very quickly.
So now instead of having to try and guess what Google thinks are are at least on page content factors, you can build it yourself just build it just have it extract up a language and say these are the terms that you need to check the box on in order to make your on page content work well.
Another one I did recently, which was entertaining was with all of gardens Glassdoor reviews.
How many of you love reading Glassdoor reviews have you dread reading last or views kind of a mix Olive Garden has 2500 Glassdoor reviews.
There’s a decently large number of them and I was curious what do people say, well, in addition to the usual things you would expect from a chain like pay hours sometimes bad managers.
It was soup salad and breadsticks which is 1199 and I think 799 or Tuesday’s customers love it right customers love it employees hate it employees hate.
This dish with a passion rivaled only by politics right people will say things like you are a breadstick slave right because but if you were trying to understand your customer if you’re trying to understand your employees, you’re trying to understand your market using text mining using this topic modeling as a way to dig in and find the unexpected that can then inform your marketing and lastly is prediction being able to predict the future because let’s face it.
As humans, we are fairly predictable.
Most of the time, most of the time you do the same things you you keep the same work week you have the same holidays and so you can forecast based on what people say and do with relative accuracy.
If you were to take data out of like Google Trends.
For example, you can forecast that take five years of backdate and forecast of the year forward and then instead of having to guess when a search term is going to be most popular, you can actually go and just find out and plan your calendar around if this is one for Bitbucket.
DevOps and get lab three companies when is each company you’re going to be most searched for in the next year.
If I am the company in red.
I know when the periods when I need to up my PPC bidding on there on the competitors brand names to try and take some of the wind out of their sails.
That’s what you can do with predictive forecasting.
You can even turn it into a keyword level forecast by week so that if you are creating social content you’re creating on page content you’re creating video content you know what you need to be talking about this week there’s no laser on this thing.
So I’m just gonna point this is again an example, if it is the week of May ninth so week of May 6 for this particular company.
The four things they need to be talking about our marketing courses media training public relations courses and social media training they shouldn’t be bothered with the others because they’re not going to be the most searched for things this week.
Now imagine how useful this is for your company to be able to forecast when things are going to get hot.
We did one in the fall for.
for Outlook out of office is a very popular search term.
What are you doing when you search for that term anyone you’re going on vacation you like how do I turn this thing that I want out of here.
I get me out of this place right and so we involve that search fine goes the highest we know okay don’t send email that week because no one’s going to read it.
So I ran a forecast that when in the next six months is going to be lowest.
It was the week of January 18 of this year it was when that term bottoms out so I sent it I dusted off an email campaign.
I ran the previous year.
Send it the exact same campaign just that week of this year 40% greater results, a year over year because I got the week.
Last year I was two weeks late.
And so that’s an example of how you use this to make real impact.
So that’s today what’s next.
How’s things going to change one thing we’re gonna need fewer humans sorry Coca Cola is doing this for example, they’re using bots and AI to create ads better.
A million at a time and testing them all in market because it takes forever in a day for creative team sometimes to turn around, you know, and ad campaigns of coke just does a million of them 999 9990 of them are going to suck right and they know that, but the 10 that do really well they can then retrain the software and say do another million though of these just like these 10 and so instead of having to wait six weeks for the creative team and legal the bot just does it programmatically and they can find what works much much faster if you do it with a template today email template website template Instagram template.
If you’ve searched for these things a machine will do it without you tomorrow.
Right, so that we’re not going to need nearly as many people to do the same amount of work, our marketing is going to get much, much faster.
Another example the big iron stuff.
How many of you have had the pleasure of going through a financial audit at your company.
Three months 500 Junior analysts crawling up in your business creates a paper everywhere.
It’s not a great time for most people, IBM said I wonder if we could do this so wonderful machine could do so they trained Watson on what a publicly traded company should have and as financials and they took the result, they took the data from this publicly traded company and said okay, Watson go take a look inside Watson completed the audit found 99% of what human auditors found found 30% of what human auditors missed it was done in seven minutes.
So instead of three months.
million dollars and 500 staff, you just now need one machine to do it for you.
Now think about that same kind of power coming to your marketing.
That’s what’s coming your market is going to get hyper personal.
How many of you use personas in your marketing you won’t have to do any more because you will have a persona of one we can actually do true one to one personalization.
This is one of my favorites.
This is Campbell’s Campbell’s food group has this interview.
active with the weather channel and Watson you tell you talk to the ad you the type or you actually use the microphone.
Tell your at the the ad what your favorite ingredient is and the ad comes up with a unique recipe for you made with obviously with what the sponsors content product is they do when you these recipes or kitchen an outcast and tested.
Please use your judgment when you bake them.
But the point is, I’m able to put talk to an ad tell it what I like and it will come up with something that’s unique to me.
I don’t have to figure out like, you know, Charlie CEO persona.
No, it’s gonna be content just for me and the barriers to entry for everybody are going to get much, much, much higher.
So how many of you have heard of Oh gosh, I can’t remember the musicians name IBM did this experiment with a YouTube musician said let’s try and make a hit song so they gave Watson, the corpus of the English language and every Billboard Top 100 song ever and said let’s make a top so.
