What is Predictive Marketing?
Predictive marketing is a modern practice of marketing automation that predicts the actions of your contacts based on information gathered from their profiles and history of your store and interactions with your company.
With the programmed end of cookies that allow to identify the actions of visitors on the websites of e-commerce companies, the current retargeting will be less and less relevant because the visits will not be traceable anymore (except for Google).
To continue to send hyper-personalized emails to your customers, you will have to predict the expectations and desires of your customers rather than reacting according to the visits on your site.
To make a prediction, it is necessary to collect a lot of customer data and the processing of this data requires a lot of computing power. It is therefore unthinkable to process this information manually. But AI is there to process it.
To be more precise, predictions are made by machine learning algorithms (a subfield of AI). These algorithms can classify a customer's activities, identify similarities with other customers, predict an action and modify their operations based on the actual results of the prediction.
What is the difference between AI, machine learning, deep learning and reinforcement learning?
It is common to confuse"artificial intelligence" (AI) with"machine learning". Worse, some web companies falsely claim to do artificial intelligence.
Artificial intelligence refers to the ability to design computers that have the same behaviors as humans. It is a very vast and complex technology for companies because the human brain is difficult to understand and therefore difficult to reproduce.
It is therefore presumptuous at this time to claim to be doing AI in digital marketing.
Among the different fields of AI, machine learning is the ability to create algorithms that use a large amount of data to predict events. This predictive technology is very useful in the field of digital marketing because it allows you to anticipate consumer behaviors based on known information about their profiles and their history of actions on your website.
The subject generates a lot of fantasies when it is simply a matter of more or less intelligent sales programs (algorithms).
These algorithms have the ability to learn, in a more or less guided way. There are 3 main families of learning :
- supervised learning, which uses categorized data sets,
- unsupervised learning, where the algorithm learns by itself by creating "paterns" from the data,
- reinforcement learning, where the algorithm improves according to the interactions of the environment;
Kiliba uses many algorithms to predict user behavior and send hyper-personalized emails:
- Classification: assigning a category to each customer based on the user data collected (the data is known and tagged, this is called supervised learning)
- Clustering: assignment of a customer category based on observed actions and results (unsupervised, data is not tagged in advance)
- Recommendation: product recommendation based on customer visits and purchases.
These algorithms use logic coded by a developer to make a prediction of behavior.
Deep learning is a subset of machine learning algorithms. Unlike previous techniques, these algorithms learn by themselves how to work to best make their predictions and improve over time. They try multiple combinations from the raw data and compare the results to refine their operating logic. In reality, nobody is really able to explain the logic of the algorithm created, we are just able to measure that it works. This is the beauty of the technique and the concern about its control.
Reinforcement learning is a distinct subset of algorithms based on the confrontation of these predictions with reality. By "rewarding" correct predictions and "punishing" incorrect predictions, this type of algorithm is able to improve over time by modifying its logic to optimize the number of correct predictions.
What are the advantages of machine learning?
These algorithms allow you to recommend products from an online store based on consumer profiles. These recommendations make it possible to create hyper-personalized emails, where each email is unique and the content corresponds to the buyers' expectations.
For example, Kiliba optimizes email conversion rates through multiple analyses including:
- Predictive product recommendation based on customer visit and purchase history
- Identification of complementary products to a purchase
- Identification of gender (male/female) by clustering when the customer does not provide this information
- Analysis of customer satisfaction (level of email harassment) to reduce if necessary the marketing pressure (Kiliba sends less emails to a customer if the algorithm predicts that this recipient feels harassed)
What are the advantages of deep learning?
Machine learning requires coding an algorithm's operating logic, thus relying on the development team's mastery of the mental process of a customer's purchase decision. Human behavior being complex, it is by nature difficult to understand and anticipate.
Coding an algorithm requires regular measurement of the results obtained to modify the algorithm's logic and retest. Deep learning can be used to go even further in product recommendation. Indeed, the deep learning algorithm will decide itself on its decision making logic and will improve itself over time.
What are the advantages of reinforcement learning?
Reinforcement is an approach that takes into account the interactions of the environment.
In the case of email automation, reinforcement allows to automatically improve future recommendations by learning from the actual behaviors of email recipients: email opening, click, store visit, purchase.
If the actual behaviors match the predictions, the algorithm has no reason to change because it is effective. But if the email recipient doesn't click on the proposed products, the algorithm will try to improve for the next recommendations by changing its working logic.
Read more on the same topic
How to set up a marketing automation software?
How to make a successful email campaign in 5 steps?
What are the best triggers for marketing automation workflows?
Optimising the performance of email marketing