How sticky.io Recovered $8M Per Month Using No-Code Automated Feature Engineering

The analytics team at sticky.io had a mandate to apply Predictive Analytics to optimize these flows to achieve net benefits to customers. Justin Shoolery, Director of Data Science & Analytics, wanted to quickly analyze customer data and determine how to treat customers and their transactions in more profitable ways. The analytics team knew that Predictive Analytics, using Machine Learning (ML) was the ideal technique to spot specific patterns from the data from the hundreds of millions of transactions they were seeing. However, with so many partners and customers in the ecosystem, the amount and other kinds of data were simply overwhelming. The team’s biggest challenge was how to make sense of data since they were discovering new data tables and new elements every day.

sticky.io’s team had a good idea about which data they should be looking at, the things that might be correlated, and what they could predict. But actually extracting the data out of the data warehouse in a format suitable to run a statistical model took quite a bit of effort. The added effort was exceptionally high when Justin ran a model and realized a predictor didn’t work well and needed swapping. Given the database’s complexity, Justin found himself going back to the drawing board, writing new queries to join new data.

Justin and his team evaluated several Predictive Analytics and Machine Learning platforms, including BigSquid and DataRobot, but found that the amount of data prep needed to get value would be too high. According to Justin, “I was spending 95% of my time wrangling data. I just didn’t see value in automating the last 5% of my process when I need more help in making sure my data pipelines are working properly.” The structure, complexity, and nature of the data used by sticky. io meant that generating and evaluating value was more critical than merely automating the process of developing predictive models. dotData’s unique combination of no-code ML automation software and full-service approach to customer onboarding offered a unique combination.

dotData made the whole process convenient and easy. The team connected dotData to the database and moved quickly from ideas to tests — like testing if processing transactions on a particular day may be more favorable. According to Justin, “Before dotData, querying data, finding the right format, moving data to a new hosted cluster and testing models could take more than a month — only to find out the model did not work. Now I can offload most of that work and just focus on uncovering patterns and finding viable models. With dotData, we feel that finally we have a solution that is very promising.”

What mattered most to sticky.io was dotData’s focus on accelerating the process of combining information from multiple tables and columns through automation. The Feature Engineering functionality ultimately made the difference for Justin’s group. Feature Selection, particularly on complicated databases with many tables and joins, accelerated data science workflows and enabled the data team to explore millions of features across thousands of table columns in record time. The data science group is very excited about using dotData, and are hiring more members to the team to leverage the dotData platform.

The analytics group has already pitched seven other use cases, and more are in the pipeline. The team has many ideas about how ML-powered predictive analytics will help sticky.io optimize the way they funnel transactions through their platform and drive insights for customers. Among the models Justin’s team is considering are predicting revenue, identifying customer churn candidates, possible complaints and even modeling which clients might be better suited for specific product promotions. In the end, the possibilities are endless, and Justin and his team are confident that dotData is the ideal partner to help them explore ideas and build valuable models for their rapidly growing business.