In this week’s episode of In-Ear Insights, we eavesdrop on our MarketingProfs B2B Forum session talk about AI and analytics. When only 30% of CMOs use the analytics their organizations generate, and 70% of companies ignore their data, we know there’s plenty of room for improvement. Tune in as we explore AI and how it’s changing the value proposition of marketing analytics.
Listen to the audio here:
What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for listening to the episode.
Thank you very much. Good afternoon, everyone. Alright, so this
is the food coma slot, we have intentionally lower the room temperature by 15 degrees, just to make sure
that the carbohydrates don’t get the best of, you know, actually, they are going to try and make it slightly warmer. But just a little.
today’s talk is all new. What is in the app is not correct. So if you’d like a copy of the slides, you can go to the aptly named Where can I get the slides.com to get the slides
and you don’t need to take a picture of its Where can I get the slides.
So let’s talk about analytics. Specifically. Let’s talk about the problems we have with analytics, b2b,
b2c, marketing sales doesn’t matter
we have five problems with analytics these are the five these volume variety velocity
veracity value. Come on in find a seat.
The first problem we face is a volume problem. The amount of data that is coming at us is hitting a record numbers right now this year according to Seagate, and IDC, we as a civilization will create 30 zettabytes of data.
Now zettabytes is a really really difficult number to get your brain around because it sounds like it’s made up it actually isn’t made up if you were to start binge
watching Netflix start binge watching Netflix in the EEOC in area 55 million years ago and he didn’t stop for anything.
You would just about get to one zettabytes of data today that’s how much a single zettabytes and we’re creating 30 of these year what’s going to happen in the next seven years is even more astonishing. All these lovely devices that we own that I have sensors and everything we could possibly put them in are going to create 120 zettabytes a year by 2025.
Some of that’s going to find our way into marketing data. But even on today’s practical skill.
How many of you have ever tried to download Facebook’s analytics from your Facebook page? Is that fun? You get a spreadsheet with 14 tabs. Right. This is the lifeblood is very, very difficult to use. So volume is our first problem. Our second problem is a variety problem. The amount of types of data we are running into is continuing to grow and expand
in just 60 seconds on the internet. In 2017, we had things like 16 million text messages, 452,000 tweets, 15,000 gifts and messenger
3.5 million search queries in in 20 1818
million text messages. 266,000 hours on netflix. 1.1 million swipes on Tinder I’m guessing mostly left.
as marketers though, we have an even bigger, more pressing problem. How many people love this landscape by Scott Brinker with 8000 marketing technology companies audit. The problem isn’t the two companies. The problem is every single one of these bloody companies has their own analytics and they don’t talk to each other.
Our third problem is a velocity problem. The speed at which data is coming at us if you were born know if you’re graduating college
this year in 2018. You were born in 9798, during which time there were about 4 million news stories in your lifetime. This year, we are on pace to track to about 110 hundred 15 million news stories. That’s 200,000 news stories a day. Even if your company does a bang up job on the front page of The Wall Street Journal, you are one of 200,000 stories, there’s no way that we can keep up with that much data individually. Our fourth problem is a veracity problem. And I don’t mean fake news. I do mean the ability for us to trust our data. This is a fun example. A few years ago, Google Maps said if you try to go from Topeka to Tokyo.
Step 27 will be kayak across the Pacific
Ocean, which is a really, really bad idea.
You can shouldn’t but more close to home for b2b marketers. How many of
you have this person in your marketing automation
system test test with test at testing calm, right. We all have terrible data let littering our systems. And because of the speed at which we’re getting the data, we were able to keep up with a less than less. Finally, we have an analytics value problem. We R CMOS are boards, our stakeholders don’t really trust our analytics. And as a result that percentage of time marketing analytics is used in decision making has actually gone down. It hit a peak in February of a 42% of organizations, but it’s gone down to 35% and
disturbingly b2b services is the lowest fall by b2b product,
right? So b2b marketers don’t trust and don’t use their analytics.
