The Marketer’s Guide to Practical AI

“Big data is like teenage sex: everyone talks about it, nobody really knows how to do it, everyone thinks everyone else is doing it, so everyone claims they are doing it…” — Dan Ariely, 2013.

I love the above quote about big data because it perfectly captures the atmosphere surrounding new, compelling, but poorly understood technology. The general excitement and ignorance teenagers have about sex is compounded by group dynamics. Exaggeration and misinformation spread like wildfire, because the eagerness to seem capable and experienced exceeds the ability to actually become those things.

That heady combination of excitement and ignorance can be amply found in discussions about Artificial Intelligence (AI). Everybody knows that machines have been making impressive advancements in the areas that used to be solely the arena for human intelligence. Those advancements are in the process of expanding the kinds of tasks that machines can perform, and the resulting automation will have profound impact on work and society at large. It’s exciting, scary, and relatively technical — a little bit like sex for a teenager.

The problem is especially acute for marketers. Marketing has become ground zero for the explosion of massive amounts of newly available data. New AI-based techniques in automating the recognition and utilization of patterns in data present incredible opportunities to marketers. It’s understandable that marketers largely lack a deep understanding of these new techniques, but that ignorance will become increasingly costly over time.

We at Blackbelt want to help and the first step, as in any area of knowledge, is establishing terminology. Following is a quick list of AI-associated terms that are relevant for marketers, what they mean, and what you need to know about them:

Artificial Intelligence.

Definition: A branch of computer science dealing with tasks that historically required human intelligence.

What you need to know: Artificial Intelligence is a very vague term, applying to anything that computers do that mirror human capabilities. Machines have historically been much worse than humans at tasks involving the rapid recognition of useful patterns in data, so often the use of AI in business today refers to these tasks in particular. Image processing, language processing, autonomous decision-making, and predictive analytics all generally fall into this category of tasks, which contains the bulk of key AI advances today.

For example: “All of our competitors say they’re using artificial intelligence in optimizing their campaigns. We have to do something to keep up!”

Machine Learning.

Definition: The area of artificial intelligence dealing the automated extraction and application of patterns in data.

What you need to know: Machine learning is really the money term in AI. Automated pattern extraction, for use in prediction, clustering and classification, is the area that has been advancing quickly and showing the most significant progress.

For example: “I want to see if we can’t apply a machine learning algorithm to our marketing campaign data. I know nobody is optimizing those campaign, and it would be great to see what kind of value we can get.”

Data Cleansing.

Definition: The process of detecting and correcting errors in a data set.

What you need to know: In the real world, data is messy, and cleaning it up is often the most time-consuming part of using it in machine learning. Inconsistency in how data is collected and recorded causes inconsistency in the data, which prevents it from being analyzed effectively. When a human does manual analysis, they notice and clean this stuff up. When you have an automated process it still usually requires that someone manually identify and fix the issues with the data. For a big data set, this can be a ton of work.

For example: “We’ll move forward with this machine learning project, but we have to make sure we reserve enough time for data cleansing. Nobody’s looked at that data in years.”

Algorithm.

Definition: A procedure that solves a mathematical problem.

What you need to know: An ‘algorithm’ has an almost magical power for some, but it’s just a process. Here’s an algorithm for figuring out the tip on your dinner bill: (1) subtract the tax from the total, (2) multiply the result by 20 percent. Of course, algorithms can be quite complex, and in machine learning you’ve usually got an algorithm that the computer uses to get a different algorithm. Because ‘algorithm’ is a very general term, it can apply to a lot of things, so for that reason it can get overused. However, it really just means ‘procedure’.

For example: “This vendor keeps talking about their amazing algorithm, but who cares? I only need to see that it improves results.”

Model (The mathematical kind).

Definition: A representation of a real-world system in mathematical terms.

What you need to know: Mathematical models are used in the same way that physical models are used — they allow you to play around with something unwieldy/fragile/dangerous/important without so much risk. Want to know what will happen if this airplane flies into a cliff? Let’s try it with a model! Oh. That’s what happens. Want to find out what will happen if you triple your ad spending? That’s what the mathematical model is for. In AI, a model of your given data set is a normal output for several machine-learning algorithms. That model can then be used to classify or predict future data.

For example: “Our machine learning algorithm made this model from your data, and now we can update it as more data comes in so it just keeps getting better.”

The key thing to remember here is that knowledge is power. The more you learn about AI and machine learning, the better you can navigate this new world of marketing. You’ll also be less likely to make the awkward — and sometimes really impactful — mistakes that teenagers make with sex. Aren’t you glad you left those years behind?