On and they worked with this musician to kind of tuned up and provide a little bit of that human judgment and and they did the week it was released it went to number one on Spotify charts and it stayed there and stayed there and it stayed there if you were a musician who was not that person you are now suddenly essentially locked out of the number one spot because the part the musician with AI on their side essentially has a massive competitive advantage.
This is going to happen across all domains.
So not just music, but every single industry.
So my advice would be to start learning this stuff.
Now the organizations, you want to take a look at what we call the magic Microsoft Amazon, Google IBM these four companies have AI technology that is relatively inexpensive and if you have any development capability your company or you want to learn it yourself as Eric Angus been doing for a Stone Temple, you can learn it yourself and start building applications with these things.
getting easier and easier if you want to learn to program in these languages.
The two languages to focus on or is a language called our and the language called Python.
These are both great languages to get started in.
How do you prepare your career as a mere human for AI number one you need to focus on multidisciplinary skills, this is these are the top 10 most in demand hard skills.
According to LinkedIn as a in each discipline every one of the in every one of these top 10 disciplines.
There’s a whole bunch of things that are repetitive that you can do SEO and SEM number eight, by the way.
Good job, everyone.
But think how much of SEO is repetitive right.
How much is the same process over and over.
Now if you are somebody who is skilled and user interface design and data presentation and SEO your career is much more insulated from AI because you’re crossing disciplines, you’re doing lots of things that are not just one discipline.
If you are have background in statistical analysis and data mining mobile development and network security, you’re in a much better place.
Why so focus on building your career to be multi disciplinary be able to think algorithmically think how could I turn this problem.
I’m trying to solve into a system that’s instead of just solving one problem, learn to think like a machine.
How can I build a process around this.
How can I train the machine to build a process around it that will free up your time to solve newer and better problems.
Learn to oversee the machines.
So one of the careers that will exist for about five years.
Coming soon is proving the models that machines come up with this is a famous example from a couple years ago, a police department attempted to create a predictive algorithm to predict whether a criminal would re offend would commit another crime, the algorithm was 20% accurate, you would have been better off flipping a coin but mysteriously the algorithm flagged African Americans is five times more likely to reoffend than Caucasians why someone built their bias into the algorithm someone built their bias either into the algorithm or the training set and the machine came kept kind of with the wrong answer.
over and over and over again this is going to be a critical problem in the next five years for everyone to solve.
How can we avoid bias in our in our machine learning in our marketing in our business and is it going to be a legitimate career for some time and be outcome focused, think about process.
Think about the outcome you want the machines are going to do the heavy lifting.
This is a fun cheat sheet for learning how to do deep learning with Charisse I used to have this hanging in my office wall and I would check off a little things as I learned that like wow it’s really difficult.
I finally learned this thing.
Then I went to IBM thing and they released a drag and drop interface for it like well.
So now instead of having to learn everything on that cheat sheet.
All I have to do is go into the interface drag and drop the deep learning system I want so I don’t have to think about learning the code anymore.
I just have to think about how I want to solve the problem.
And we’re going to let the machines do the heavy lifting.
There will be two jobs left in the future right you you will match the machines in the machines will manage you by the.
The way we’re already there.
I was at the local supermarket.
The other day and I was walking on I saw this guy was a little cleaning carton had like a little grocery scanners like why is the guy with a cleaning cart have a little grocery scanners shopping while he’s clean store at the top of each while he has he pulls out a scanner scans the top.
They all walked the aisle cleans it gets the end of the aisle scans again he is being managed by machine how fast or slow as you go through the if he goes through too fast.
You didn’t do a good enough job right he oversaw if he goes to slow.
He’s unproductive and all that data goes back to the system and the system says yes keep this employee or get rid of them.
So think about that you’re already being managed by machines were already being managed by machines in our homes, we talked to Alexa, and Google and Google is literally our second brain.
We don’t remember anything.
How do you prepare your company for AI, you’re going to need three kinds of people.
People love the expression data is the new while I think it’s a great expression, because you ever held crude oil.
It doesn’t do anything useful and stains everything.
It makes a.
A thick black smoke when it burns very unhelpful stuff you need to refine the oil, you need to to extract it refine it and then turn and deliver it to somebody useful.
So the roles.
You will need corresponds to that process.
You’re going to developers people who can take data from where it lives in your enterprise and be able to extract it this is Steve Ballmer screaming his wedding profusely at a developer conference so you’ll need developers to extract your data.
You will need data scientists to process that raw data and turn into insights and analysis and and these are the people who love equations like this.
By the way, if you are looking for the secret of social media influence.
That is how you graph social influence.
Take a vector.
I’m just getting and you’ll need marketing technologist people who can take the refined product and actually put it into market, put it to use put it into into the business.
So that’s who you’re going to need to be able to make use of this future this future will be here for.
faster than you think it is already here in unevenly and it’ll be just a matter of time as vendors make things easier and easier for us to be able to have access to these superpowers.
But what question you have to leave yourself with is are you ready to be an AI powered marketing superhero.
With that, thank you very much.
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.