Campaigns, product launches and social media trends: Retail use case

In this article I will mainly focus around one problem statement which I took care in one of the consulting assignments.

Business case (Industry: CPG)

The requirement of the client was to come up with a social media strategy aimed to clearly chalk out an end-to-end framework and a roadmap for activation of relevant levers to understand consumer voice in a more effective manner.

For ease of discussion, I have simplified the objective and divided into 3 parts:

·       Campaign performance

·       Product launches

·       New trends

While working on these kinds of assignment, invariably you have to come up with multiple strategies and frameworks depending upon your past experiences, Existing data strategy and architecture (if any!) and Industry best practices and benchmarks.

Strategic Roadmap & Execution (Consulting perspective)

This may sound straightforward but there are lot of dependencies like client’s social media strategy, existing data strategy. We need to have a clear view of various strategic objective and the levers to enable the same, entailing framework which would enable the use cases pertaining to campaign, launches etc.

Before decoding levers for campaign performance, we need to figure out the right KPIs and frequency of reporting to start with. Campaign may run offline or online but the feedback can be obtained on social media. If campaign running of client’s owned media assets it is much easier to obtain the data and come up with reporting strategy. However, using owned media assets reach of client’s initiatives are limited to very small target segment.

In fact, in case of product launches and new trends identification we better need to understand what people are talking in social media as a whole, not limited to client’s owned media assets. This is a hard nut to crack, in fact all the social media listening tool are not state of art when you talk about this use case.

It is very important to keep focus away from tools and technology when one is starting to carve out a strategy. Tools and technologies are simply enablers, that’s it. They are equally important but there as to be an order in which problems are solved.

Technology (AI/ML)

Once we are done with the high-level framework including data strategy, we can talk about activations of relevant use cases. From technology perspective, we can divide the problem into three parts text, images and video.

Considering we would be scraping publicly available data from social media platforms, we need to have data strategy capable of handling terabytes of data. Considering the amount of data, its pretty much straightforward to decide algorithm based on neural nets or standard ML algos like Logistic regression, na?ve Bayes etc. we can use this to broadly solve below use cases like:

·       Sentiment

·       Contextualization

·       Segmentation

We can use feed forward neural nets, LSTM (long short-term memory network) or even logistic regression to execute these use cases. Traditional language models are frequency based and doesn’t account for context. For ex: usage of book as a book or in context of booking a flight. LSTM does solve this problem but needs lot of tuning and data for the same.

For images / video, we can get this data from social media platform and can identify objects in an image and come up with a similarity-based dictionary model which would classify image into multiple categories depending upon the content. For ex: we can have an image uploaded by a FB user with him enjoying a product along with friends during dinner. This information can be extracted from the image using various models like YOLO, SSD etc. It is preferable to use transfer learning in case of images considering they have been trained on millions of images and are better equipped to understand the patterns in images.

Once we are done with text and image analysis we can club both the information and come up with consolidated data set which would then be used to solve the mentioned use cases.

This is merely a high-level view / approach to solve these problems but it’s always better to start with solid framework and build use cases over it.