What is Machine Learning & How is it Used in Advertising?

Machine learning is a subfield of artificial intelligence (AI) and computer science that focuses on the use of data and algorithms to mimic human learning. It aims to improve its accuracy over time by imitating how humans learn.

Every part of a client or customer’s experience—from correctly locating relevant goods to providing essential moments of delight are augmented and even customized using machine learning. Airbnb, for example, applies machine learning across its business, including search autocomplete, recommendations, pricing suggestions, marketing, fraud detection, payments, customer service, auto-categorization, and amenity predictions.

In simple words, machine learning is a scoring method. Machine learning is mostly used to teach machines better outcomes for moments (like the recommendation engines on Google) by testing, analyzing and learning to make predictions.

Why CatapultX Knows Machine Learning

CatapultX was nominated by The Drum for Best Use of AI and Machine Learning in 2021. Not only is CatapultX’s On-Stream AI technology award-winning, the results have been impressive for both advertisers and publishers. Video publishers have seen up to +25% revenue growth on their video monetization strategies and advertisers have seen higher click through rates and conversions from On-Stream ads over pre-roll ads. As CatapultX’s machine learning algorithms continue to learn, they’ll become more able to analyze and predict content that drives success. Read on to learn more about machine learning.

What’s the difference between Machine Learning and AI?

Machine learning is a part of AI. Machine learning is the ability to teach a computer to make decisions. AI is broader since it also involves a system's ability to intelligently perceive data (e.g., natural language processing or voice/image recognition) or to control, move, and modify things based on learned information, whether it be a robot or another connected device.

Types of Machine Learning at a glance

Supervised Learning

Supervised learning, also known as supervised machine learning, is the process of training a computer algorithm to identify or predict outcomes by using labeled data. The model adjusts its parameters until it has been properly fitted based on incoming data. This is because of the cross-validation procedure, which is used to guarantee that the model does not overfit or underfit. In addition, supervised learning is a type of machine learning that guides humans to make decisions. It's used by businesses to solve a variety of real-world issues at scale, such as classifying spam in a different folder from your inbox. Unsupervised Learning

Unsupervised Learning

Unsupervised learning, often referred to as unsupervised machine learning, is the process of analyzing and clustering unlabeled data using machine learning techniques. These algorithms reveal hidden patterns or groupings in data that may not be seen by humans. It's ideal for exploratory data analysis, cross-selling ideas, customer segmentation, and pattern recognition. Principal component analysis (PCA) and singular value decomposition (SVD) are two frequent methods for reducing the number of characteristics in a model. Other unsupervised learning algorithms include neural networks, k-means clustering, probabilistic clustering methods.

Semi-supervised Learning

Semi-supervised learning is a hybrid form of supervised and unsupervised learning that combines the best features of each. It employs a smaller labeled data set to help classify and extract features from a larger, non-labeled data set during training. Semi-supervised learning can be used to tackle the scarcity of labeled data (or not having enough funds to label all the data) that may plague supervised learning.


Reinforcement Learning

Reinforcement learning is a behavioral machine learning technique that works similarly to supervised learning, but the algorithm isn't trained on sample data. This model learns by trial and error as it goes, getting better with each attempt. The ideal solution or policy for a given scenario will be reinforced if a sequence of successful outcomes is observed.


How is Machine Learning used in advertising?

As marketing becomes more data-driven and reliant on machine learning algorithms, businesses are looking for ways to employ these technologies to improve their advertising efforts. Recent advancements in AI technology have made it possible for machines to learn from and make decisions based on large amounts of data. This has led to the development of various machine learning models that can be used in marketing, such as predictive modeling, natural language processing (NLP), and deep learning.


Advertisers don’t like to just throw things out and see what sticks anymore. Machine learning helps advertisers to forecast or predict performance of advertising campaigns. Many DSPs or ad servers across the internet include a modeling or forecasting tool - or if they don’t they usually provide an estimate of the daily reach or daily results that one could see from an ad campaign. This is achieved through machine learning.


As we stated before, Google uses a recommendations-based machine learning approach for predictive text, but ecommerce advertisers everywhere are using a recommendations based machine learning algorithm to help serve product ads. By learning and predicting which types of users will be interested in which products, the product ads can include the types of products that that person would be interested in - there’s no point in showing Joe the painter a set of golf clubs.


One of the most important aspects of marketing is knowing your audience. The more data you have on people, the better you can customize your message to them. Machine learning has helped marketers with this by providing new ways to identify and target audiences that are most likely to convert into customers. One example of how machine learning is used in advertising/marketing is through targeting ads based on probability of purchase. By using machine learning, advertisers can select targeting preferences based on the likelihood that someone will purchase a product in n amount of time.

The issue behind this is that this may sometimes result in an invasion of user-privacy. For example, Target fell into hot water when an article was published about their machine learning algorithm predicting a teen’s pregnancy. While this story may or may not be completely accurate, it still opens up the question on whether our data should be shared or used for marketing purposes to that level. It’s most  likely that the girl purchased a pregnancy test, maybe some nausea pills later, and the machine learning put those two data points together to sign her up for the next maternity mailer.

Contextual Targeting

The machine learning used by CatapultX uses moments to predict sentiment and likelihood for interaction. The algorithms intake data on the content with ad spaces available - what is the video about, what objects are in the video and how might one feel about the moment within the video. By using machine learning, CatapultX’s proprietary algorithms can match the moment up with an ad that is related to the content and more likely to garner interaction or ad-recall. Since learning is continual, the more impressions that the algorithm sees, the more likely it is to get it right.

What is the future for machine learning in advertising?

The best marketers and advertisers will be generating insight more quickly with machine learning in the future. This means that marketers can test and iterate even faster in the near future. Advertisers and agencies need to plan for numerous versions of creative early in the campaign and reserve budget and time for side-by-side testing so that brands can get the most out of AI insights even faster.

We’re excited to see machine learning continue to be integrated across marketing and advertising strategies in the years to come.