Over the years, marketeers have had to keep up with the advancements in technology and changes in human behavior. Although the change is getting costlier every day, not changing can prove to be even more costlier.

It is expected that spend on Marketing activities will cross the mark of $560 Billion in this year alone. Further, more than 30% of the budget goes into content creation.


The Problem

In a survey of 1,000 marketers worldwide by Rakuten Marketing, respondents estimated they waste an average of 26% of their budgets on ineffective channels and strategies. In another study, it was found out that Creative Quality/Design is 4 times more important than Media Plan Quality in driving sales.

A critical component of success in the current environment is effective strategy and the elimination of wasteful spending. Brands need to design with conversions in mind if they want to stand out from millions of marketing creative agencies and competitors. With the advancement in technology, marketeers have been deep diving in finding strategies based on audiences, platforms etc.

Using the progress made in deep learning (image processing) and machine learning (prediction and recommendation), We (Team: The Mean Squares) devised a solution that not only unlocks HOW a marketing creative or campaign image performed in a way but also helps in predicting the NEXT. Just to brag here, this solution won the Runner-up Position at Publicis Sapient EXPO’19.


The Solution:

The solution is designed with two Parts:

I: Creative Analysis: Unlocking HOW?

II: Creative Performance Prediction & Recommendation: What’s NEXT?


Let’s talk about them in detail:

I) Marketing Campaign Creative Analysis: Unlocking HOW

Before getting into the details, try answering the following:

Image Source: Google
Image Source: Google

This part of the solution helps marketeers measure the impact of particular attributes in campaign creative. It can help identify top combinations of attributes or even test our hypothesis. For example, we can see the impact of using a model/celebrity in our creatives; If we see it doesn’t benefit the performance, we can simply save some cost there itself. Another example, we can measure the importance of stress keywords or Call To Actions (CTA) in driving engagement for campaigns ads and creatives.

The Process:

1. We take data from one of the platforms that a brand is using for running campaigns.

2. The data is pulled in two parallel stream, one containing the creative performance and other stream passed through Google Cloud Vision API to identify features/attributes present in the creatives

3. This data is then cleaned and joined.

4. We use this data for generating data driven insights

The Process

The insights after the analysis are then used by the designers for data driven and intelligent designing.


II) Creative: Performance Prediction and Recommendations: What’s NEXT?

This part of the solution helps a designer in preparing for the NEXT. The data extracted and transformed in the first part is used to create a machine learning model which helps in prediction of Click Through Rate (CTR) performance of an image. A designer can input a design that he has created and find an approximate prediction of CTR.

The solution also identifies the impact of attributes in a creative’s performance and gives recommendations for optimizing the performance of creative.

The Process

Below is the video, our team, “The Mean Squares” presented at Publicis Sapient’s EXPO’19 which won the Best Promotional Video Award.

Special Thanks to Gaurav Thakur (Publicis Sapient)

And finally, a sweet picture of our team, The Mean Squares after winning the Runners Up position

From Left to Right (in black): Jatin Pasricha, Kulbhushan Verma, Gaurav Thakur, Utkarsh Tripathi, Myself (Sumil Mehta). Binay Chandra was missed

Leave a Reply