A couple of weeks ago I had the pleasure to go to one of Israel's largest annual marketing conferences: All Things Data. The first speaker up was Jim Sterne, a “professional explainer” and the author of 12 books (and counting). His newest book and the topic of the lecture was “Artificial Intelligence for Marketing”, a daunting way to start my morning since data science, math, and coding are far from my vernacular of know-how.
Lucky for me, Jim must have foreseen that look in my eyes and the eyes of about 100 or so other marketers sitting in the room. So he started from the beginning.
The Basics of AI
Artificial intelligence (before we go into what it has to do with marketing), can be broken into 5 sections:
- Robots - Probably the most well known form of AI from the movies, robots nowadays are doing things that I can’t even imagine doing myself. Like this backflip for example.
- Face recognition - We’ve already seen this used both in the movies and IRL to sign in to portals, as well as knowing which friends to tag in photos on Facebook.
- Natural Language Processing - Think of your friends Siri, Alexa, or a bot. They can take words you type or speak, make sense of what you’re saying (or at least trying to say), and offer you a response based on your requests.
- Computer vision - Now, imagine you give a computer a set of laws about what a chihuahuas looks like. Now try to give the computer this picture game below, with a mix of chihuahuas and muffins. As humans, this is pretty funny, and while it may take us a second to look at each picture closely, we’ll most likely ace this test. A computer? Well a computer may be faster at spitting back answers, but they can be pretty bad at these sorts of puzzles. But they’re slowly getting better due to...
- Machine learning - This is the part of AI that allows machines to learn from their mistakes (something us mere moral are still grappling with). Let’s break down how machines do it.
3 Needs for 3 Deeds
This next part of the presentation is what Jim referred to as “3 needs for 3 deeds”. When I was jotting down my notes during the presentation, I ended up making a table that looked a little something like this:
|3 Needs||3 Deeds|
Let’s make sense of my scribbles. In order for machines to be good learners, they’re gonna need a lot of data. Lucky for machines, we live in the era of overwhelming amounts of “Big Data”. Unlucky for machines, most of the data we have is messy. So we’ll need to clean that up so it can be used by the computer properly.
But data with no goal, is mostly pointless. Think of the computer asking you, “based on the data you gave me, what would you like me to figure out?” You answer or goal can be anything from, “let’s detect which of these cells are cancerous” to “should this automatic car stop, go, or yield?”, or in the case of marketing, “which of these people should view this ad?”. Based off the goals you set, the computer will then be sent on its way to make sense of all the data in front of it, reder which of the data is important in completing its task, and then using that data to find a solution. For our last example of a goal relating to ad placement, the computer may use pieces of data such as time spent on a website, number of pages viewed, previous purchases, and things of that nature in order to determine who should see a 15% off coupon ad.
Once the machine starts trying to accomplish a goal, you can offer it the control to, you guessed it, detect, decide, and revise what data was helpful in its mission, what it can do without, and what other pieces of data may be beneficial in completing the task. Going back to our example, the computer might learn based on real-time responses that visitors chosen to see the ad should spend more time on the website and have viewed fewer pages in order to receive the pop-up offer.
The Machine’s learning curve
Just like people, machines have different ways in which they can learn, each coming with their own pros and shortcomings. Let’s take a look.
- Supervised - In supervised learning, we offer the computer a training crash course where we offer it a specific data set’s input and correct output. This is explicit so it’s a great way for a computer to learn, but it can sometimes be tedious to have a human organize these data sets and babysitting their progress.
- Unsupervised - This is where Jim started talking about correlation and causation. In these kinds of learning models, we don’t offer the machine ‘right answers’. We leave it up to the machine to find and create categories based on data we’ve given them. Maybe a machine will look a crowd of people and divide them up as men vs. women, or those wearing hats vs. those who aren’t. You get the idea.
- Reinforcement - Anyone who ever took an Intro to Psychology course learned about human reinforcement and remember something about a rat and a lever. This one’s kinda similar. Here, machines are given data, a goal, a ‘lever’ of what they choose to do as behavior with the given information, and lastly are given feedback (although not always instantly) on if their behavior was ‘good’ or ‘bad’ in order to learn better for next time.
AI for marketers
Throughout the talk, Jim mentioned numerous ways in which AI is beginning to play a big part in how we do marketing. For starters, he talked about x.ai, a scheduling bot that will send emails for you to help you tee-up, reschedule, or even cancel meetings for you.
Another cool example of companies harnessing AI is GumGum V.I., a company which uses computer vision to scan BILLIONS of photos across TV, online stream, and social media to ones relevant to your brand. The company can then give you serious insights on what campaigns work best in regards to gender, household income, time of day, top performance verticals, etc.
So what else can AI do better than human marketers? Jim gave a pretty nice list of marketing tasks we can hand over to ML, including testing, lead scoring, personalizing content (something I love doing with HubSpot Sales), inbound email sorting, social media monitoring (think Hootsuite), and many others.
Going to All Things Data, I had the thrill of listening Jim, a man who’s seen in the marketing arena move from the streets, into digital, and now to AI. Walking in, I had a good feeling I was going to be lost in a sea of data for an hour, only to walk out dumbfounded and more confused than ever. Instead, I learned from Jim that AI is not only something that can be tangible to the average marketer, but is something we are already seeing before our very eyes come to life.