Using Machine Learning to Connect Brands with Influencers

Sammanfattning: With the increase of social media users, marketing through social media channels is becoming more and more important for large and small businesses around the globe. This form of marketing can be done in different ways. For example, through ads on social media platforms or by using influential opinion leaders (influencers) to reach out to a large number of customers. It can be difficult, especially for smaller brands, to navigate through the vastness of influencer marketing and find the right partner. This master thesis investigates how a process for brands to create a deal for a product/service they want to promote and how machine learning can be applied to recommend suitable influencers. Finding the right match is necessary for brands looking to take advantage of the rise of social media marketing. The thesis includes research of potential input and output for a machine learning algorithm, suggestions regarding which machine learning type and machine learning algorithm could be used, and presents a design of the deal creation process. The master thesis is conducted with the support of a startup company called Splick. Splick connects brands and influencers with an easy-to-use platform. Collaborations through Splick can be done with the help of affiliate links that track everything from clicks to sales. Every partnership is different since it depends on the brand, the influencer, and the type of product/service the brand is trying to promote. This makes it necessary to take a wide variety of metrics and benchmarks into consideration. The thesis found that a combination of metrics from the participants of the deal and the deal itself, metrics from various social media platforms where the influencer is active, and benchmarks from Splick’s affiliate link service would all be useful. The suggested output of a machine learning algorithm is common points of interest found in partnerships with similar brands and influencers that occurred in the past and turned out to be successful. Out of the machine learning types and algorithms analyzed, an unsupervised learning approach using Neural Network-based clustering seemed to be a viable way to proceed. Further research would be necessary regarding the presented machine learning approach to create a more clear plan for development and implementation.