SOLVER: Customer Lifecycle Management
How to identify high-value Millennial customers considering the lack of available and useful data on that unique generation.
Fuse bank-customer data with TROVE Data and apply machine-learning models to the new data superset.
1. Key insights:
2. TROVE machine-learning models used three months of banking-activity data to accurately predict which customers would be “high value” at 24 months post-open with 85% accuracy.
There’s something about Millennials. While often poked fun at for their lifestyles and work habits – albeit typically by an older generation, the Boomers – Millennials have something that is no laughing matter and that every business wants – numbers. At 74.5 million strong (according to Pew Research), the generation born between 1981-1997 is due to surpass Boomers in the U.S. population in 2019.
That’s a lot of people with a lot of potential buying power, but getting to know this group, and the individuals who comprise it, can be tricky. While some Millennials have struck out on their own, leaving the beginnings of a data trail useful to marketers, many others have yet to leave the nest, blending into their parents’ households where they are harder to see, parse and track.
The opaqueness of Millennials is a major challenge for companies looking to attract the best customers among them, those with the highest potential to engage with their brand, stay loyal to it, and profitability transact with it over their lifetime. A large regional bank wanted to understand its Millennial customers better to see if there were behavioral characteristics it could identify to better predict successful long-term banking relationships. It turned to TROVE Predictive Data Science for help.
“We wanted to see if we could help identify behavioral characteristics that could act as reasonable indicators of Millennials being a good, long-term, profitable customer for the bank,” said Dr. Adam Stotz, Chief Technology Officer, TROVE. “In the course of developing approaches to solve this problem, including using machine-learning models and our vast repository of consumer demographic and psychographic attributes called TROVE Data, we learned a lot that may help sharpen the bank’s targeting of Millennial customers.”
Putting Prediction to Work
Working with the bank, TROVE put together a Proof of Concept (PoC) focusing on a select geographical region’s Millennial segment. Exploring the bank’s data and TROVE Data, TROVE quickly came to a realization that delivered one of the PoC’s key insights: there wasn’t much data to work with.
“We knew going into the project that finding meaningful data on the majority of Millennial customers was going to be a challenge, and our initial assumptions were quickly verified,” said Dr. Kamil Grajski, Managing Director, Data Science, TROVE. “But that finding quickly led to a more meaningful discovery. By delving into our data set, we found that there were really two groups of Millennial targets – the ‘independents,’ those living outside their nuclear family household, and the ‘dependents,’ who were still at home.
“Furthermore, we found through our own data modeling that behavioral characteristics of the family household – not just of the individual Millennial – could be very helpful in predicting individual banking behavior.”
This was an important finding for the bank, a novel one that offered the opportunity to tune their segmentation and targeting approach. Simply put, Millennials were not a homogenous group. Moreover, the ‘independents’ whom they had traditionally targeted were far outnumbered by the ‘dependent,’ harder-to-know group, essentially revealing a large untapped market.
Having uncovered this new target group hiding in plain sight, TROVE went on to demonstrate the power of its advanced machine-learning models. Using just the first three months of a Millennial customer’s banking activity data and the 650+ demographic and psychographic characteristics in TROVE Data, TROVE was able to correlate specific attributes with higher and lower performance, predicting the value of an account at 24 months from just the first three months of banking activity data with 85% accuracy.
“Overall, we found that those Millennials who led a multi-faceted financial life or were members of a relatively financially sophisticated household – demonstrated by activities such as purchasing auto insurance, investing in identify theft prevention programs, and even reading magazines related to financial health – outperformed those who did not or were not,” noted Grajski.
“But the results didn’t stop there,” added Stotz. “Through this work, we were able to demonstrate the added value of an ‘ensemble’ machine-learning approach, complementing the regression and other modeling techniques the bank traditionally employed. These are new tools and techniques that banks are just starting to get their arms around and which show great promise when mining very rich data sets like the ones we used on this project – a fusion of bank data and TROVE Data."
Banks often rely on the same five variables in their personas, a.k.a., “The Common 5”: age, occupation, household income, marital status, and educational level. By being able to mine a much deeper database and broaden the exploration to include demographic information such as “matching the profile of a purchaser of identity-theft services,” TROVE delivered novel insights to fuel the bank’s expert marketing teams.
“The bank appreciated that we proposed new attributes for consideration in their customer value calculations,” concluded Grajski. “Not only did they find our ‘out of the box’ thinking illuminating and useful, it also hit the mark in another way: the traditional demographic determinants, such as age and household income just aren’t as statistically significant in specific geographies where those values aren’t that varied. Our findings weren’t just novel, they were incisive and led to much greater prediction accuracy.”
The bank’s initial foray into advanced predictive data science with TROVE has been a success. TROVE:
Combined, these insights were applied to a successful PoC that accurately predicted 24-month performance from the first three months of banking activity data with 85% accuracy.
Armed with these initial insights and successes, the bank is currently in discussions with TROVE to deepen its exploration of Millennial groups and data, as well as partnering with the company more broadly to explore ways it can infuse predictive data science throughout its business.
See the potential of a large Millennial group that had been "hiding in plain sight."
Develop useful persona characteristics beyond age, income and education level.
Gain experience using new modeling tools and approaches.
Increase prediction accuracy.