Why? Well, it’s all these problems,
all of our analytics
have specific value, but the value changes based on how we use it. When you look at the hierarchy of analytics,
the base level descriptive
analytics, the answers the question, what happened has very little value by itself, right? There’s no value to someone backing up the dump truck and pulling the data all over your desk. Like, that’s not a fun thing to have happen. So it’s very low value by itself, we start to create more value when we have diagnostic analytics. Why Did something happen? Can we answer why you abandon this one? Can you answer why you didn’t call back? the salesperson has called you 15 times today,
that’s sort of medium value,
we get to unlock the value of analytics. When we get to things like predictive, what will happen next? How can we predict and forecast and plan and then we get to prescriptive analytics where our software, our tools, and our teams help us understand what to do that’s very high. And the ideal is proactive analytics when the machines do it for us. And unless you’re Jeff Bezos, that’s probably not where you are. But here’s the challenge. You can’t skip steps on this ladder. And almost everyone, This room is stuck here. We can’t find the data, have no idea where it is, we can’t make use of it. And as a result, this is how we tend to use analytics wrong.
So what’s the solution to this problem? The solution is partially artificial intelligence and machine learning. And I say partially because it is not the complete solution. AI is not magic, right? It is not fairy dust is not unicorn poop. It is nothing that is going to create something from nothing. What AI promises. And what it can deliver on today are three benefits, acceleration, accuracy, and automation. Acceleration means we get the result faster. When you think about 200,000 new stories, how long would it take you to go through those new stories and find mentions of your company? Probably the rest of your life, right? But the machine can do it very, very quickly.
The second benefit is accuracy, can the machine do a better job than we can?
Yes, because he humans as a whole are terrible at things like accuracy. And third, how many of you have a job or a task or thing you do in marketing, that it’s pretty much the same thing over and over again, you know, like, you have to schedule social media for the week, and whatever. It’s just not fun. It’s not rewarding. It’s not exciting. It’s the same copy paste graph that you’ve been doing for 20 years, right? putting data into a spreadsheet? Well, the machines can help us do that stuff really well. So those are the benefits of AI.
Let’s set some quick definitions. What is this thing because everybody in their cousin has is talking about it. The analogy that I’ve heard that I really liked that my CEO tells me is profoundly and appropriate
is that AI is like teenage sex, everyone’s talking about
it, but no one’s actually doing it.
Artificial Intelligence is teaching machines to learn like humans, when you look at how a human being developed, we start with basic, it’s very simple algorithms, how to see how to here we develop land which and eventually develop higher cognitive function, by the way, is the parent of a teenager, I appreciate the fact that the cognitive function drops off when they hit 13 on this chart. When we look at machines, and how they’re evolving, we start with things like statistics and algorithms evolved to machine learning, get too deep learning and will eventually get to what’s called general purpose AI. When a machine becomes sentient. And can think for itself this is a long ways off just to set expectations. Skynet will not be taking over your marketing. Tomorrow,
all AI begins stats, and probability AI is math, not magic. Once you have the basic statistics to be able to understand your data, you build algorithms, these are the things that you do repeatedly over and over again, every time you’ve done a B testing, you are using an algorithm you used
an algorithm this morning, I guarantee you did because you probably put
the same general article of clothing on first, some people put on the bottom first up at the top on first a few people put their socks on first. But it’s generally the same, you have an algorithm them a way of doing things. This is where additional software stops, you make the thing it does the thing machine learning by virtue of the name means that we are teaching the machines how to learn, we give them data, and they give us software. Instead of we give them we make the software and they give us data, which is traditional software. So for example, suppose you had a table full of colored blocks, you could use machine learning in two different ways to make sense of this, if you knew what you were looking for, you’d be doing what’s called supervised learning, which is where you tell the machine, hey, this is the color red. And you do this a million a billion or trillion times until the machine gets this is the color red it can get pick it out. The most famous example of supervised learning
is probably IBM. Watson Watson was
there was a woman in Tokyo who was had contracted a form of leukemia, and she just was not responding to treatments, you kept getting sicker and sicker, the doctors couldn’t figure out why. So they saw IBM and the University of Tokyo took her entire genome plus 233,000 ecology journals and said, find the match, find out why this isn’t working Watson sequence the genome and read all 230,000 journals and said, You’re treating the wrong kind of cancer, you’re treating the wrong kind of leukemia, they changed the treatment, she made a full recovery. What’s interesting about this is not just that Watson did it accuracy, but it did it in 11 minutes.
