Engineering Real-Time Predictive Analytics Delivers Millions in Additional Value to Retail Store Chain
Case Study: Data and Analytics
SITUATION & BUSINESS CHALLENGE
A retail department store chain was struggling to keep pace with advancements in data and analytics as its D&A platform had become stagnant and therefore difficult to maintain, re-engineer or automate. Product data was scattered in numerous locations, and data models were so out of date that the website’s personalization services team had not run new algorithms for several years. Furthering the issue, Data Scientists and Data Engineers worked in different groups rather than in a cohesive team. Altogether, these issues meant an inability to produce real-time predictive models, which impacted the shopping experience for customers and resulted in a correspondingly large amount of lost potential revenue.
Leadership realized the need to modernize D&A capabilities in order to remain competitive in e-commerce and keep pace with consumer expectations for the digital shopping experience, but lacked the expertise to lead the effort. The retailer turned to AIM Consulting’s Data & Analytics practice for skilled resources to help them solve their challenges.
An AIM D&A team consisting of a Principal Consultant and Lead Data Scientist began a six-month engagement with the following goals:
- Modernize the company’s D&A platform, moving toward automation and the cloud
- Clean the data and move it to a more accessible location that would be easier to update
- Create team cohesiveness between data scientists and data engineers
AIM began by assessing the company’s current D&A platform, a maze of environments and outdated tools. For example, one existing solution leveraged hacked together on-premises servers and Cron jobs for data movement. There was some modernization where the company was using Amazon Elastic Compute Cloud (EC2), but only in a pure base-level instance and not in a cluster environment.
The AIM team reconfigured Python scripts so the system could download and automate much of the workload itself, removing the need for the company’s data engineers to schedule tasks. The team also simplified the D&A architecture and moved much of the workload to the cloud. In doing so, AIM enabled the company to move to a middle ground of leveraging existing infrastructure and updating the right elements to rapidly enable new data science projects.
A Surprising Massive Value-Add
Organizational change often spurs new ideas and opportunities. As AIM started to pool the scattered data to a central location, a company IT Director wrote a sample script for the search and recommendations team that would add visual recommendations to the company’s website shopping experience. As no new algorithms had been introduced in years, the Director had no real expectations or deadline in mind for implementation.
The AIM Lead Data Scientist, having prior experience writing visualization recommendation algorithms, took on the project. The original script, written for a sandbox environment, was generating mismatched experiences like maternity clothing recommendations for non-pregnant women. AIM converted the script to production-level quality, adding another algorithm that re-ranked results based on product attributes, solving for previous mismatching issues.
The final algorithm is a combination of visual similarity of images derived from product data with actual product information. Now, when customers view a product on the shopping site, instead of seeing “frequently purchased” recommendations, they see well-matched items that are visually similar to the item they’re viewing.
AIM rewrote the script for production within a week, then the company’s data engineering and development worked the script into production. Within three weeks, the script was running in real-time on the company’s website and mobile shopping apps.
A/B testing revealed that the AIM team’s algorithm would result in a $4.16 million immediate impact to revenue over a 90-day period, an astounding value-add for a company that was already receiving great value from the original project charter of modernizing data-science capabilities.
With a modernized D&A platform and new algorithms leading to significantly improved conversions, the client is now capably equipped to take on new data-science initiatives. Data Scientists and Engineers are also now working more capably as a team, producing real-time predictive modeling for the first time in years.
With platform modernization reaching final stages and the company’s data teams working more fluidly together, AIM had additional time in the contract to begin work on modernizing the company’s A/B test procedures. Normal A/B testing within the D&A group was requiring several weeks for results, leading to lost revenue opportunities. AIM is now designing a new test platform that will enable testers to determine which models perform or underperform in real-time. Poor performers can be reassigned or eliminated, allowing more illumination and attention on high-performing models.
The client has extended AIM for an additional six months to finalize the new A/B testing platform and develop new insights for how the personalization services team can work with customer profiles.