When you’re working with Artificial Intelligence and Machine Learning, there are some terms that you’ll hear over and over. Neural Network is one of them. As a marketer, you probably don’t need to see how the sausage is made when it comes to building and maintaining neural networks, but it’s essential that you have a solid foundation of what they are, how they work, and where/when you should use them in your business.
WHAT ARE NEURAL NETWORKS?
First, let’s start with what a Neural Network is.
A Neural Network is a computer system modeled on the human brain and nervous system. Thus the “neural” part. (In the interest of full disclosure, many techies would change out “computer system” to a “set of algorithms” or “a collection of units or nodes called neurons.”)
Our (human) brains are responsible for many survival-type things such as movement, digestion, and, you know, breathing. They also help us by recognizing patterns behind-the-scenes and making decisions. (We make over 35,000 a day.) The brain takes in information, processes and/or assesses it, and then comes up with a conclusion. The result could be acting on the information, storing it, discarding it for lack of importance, etc.
Even the fastest, most high-powered supercomputers with all the bells and whistles have a hard time mimicking what we do. This is not surprising because we don’t even have insight into everything our brains do for us. We’re very, very, very, VERY complex.
Neural networks are the “human-like brains” of the computer. Their algorithms allow them to learn from their own experience. As humans, sometimes we get it right, and other times, we get it wrong. Neural Networks help computers react as we do. And like us, they make errors at first, but they learn from their mistakes, and over time, they keep getting better. (Okay, this may be a generous comment when it comes to some humans, but Neural Networks can definitely improve.)
HOW DO NEURAL NETWORKS WORK?
Neural Networks have been around since the 1940s. Over time, they’ve matured enough to offer marketers real, actionable benefits.
Above, I mentioned that you don’t need a step-by-step of how Jimmy Dean makes your sausage, but I wanted to give you a quick look at the recipe just in case you ever end up on Jeopardy or something.
Neural Networks have an input layer, a hidden layer, and an output layer. The input layer is where data is entered. The output layer is where the output (processed information) is presented. The hidden (middle) layer connects the input layer, and the output layer, kind of like the cheddar does in a grilled cheese sandwich. (I’m from Vermont. Cheddar is the only option.)
The hidden layer consists of units (artificial neurons) that take the plain, ole, boring input data and turn it into something magnificent for the output layer to present. The magic? It happens in the cheesy middle hidden layer. Then, it goes to the output layer for presentation.
Neural networks learn by processing examples (aka training) or performing tasks. Critics feel that this is one of their shortcomings. The training can be time-consuming, especially if you have data quality issues.
There are all kinds of Neural Networks: Hopfield, Boltzmann Convolutional, Perceptron, Long Short-Term Memory, Networks, and so on. The two that you’re most likely to hear about are Recurrent (most common due to their greater learning ability. Data flows in multiple directions. Used a lot in processing language) and Feedforward (information travels in one direction from input to output.)
WHAT ARE THE BENEFITS OF NEURAL NETWORKS?
They reliably handle oodles of variables. They’re effective at gathering and extracting information.
They learn. They’re solid at figuring out the relationships and the connections between the data.
They’re flexible, and they adapt.
They often make more complex/sophisticated predictions and identify valuable trends.
WHAT ARE THE DRAWBACKS OF NEURAL NETWORKS?
From a marketing perspective, they can be too Black Box-y for my tastes. It’s often hard to find out how it arrived at its prediction. They can also take a lot of time to develop; technical people get enamored with them, so they sometimes ignore the simpler decisions, and they can be pricey. You may also find that they need more data than traditional algos, that they take FOREVER (read: several weeks) to train, and that they can be slow.
My complaints are on par with a person who squawks ad nauseum about the restaurant while you’re eating dinner there and then says: “we will definitely come again.” As a marketer, I like what they’re doing for us now, and I love what they’ll do for us in the future.
WHERE ARE NEURAL NETWORKS USED IN MARKETING?
SO. MANY. WAYS.
Here are some of the most common:
Classifying information and clustering huge amounts of data very quickly and reliably. (Despite the fact that critics yap about the speed, this is a legit benefit. They’re not always the best way to do things but when the project is right, they are speedier than what you usually do/get.)
Quickly identifying patterns that are too complex/time-consuming for humans to pinpoint. (They also find stuff that we don’t see.)
Making predictions and recognizing trends.
Improving engagement by manipulating (positively) rewards systems, loyalty programs, etc.
Assessing credit risk, fraud detection, payment plans, and special pricing
Detecting SPAM and content that doesn’t meet standards/policies
Geoanalyzing – You see this a lot in Retail. They use historical data coupled with coordinates (location), weather, local stats, and so on to predict demand, product assortment, amount of shelf space, optimizing store layout, and all sorts of other things.
Inventory management – yes, Inventory is typically an Operations thing, but when you start doing a lot of AI/ML projects you realize that you need to account for it as well, especially in your search function and product hierarchies
And the list goes on….
How are YOU using Neural Networks in your business? Have a tip you’d like to share? Have a question you’d like to ask? Tweet @amyafrica or write email@example.com.
A Down-and-Dirty Definition. (Read more about these here.)