So that’s supervised learning, you’ll see this much more and things like lead scoring and b2b analytics, is this a good lead, unsupervised learning means instead of knowing what we want, we want to figure out all the ways that we could explore something. So you’d have, you can explore these blocks by colors you do by shapes by sizes and things like that. We want to know what’s in the bag. As b2b marketers think about how much data you have that is not in the side of a nice tidy table, you have Customer Service Center logs, you have CRM data, you have your inbox, which is probably the least structured thing in your our entire lives, we have photos, we have video, we have music can audio, how do we make sense of it all we use what’s called unsupervised learning to categorize stuff. Here’s an example I did
taking 2600 articles and turn it into a chart. So I don’t have to read all 2600 articles, I can just have the machine summarize and get a sense of what people were saying about this customer, machine learning math and stats, not magic, which also means that once you start to learn all the different names and the labels and the techniques and how they work, you will be in better position to evaluate things like vendors, when you’re talking to a vendor in the learning lounge has what’s called this year at the expo hall. The third branch of machine learning, which is still relatively new, and not super practical yet, for b2b marketers, is deep learning this is when you tie lots and lots of machine learning together in like layers of pancakes. And you get what’s called Deep learning. And this is where the future is going. Because we get machines can do things better than we can machines using deep learning can read lips, Bowen, humans, now they can understand they can see how many of you use Google Translate, okay, have you ever been using for at least two years? Okay, did you notice
you may have met us, it’s been a while now. But our two years ago, Google Translate got substantially
better. Like overnight, what Google did was they, they took away the old algorithm and said, okay, and we’re going to have this deep learning software tried to understand human language. So they took 103 languages, all of them and stuff them into the this system called DeepMind. And it figured out that underneath our individual language is a metal language is a proto language that the machine could translate. So when you translate in Google Translate, you go from English to Google, to Chinese, or from Swedish to Google, to Danish, which also means that if you were to put in sentence fragments and different language that would still spit out the proper language at the end. That’s how it the power of deep learning to be able to make sense of things. We see this, most as b2b marketers, and SEO Guess what, nobody in the world, including Google has any idea how Google search algorithm works, right? Anyone who says they do doesn’t understand machine learning, because that album has gotten so complex and so big, and so deep over the years, that we don’t know what’s inside, we only know how to tell it’s doing the right or the wrong thing. That’s the layout of machine learning and AI, how of
algorithms, machine learning and deep learning. And like I said, the reason for this is so that you
start to develop a BS detector.
When a vendor to start spouting off terms, you can start to ask, tell me more about how this works. We want to help you make a little bit smart about these things. To give you a sense, in terms of analytics, when are you going to likely start running into machine learning, we go back to our hierarchy in the first few stages of the ladder of analytics, you’re still doing just good old fashioned stats and math, quantitative data, right? You’re not going to be doing machine learning at the Hey, what has happened last week, where you start to see this as in things like predictive analytics, where you have statistical analysis to figure out what’s likely to happen. And then you start getting into the machine learning and deep learning. If your organization is down here at where’s our data, we can’t find it. You don’t have to be terribly concerned about trying to launch of machine learning initiative that transforms the whole company today, you first need to find your data because nothing will work without that. Now, what is AI good at and what is it bad in 2010, at the time Secretary of Defense, Donald Rumsfeld was somewhat mocked for saying it is conference call that there are known knowns and unknown unknowns. But this is a really good way of categorizing tasks that machines could do
when you have known knowns, which is we know the problem
and we have that we know we have the data to solve and we know that there’s a process to it. That’s what AI is really good at. It’s really good at acceleration, accuracy and automation. When we don’t know what we know being that we don’t know what the problem is, clearly, we don’t know where our data is, like, for example, I was talking to a company earlier this year, their sales department refused to give any data to their marketing department, same company, but for some reason, I don’t know if people are weird.
