We've already discussed supervised and unsupervised learning in advertising, but there is a third of the big three fundamental machine learning paradigms: reinforcement learning. In this blog, we'll discuss what reinforcement learning is and how it can be applied to marketing.

What is Reinforcement Learning?

Reinforcement learning is a type of machine learning that is concerned with how software agents ought to take actions in an environment so as to maximize some notion of cumulative reward. In reinforcement learning, an agent faces a series of decisions. For each decision, the agent receives a numerical reward or punishment that reflects the value of its action. The goal of reinforcement learning is to find a policy that tells the agent what action to take in each situation so as to maximize its total reward over the long run.

There are two main types of reinforcement learning: positive reinforcement and negative reinforcement. Positive reinforcement occurs when an agent is rewarded for taking a certain action, while negative reinforcement occurs when an agent is punished for taking a certain action.

How Reinforcement Learning is Used in Advertising

Reinforcement learning can be used to train agents to perform all sorts of tasks, from playing video games to flying drones. It has also been used in marketing applications such as optimizing website design and online advertising campaigns.

One potential use of reinforcement learning in marketing is to optimize website design. A reinforcement learning algorithm could be used to test different versions of a website and see which version results in the most customers taking the desired action (e.g., clicking on a buy button).

Another potential use of reinforcement learning in marketing is to optimize online advertising campaigns. A reinforcement learning algorithm could be used to test different versions of an ad and see which version results in the most people clicking on the ad. Here are a few ways reinforcement learning is used to optimize campaigns:

• A/B testing ad creative through performance measurement
• Deciding advertising placement
• Optimizing for times to serve ads
• Determining how much to bid on ad placements
  

Typically, an advertising algorithm is built from multi-agent models of reinforcement machine learning - meaning that the algorithm is trained to optimize and serve the best ads to the best person and place through positive enforcement in the form of clicks and conversions.

In particular, reinforcement learning is only well suited to problems where there is a clear notion of reward and punishment. For other types of marketing problems, other machine learning paradigms may be more appropriate.

How Does CatapultX Use Reinforcement Learning?

CatapultX uses reinforcement learning to improve the performance of online advertising campaigns. We use reinforcement learning to test different versions of ads and determine which version is most effective at getting people to click on the ad. The most innovative way we use reinforcement learning is to learn what types of moments within videos generate better brand lift or engagement. For example, a clothing company selling sweaters with quirky designs could find that the best engagement for their on-stream ads comes from moments in videos that include cats, which would trigger our algorithm to find more cat moments.

The jury is still out however whether cat moments for a quirky sweater company means that cat owners are quirky, but we'll let the machine learning figure that out.

Let us know what you think about reinforcement learning and what topics are of interest to you in the future!

‍