The problem

One of the biggest clothing retailers in the United States wanted to understand customer behavior and sentiment towards all their email campaigns and tie it back into their sales KPIs. With a primary online presence and a strategy of engaging with the customer in sequence, which ensures a loyal customer base for the retailer, it’s imperative that it has a strong connection with the customer.

With a constant focus on customer acquisition and engagement obsession, the retailer asked Cognitive BI to see if we could answer what type of ads would work, and how we can set up a framework to measure the impact of these marketing mailers with online sales figures.

The cognitive solution

The problem was a complex one, with the traditional click-through-rate almost saturated. There was a dire need to look at a new innovative approach to understand what triggers a customer. The business owners and creatives working in silos was another barrier that needed to change. A three-stage solution was implemented to tackle this problem:

1. Fusion of non-existent data with marketing data

• Extract meaningful data points from marketing creatives that actually have an impact on customer behavior – color schemes, textual content, infographics used, branding in messaging, classification of images etc.
• Fuse the above data points as part of current existing marketing, clickstream data, and customer profiles
• Framework for segmenting customers based on their preference of creatives (psychological human bias), customer demographics, and product preference segments

2. Data-driven creative guidance

• Reinforce creatives using data-driven insights from the above framework
• Segment creatives based on customer segments to pick and match relevant content dynamically

3. Measurement and scale

• Marketing effectiveness playbook to measure incremental lift from a segment and month level KPIs
• Deployment strategy to measure effectiveness from a baseline using an A/B strategy across geographical regions

The cognitive design

The solution was built on the AWS to run computationally expensive neural network algorithms for image and textual feature extraction, and integration with clickstream and RedShift data sources of marketing data.

The text extraction module was a trained Tesseract OCR engine to extract the words found in the banners with company-specific fonts, so as to extract information on the type of campaign, offer, and language. The object extraction module was a trained SSD and YOLO based models to identify different objects and icons present in the incorporated banner/image. The output from the text and the object detection modules were processed to create variables for the driver analysis model. The raw text data was analyzed to extract the banner category, offer type, special word presence, while the object detection solution was used to yield the presence of important objects.

Key insights

The implemented solution enhances the current methodology with forward-looking views for marketing strategies.

• Increased customer acquisition rate by 1.2% in two consecutive quarters, boosting online sales specifically in the clothing section
• Removal and optimization of over 1000+ non-performing creatives that had minimal/zero impact
• Insights on what creatives work for email campaigns – logo branding in emails, choice of warm colors, choice of % discount vs. $ amount, beaches as opposed to mountains

This solution was a significant step towards an entire business process managed to be data driven strategy – from content creation to measuring sales performance.