top of page
  • Writer's pictureStratzie

How Image Recognition Classification Method Can Help Lift Brand Creative Standards

Updated: Jul 16, 2020

Machine learning modelling is known for predicting performance of marketing activities, however, this is a story of how we developed an image classification model that predicts the performance of Instagram advertisements and provides an opportunity to uplift creative standards. “Will consumers engage with this creative element?”, “and to what extent is the emotional connect?” were the questions in the minds of our researchers?

We selected a tea brand's Instagram image adverts for this research. And we felt that there was scope for improving the engagement rate (viz. likes, comments and saves) of Insta adverts. We set three research objectives: 1) To identify high performing Insta adverts. 2) To build a profile of high performing campaign elements. 3) To explore building a database of creative campaigns, especially those that have an emotional content. This brand would execute different type of campaigns – festive, new product launch, new offer, sales offer etc. Our initial scan showed that the engagement rates varied significantly between different campaigns.

We developed a 3 Step C-P-M Framework or simply known as Classification-Prediction-Measurement Framework.

In the 1st stage of classification, historical Insta adverts dating back to 2018 and 2019 were classified using natural language processing methods and image recognition methods. Three classes viz. High Performers, Medium Performers & Negative Performers were formed based on factors like a presence of a primary image, headline text, number of likes, offer prominence etc.

In the 2nd stage, new Insta adverts were checked against i) the above classes and ii) automated eye tracking model. Prediction scores revealed the class to which the new Insta advert would be tagged. This framework allows creative time to revise the new adverts.

In the 3rd stage, upon release of Insta adverts, the performance key performance variables like no. of likes/shares are correlated with in-store KPI viz. average sales per square foot in order to understand the impact of lift on in-store KPI.

Also, due to application of automated eye tracking and machine learning models client teams are able to detect creative concepts that has maximum emotional impact on consumer emotions. This enables in lifting of creative standards in what we term it as ‘Creative Idea Leapfrogging’.

10 views0 comments


bottom of page