That’s an unknown unknown. On the other hand, there are things known unknowns, we know what the problem is, but we don’t know where the data is. or we don’t know how to get it. We don’t know even know what data we need. That’s a data science problem. That’s not a machine learning problem.
And then there’s the unknown unknowns, which is who knows what’s going on.
But believe it or not, that’s a good thing, because that’s where we still have jobs, because we need to do be able to do things that create and talk to other human beings, which is my least favorite thing to do.
Machines are bad at stuff like empathy, right? You can simulate it, but they can’t actually be empathetic. They’re bad judgment, big complex judgments, should we change fire this client is this aggression, corporate strategy. Overall, machines can’t do that. Because there’s too many variables is too complex a question machines are bad at general life experience, they can’t cross domains very well. And mostly machines are not good at relationships. The exception is, if your customer experience is so terrible that a machine is better, like, say, the Department of Motor Vehicles, I think we’re all agreeing with us love to never talk to another human again.
So let’s look at some practical tactical applications of how you could use AI today, or how it’s being used today, in the field of analytics,
we use it to answer three questions Who, what, and when. And if you have these questions, these are applications that you would want to use artificial intelligence for, for example, in who all of you are at this wonderful event, the marketingprofs b2b form when we look at every conversation people are having if you wanted to find b2b marketing influencers there’s a number of different software tools that can do that. But when you use a certain type of machine learning it’s called network graphing technology you can isolate and determine Hey who’s who’s being talked about the most now if you wanted to say talk to one of these folks to get them to represent your product or service for influencer marketing you might want to look at the people who are talked about the most like Ann Handley because should they just mentioned your product? Everyone looks to them for the answer, right? This isn’t stark contrast how most influencer marketing is done, which is who is the loudest? Or who has the largest number of followers, you know, focus on who is talked about the most, because that’s who everyone else looks up to. And the output of this is not just a cool chart, but actual data, you can go and use and hand off to a team of people or an agency to work with. So that’s just one example of how you could be using machine learning to answer who questions also applies to things like lead scoring, who is a good lead, every variable that you collect right now could be used to put together a formula saying, These are the things that indicate someone might be a good lead, and it’s probably more than just do they have the budget, authority need and timeframe, there are a whole bunch of other pieces of data you can gather, and you probably are, are already gathering them that you can use.
The second thing we can answer machine learning are what question
how many of you currently feel like you have an amazing solution? And for attribution analysis, you know exactly what the most valuable things you do are.
So attribution analysis, at least from a digital perspective is something that you can solve with machine learning techniques. There’s many techniques, but one of them is called mark off chain modeling, which is just a fancy way of playing Jenga with your conversions. Every step somebody takes from beginning to end, whether it’s attend a webinar, read an email, click on a tweet are things that your system and software are tracking, as people convert, as you gather data, you can then start to run simulations of conversions, pull away a step and see if that conversion falls apart. This is an example using that same attribution model to figure out what are the most important channels in this case for this company was email was the most important channel, even though in the regular analytics tool, they thought it was organic search, what actually causes conversions the most, the thing that you took away their marketing would collapse is email then referrals in their comments on their blog, then organic search. A Funny Thing Happened actually there, this company was all in on Facebook, like, yeah, we’re going to do Facebook marketing, grow our cut our, our community of that of b2b marketers there. And but they, what they found was, when they looked at their analytics, it was actually
Twitter that drove that push people over the line much more often than Facebook. So they were able to realize maybe they need to change their tactics. Another fun example of using machine learning today is reverse engineering, Google, we don’t know what’s in Google’s algorithm. But we can know what the algorithm spits out. So when you type in a thing, like say, industrial, concrete, right, or medical devices, and you get a certain number of results, if you were to take the content of all those pages that Google ranks highly for all your terms, and stuff them into a text processor, you could pull out and say, hey, these are the terms that all these pages have in common, right? So we don’t know what’s in Google’s algorithm. But if you know the output, you can reverse engineer the outcome and say, okay, so for our industrial concrete site, we know we need to talk about line we need to talk about summit mixtures, we know we absolutely must talk about curing time, etc. And that will help improve your SEO because it creates more authority on your site. Again, using machine learning to reverse engineer Google another fun example is determining the strength of your brand. Suppose that you took your brand I took marketingprofs I took 240,000 posts about marketingprofs in the last year and said what terms are most closely associated when people say marketingprofs This is a technique called vector ization and no surprise ocular marketing markets, right, but infographics, Mari Smith, Mark Schaefer, Mark trap Hagen some some influences found though in their digital campaigns, if you wanted to figure out what was your brand most associated with, you’d want to use machine learning to distill that out of all the conversation that’s happening about it. If you want to do this for competitor and figure out then Okay, where are we strong, and where our competitors strong use machine learning to do this. The third and final example where machine learning is useful for your analytics is an answering the question of when, when is something going to happen. Predictive Analytics.
