AI to Predict Advertising and Marketing Performance

Purpose:

SMRS is an education, employer, and youth marketing business supporting clients with communication and branding. They operate either as an advisor to their clients, sharing insight and sector expertise to tackle strategic challenges, or deliver marketing campaigns from planning and research to creative design and implementation.

The project focused on the education sector where the main task was to use data science to support SMRS’ marketing decisions. Specifically, the task was to model predictions about the success and impact of certain ad campaigns. SMRS recognised that a successful project could radically change the way SMRS makes decisions regarding their advertisement design and deployment.

The team of four data scientists was provided with data from two clients (Universities) spanning circa 1.5 years of marketing ad campaigns. The main task was to parse this data and decide which types of questions could potentially be answered given the amount and quality of the data. The team had guidelines as to the types of questions SMRS would like to know the answers to, further augmented with ideas and questions derived from an initial exploration of the data.

Approach

Each University shared metric data for performance, conversion and cost, allowing to build a model that could predict the conversions of a particular advertisement given its reach (i.e. how many people saw the advertisement), the website it was hosted on, and the type of advertisement.

Initially, the team explored three different machine learning approaches: decision trees, neural networks, and linear regressions. The SMRS team needed a model that could be easily interpreted for use by non-technical team members, and so it was decided to employ a regularised linear regression model (Bayesian Ridge Regression) given its ability to simplify complex data and provide interpretable predictions.

The final model uses three key variables: reach (how many times an advertisement loaded), media (which website it was on), and ad type (type of advertisement), to predict the number of total conversions.

‍

Outcome

The value of this work is substantial for both SMRS and their clients. Having a better understanding of their data and using it to predict outcomes of advertisement campaigns will facilitate potential students to find the University that best fits their needs with more ease, resulting in more student intake.

SMRS plans to use this model and the insights as an internal tool when deciding on the deployment of their clients’ advertisements.

Given the lack of a baseline model to compare against, the success of this project was gauged by providing novel understanding of SMRS’ data and delivering a baseline prediction for them to test and improve upon.

Another finding from this work is that the existing data architecture does not allow SMRS to utilise their cost data. The team provided guidance as to the data collection could be restructured in order to exploit cost data in future modelling. This will lead to further improvements in prediction.

‍

“I cannot recommend the S2DS programme enough. The structure, our team and the support we received all made the overall experience incredible.

The project also gave our business the perfect springboard into our data science journey. Of course, the work isn’t finished, it never will be. But as a direct result of the project; we are changing how we think, make decisions, invest and ultimately the performance of what we do." - Richard Badley, Head of Innovation, SMRS

‍