Objective: To study the relevance of Instagram posts in the construction of forecasting models for the variation of daily sales revenues for retail companies in the beauty and cosmetics sector. Methodology: Time series of daily sales between the years 2017 and 2019 of 10 retail companies in the beauty and cosmetics sector were considered. Methods based on machine learning were used and the forecasting models were increased with numerical variables from the official profile of the company, from the posting made by the contracted digital influencer and the characteristics of the images posted by the digital influencer were included in the models. Relevance and Originality: The study is innovative, as it goes beyond qualitative reflections on the theme and provides empirical evidence regarding the impacts on forecast accuracy from the inclusion of social media variables. A data fusion strategy (numerics and images) was also presented to forecast daily sales of retail companies in the beauty and cosmetics sector. Main results: The models proved to be efficient in forecasting and the importance of the likes and engagement variables reinforces the idea that the identification and social reference generated by the ID are important aspects in the purchase decision process. It was found that the images are responsible for adding exclusive attributes that help in forecasting and understanding the patterns of the sales series. Theoretical and methodological contributions:The study showed in a promising way the efficiency of methods based on machine learning in forecasting sales from Instagram data, especially with regard to the incorporation and extraction of image data.