We did a fun thing. Looking at just cheeses, the names of different cheeses These are 40 different cheeses. And every we can forecast from search data up to a year in advance. When each week is the most what’s gonna be the most popular cheese like mozzarella right now it’s cheddar this week, followed by string cheese, and then hover party is the third cheese this week. But two months ago, it was Hulu me because you can grill the looming
now imagine using this type of predictive analytics for your industry. So if you are scrambling in medical devices, knowing what different search terms people are searching for, and being able to forecast when interest by the audience will be highest over the next 52 weeks, would change your campaign timing, change your budgeting, change your content, all these things. For most of you in the room who are sitting in the front forward seats, there’s a card on your seat that tells you when people are going to be checking their email, because we did a look at all the different ways people type in things like outlook out of office, because you don’t Google for that term. Except when when you’re about to go on vacation. can’t figure out how to turn the thing on. So you can leave the office and go on vacation. Now if you figure out when people search for that. But at least they’re in the office, they’re reading their email, and so on that card. And for those who don’t have a card, this would be a time to take a picture of the slides with your phone. These are the weeks to send email. And these are the weekend not send email. It’s on the card. Keep that card at your desk over the next year. And time your campaigns. We did a fun thing. We figured out the worst week to send email this past year. And it was actually no surprise the week of July 4, right. No one’s checking their email. So we sent out what we call the worst converting email of the year. For the five people who are in the office. They actually got a chuckle and wrote back to us.
But you can use machine learning to predict and forecasts not just search. But anything in your analytics. You can plug this into Google Analytics and forecasts
every channel search
social referral traffic, pay per click advertising, all that stuff is stuff that you can forecast, the most important thing to do is to start now on your machine learning journey need to get started. Now because machine learning is fed by data. The sooner you’re collecting glean data, the more train of a training library you have. So how you get started. The journey is a seven step process. Again, you begin with that data Foundation, you’ve got to get your data in order. You’ve got to have good governance, you’ve got to have good compliance policies process to make sure your data is functional. Then you have analytics, you become data driven, you start making decisions, you set KPIs that have meaning, you build qualitative research capabilities, you can have you use things, focus groups and consumer surveys, or business surveys or customer advisory boards to answer the question of why something happened. Next, you start automating processes because the faster you can automate processes in your b2b marketing, the more time and budget, you’ll have to do the next few steps, which is start building your data science capabilities. Before you can do AI, you have to be able to date to do data science, to explore the unknowns, to have the ability to write code and to analyze data and look at the data you’re working with. For the conclusions gives you once you’ve done that you’re advanced to machine learning where you’re doing having machines help you with the process automation and doing those supervisors unsupervised and deep learning. And eventually, you become an AI powered enterprise where you are AI First, there are only a handful of companies right now that are AI first. And they’re mostly like Google, IBM, Facebook, etc.
How do you get started if you have money, but you don’t have time, hire people, or outsource the capability if you don’t have money, but you do have time Look at the services and software offered by companies like Microsoft, Amazon, Google, and IBM. They all have relatively affordable machine learning solutions that you can fit into your business if you really don’t have money at all. But you have a lot of time learn to do this stuff yourself. Every example I’ve shown you in this presentation of of using Sheila has been done on this trusty reliable computer here, no big cloud, no giant quarter million dollar projects, just me and these programming languages, it will take you six to 12 months to learn the programming language, and probably another six to 12 months to learn the machine learning but start now if you want to start and your company’s not ready, you acknowledge your company’s not right, that’s okay, you can still personally learn. And the thing that I recommend you do is pick one thing to measure and analyze the crap out of it. It doesn’t even have to be your company. It just has to be something that you can measure. For example, I downloaded 1.4 million medium posts to figure out what gets
the most number of claps, which is their version of likes. It turns out that on medium using
machine learning to analyze all the possible ways to look at a post it just length, you could be verbose and say garbage. As long as the really long post people like it that’s medium for you.
