Machine Learning, a major step towards predictive marketing

Yulia Spiridonova

According to Salesforce’s report [1] on the state of marketing in 2017, 60% of marketers today use artificial intelligence tools. In particular for product recommendations or to predict clients’ or prospects’ decisions.

Collectively, the tools that enable to anticipate customers’ behaviors on the basis of models that leverage data – also known as predictive marketing – are becoming key in the decision-making process of players in a variety of industries.

Machine learning techniques are gathering steam ; The field of advertising is heavily influenced by machine learning, as it is increasingly used to improve predictive marketing.


Predictive Marketing : a term that has increasingly been searched in the past 5 years according to Google Trends

It is clear that machine learning that leverages industry data in the field of advertising can handle an incredible amount of data and deliver major insights. In particular with respect to targeting options, bidding prices adjustments, placement choices for advertisements or operational improvements when choosing a media mix. This should not come as a surprise : by leveraging historical data from a large number of users, it is possible to build a model that predicts the behavior of similar users. And as a result, deliver the optimal message at the optimal time.

Using machine learning algorithms allows decision-making platforms to deliver much more accurate recommendations : by applying algorithms to large-enough data sets, one can generate a prediction that will take into account all specific aspects of those data. Those technologies let marketing teams make simulations and predictions to better understand their media mix with no need to have specialized training in media mix modeling.

Those predictions can be used to tackle a variety of problems. Below are a couple examples of process automation in predictive marketing that are made possible through machine learning :


Increased segmentation precision and targeting improvement


Users today are very demanding when it comes to the quality and loading speed of the content they consume, whether on smartphones or desktops. Advertisements must comply with those requirements or risk losing their visitors’ attention. Training algorithms to match an advertisement with a group of persons to optimize the number of resulting interactions is one of the main challenges of machine learning in digital advertising. Using the navigation history of users, it is possible to predict the view-through rate of a specific video or click-through or purchase rates.

DCO (Dynamic Creative Optimization) is a technique that personalizes banners in real time. It is used in the programmatic industry to improve conversion rates. Besides adapting the advertisement on the basis of basic parameters such as the user’s location and time of day, this technique makes it possible to pinpoint the specific preferences of users in terms of colors or graphical elements. This technique makes it possible to display a banner that is specific to the user’s preferences. It becomes all the more powerful as image recognition algorithms improve.

As predicting intent (for instance, what is the probability that a consumer look into buying a car in the coming month ?) on a user-by-user level also becomes more accurate, algorithms will recognize it and display the image directly connected to the product that the user intends to buy. This will become the perfect combination when displaying advertising messages to web users.

Bidding Price Optimization

Using machine learning algorithms to determine bidding strategies in display or search advertising is commonly used today. Given a limited budget, campaigns are run according to the recommendations of algorithms that optimize placement with the group of consumers that is potentially the most interested. Human intervention becomes limited in purchasing transactions that happen in a split second, and use of the most advanced algorithms turns into a major competitive advantage when using a programmatic platform.


Multichannel budget allocation


Using historical data, it is possible to create accurate simulations of the business impact of budget changes. Will the reallocation of media budget from TV advertising to online video impact my ROI ? How will terminating social advertisement impact my revenue ?

Forecasting media budgets remains a complicated endeavor for traditional algorithms, since many factors come into play, such as the interaction between various channels, the price of current media purchases, seasonality or the current media mix. Machine learning makes this challenge easier ; However, one should pay close attention to the volume of data available during the learning phase. With a richer media mix and an increasing number of parameters, the data volume required to obtain a high probability of an accurate prediction increases in parallel.


Predicting client loss and client’s lifetime value


Traditional client loss models are quite accurate, but their predictions often come too late for an effective action to prevent the loss of the customer. When using machine learning, it becomes possible to determine weeks in advance the behavioral insights that predict customer disengagement.

As a result, one can adapt marketing actions as soon as the first signs of customer disengagement appear, pick personalized marketing actions (such as a new message or a targeted offer) and reduce churn. One can also pick marketing actions that will increase revenue on the basis of a predicted customer lifetime value, since algorithms can determine whether it makes more sense to retain an existing customer or invest in recruiting a new customer.

Although predicting these various issues becomes very accurate, human intervention remains necessary to explain the observed phenomena. Algorithms use predictive models by leveraging pure statistical correlations but are unable to explain the causal link between events.

However, the time gain from using those tools is a major advantage for marketing teams who use the insights from those algorithms. Moreover, connecting several tools will activate the appropriate drivers according to algorithm recommendations : based on budgets, the Data Management Platform will activate the recommended targeting for the right audiences, DCO placements will be bought at the optimal prices via programmatic platforms, and the right message will be sent according to the customer’s lifetime value….. A picture that no longer seems futuristic in 2018.