It’s no secret that the advertising ecosystem relies heavily on ad targeting algorithms — the computer models that determine who is targeted with which content. The algorithms used by major platforms, along with the data that feeds into them, make them attractive places to advertise, which in turn fuels massive ad revenue for the Facebooks and Googles of the world.
But first, it helps if you understand how algorithms function.
The Way Algorithms Work
Most platforms have created algorithms loosely based on the best working example of intelligence we have: our own brains. These algorithms work using a neural network, which functions across a series of different layers:
- First, each network has an input layer, which receives the network’s inputs
- Then, calculations are made in hidden layers using the input layer data
- After that, the result of the calculations are then stored in the output layer
Is That a Cat?
Let’s break this down with a simple example, using everyone’s favorite internet actors: cats. Let’s say, for example, we want to build a neural network that can look at an image and determine whether or not it’s a picture of a cat. The input layer in this example would be all the pixels in the image. For an image that is 1080×1080, that’s more than a million pixels, and therefore, more than a million different inputs.
In our simplified example, the hidden layers would determine if specific features are present in the pixels. Do some collections of pixels resemble a tail? Whiskers? Pointy ears?
From here, the output might contain a probability figure of between 0 and 1. The closer the number is to 1, the higher chance that the image is a cat.
If the image contains a tail, whiskers and pointy ears, the output will likely be close to 1. But if the model sees a tail and pointy ears, but no whiskers, the probability drops.
The Tie-In With Advertising
Fundamentally speaking, the neural networks used by ad platforms aren’t all that different from the example above. But instead of generating a probability of whether an image is a cat, these algorithms are gauging whether a user might convert after being served an impression.
Within the hidden layers, the algorithms are working through a multitude of combinations from the input variables. These combinations can be incredibly complex, sometimes using a weighted sum of hundreds (if not thousands) of input layer variables.
How Are These Models Built?
So, how do the hidden layers know which calculations to perform and what to measure?
The simple answer is that, at the start, they don’t. They’re completely random. Over time, though, the model learns. It teaches itself what features to look out for. There is a lot that happens to facilitate this technically, but essentially, the algorithm notes which of its calculations make good predictions, and it learns to rely more on those calculations in future.
In our cat example, once the model sees enough labeled images of cats, it can recognize which features are typically indicative of a cat (tail, whiskers, pointy ears). Similarly, with advertising, once the model has sufficient conversion data, it learns which factors — and combinations of factors — make a user more or less likely to convert.
This is also why most ad platforms feature some sort of learning phase — to reflect the time it takes while the hidden layers of the model are being tweaked.
The Short-Take on Algorithms
The algorithms that platforms use to target ads can appear complex, but the fundamental ideas behind them are easy to appreciate.
Having a better understanding of platform algorithms gives us clarity around how content and digital strategies come to life in the online environment. It also gives us a chance to better work as marketers, applying our skills in a constantly changing algorithm-driven world.
If you have more questions about algorithms or just want to chat, let’s talk.