How do you prepare your company for artificial intelligence, you’re going to need three types of people, those three people well, I like the expression data is the new oil, because if you ever actually work with crude oil discovery, right, it’s like black tar doesn’t do anything. But in order to make use of it, you have to extract it, you have to refine it, you have to bring it to marketing your data is exactly the same way you need developers, people have the ability to connect all these different systems and pull data out of those systems and put it somewhere. But then you need a second second type of person, someone who can take that data, structure it architected,
refine it, process it, and get ready for business use. And then you need marketing technologists, many of you in this room to take those insights and deploy them and make business impact. So it’s developers,
data scientists and marketing technologists, that’s where you’re going to need. So you’re probably wondering, at this point, am I going to have a job? The answer is, maybe
here’s how to tip that to a yes,
number one, focus your own growth as a marketer on having multi disciplinary skills. When you look at the top 10 most in demand skills by best by LinkedIn cloud, blah, blah, blah, blah, right? Any one of these professions has a substantial amount of things that can be automated, right? If you’ve done SEO, you know, that a heck of a lot of SEO is boring, repetitive work, it will eventually be done via machine.
But if you are good at stats, and UI and SEO,
it’s a lot harder to replace you because you have capabilities across different disciplines. So look into your professional development training, what can you be doing that will give you an edge because you can do more than one thing, you’re not just a specialist in one area, your specialist and maybe two or three areas learn to think algorithmic live, and to think like a machine. So you can help design the machines. And what they do this doesn’t mean you have to go and learn how to code you just have to be able to think it when I’m facing a problem. Maybe don’t just solve the problem. But think about how do I design a system to solve this problem over and over again, without me, the third thing to prepare your career for will be to learn how to oversee the output of machines. This is a critical skill that will be an exceptional demand for the next five years, if not longer. What do I mean machines are only as good as what we put into them garbage in, garbage out. But a lot of people still have this belief that AI is magic and therefore is always right. It’s completely wrong. bias is one of the greatest dangerous we face as marketers and actually as a society in 2016 Pro publica did an article on how a police department attempted to predict whether or not a criminal would offend someone who would we offend their algorithm was 20%, right? You wouldn’t a better flipping a coin. But their algorithm predicted that African Americans would we offend five times more than they actually did. Someone put their thumb on the scale and said, I want the machine to say this and assume that because it was everyone thought was magic. That was always right. It’s not earlier this year, Amazon got Heck, a lot of trouble. They said they had to scrap their AI based hiring system because it discriminated against women. Why? Because their existing training data set all was mostly male. And so the machine learned the bias and spit out the bias. So being able to oversee the machines is going to become a critical skill
be outcome focused. Like I said, you don’t need to learn how to code that’s generally something that only really weird people like me do. For example, I used to have this lovely chart, my office will have all the different deep learning techniques I had to learn, right and write code. And I got to about here in about a year and a half. And I went to an IBM event earlier this year, where IBM said, Oh, we made that drag and drop. Oh, ok.
So now, instead of you having to learn how to code, you just need to know what the Lego blocks do, and drag them and drop them into the right order to get the outcome you want. So be outcome focused, let the machines do the heavy lifting. And you’ll be able to get up to speed and machine learning very, very quickly.
In the future, there will be two kinds of jobs, right? Either you will manage the machines or the machines will manage you. And that’s it. So decide today, which you want to be.
And with that,
what you want to run the microphone,
we want to do that or just have people yell really loud.
He comes to microphone,
or insights are very closely tied to strategy and goals, right. If you don’t know what your goals are, they can’t figure out what’s helping you lead to those goals or lead you away from this call. So a big part of insight development is knowing where it is that you’re supposed to be going. And then being able to judge whether any one thing is helping you move closer to or further away from that that’s a skill that applies regardless of whether you’re talking about data or not. I mean, if you’re talking about even something like hiring, we’re hiring the right people are wrong people. Well, if you don’t care who you higher than it doesn’t matter. But if you’re if you care very much about having things like diversity, then you can start to look for trends, new data to say yes, we are heading towards or away from that goal.
Yes, those questions, a lot of those questions like who, what and when our questions as well suited to answering if your data is good, if your data is good, if your data is bad AI, can’t do squat, because it will always spit out bad results. More specifically, one of the time series forecasting, which is prediction is something that AI has really good at. Another thing that’s really good at is what’s called driver analysis. When you have a an outcome of some kind, and you have a bunch of data. Imagine, for example, you had a big spreadsheet. And in the first column, that spreadsheet was sales, right? The number of sales you made. And then every subsequent column was things like marketing, qualified leads, sales, qualified leads, appointments, dials, and then you’re going to email marketing email sent. And then you’re looking at web visitors. And you’re looking at social media tweets and likes on Tuesdays and number of emoji, right, you imagine this gigantic spreadsheet, machine learning and AI is ideally suited to saying, if you want this outcome, which combination of variables, what interactions are likely to lead to that outcome, that you can then develop a testing plan around and figure out Yep, that’s, that’s probable. Now go find out that’s causative
one of the more interesting and complex forms of of that type of insight derivation now is called multi objective optimization, because one of the most dangerous things you can do as a marketer is to optimize for only one thing, we only want more sales, which will logically do things like, you know, hating your customers, and delivering no customer experience, because that doesn’t help sales. So being able to say, We want sales, and we want great customer reviews, and we want no complaints. And we want a good reputation. Machine learning is trying to develop capabilities to balance and and trade off for good it outcomes across the board. So one great outcome and everything else goes to hell.
Next question, who’s got the microphone
start right here?
Well, so you need to do a whole bunch of data governance up front. And so good data, there’s like six or seven different ways to qualify data, right? It has to be clean so unclean data data that’s broken eight ways to Sunday, real simple example of that would be like, you know, you you have tests at testing calm and your CRM, right, you know, that’s not good data, it has to be complete. So if your Google Analytics pixel stop working for three days on the website, and all the whole bunch of zeros, like, okay, don’t have complete data, right? Your data has to be compatible, it has to be it has to work in and out of the systems that you work with, your data has to be chosen, well, meaning that you’re choosing on you’ve done the bit, that big analysis to figure out what to actually track your data has to be comprehensive to be able to answer the questions that you want to cover it. And that’s, that’s where you’re asking is, do we know enough to be able to answer the question in a comprehensive way, there’s an exercise we actually did a Tuesday workshop called outcome mapping, where you draw out a path from your goal, and to every piece of data that links to that goal. And so it will take a long time to do the full exercise. But that’s one of those things, that’s a really good idea to do. So you can say, Yes, I have the state or Nope, I’m missing that data. It gets really interesting when you do that with like, the head of sales and the head of customer experience and the head of finance because you start going, Oh, they all care about like these 44 additional variables. And that chart gets real big. But that’s how, you know, yep, we’re taking into account everything we should be. The other thing is knowing what your outcome, what outcome you’ll you’re generally looking forward to. So it passes a sniff test, right? You say you have a goal of hiring 50% women, you’ve hired 4% women, something’s clearly wrong. So knowing like this is outside of the bounds of reasonableness, or what we’d expect is is part and parcel of that data question.
If you wanted to try a little bit hands on, I would recommend that you sign up for a free account for IBM Watson studio to do the drag and drop stuff because you’re not going to break it. You literally can’t break anything. And you can put in data sets and just start to get some experience there is there a whole series of courses from statistics through machine learning through deep running their recommend if you on that card is an email just shoot me an email, I can send you to a blog post that has a list of all the courses and books that we recommend that you use to start scaling up.
comments, Dad jokes.
All right. Thank you very much, everyone.
What help solving your company’s data analytics and digital marketing problems. This is trusted insights 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.