As smart as they are, fraud detection algorithms fall short when hunting down credit card theft.
They red-flag potentially suspicious purchases such as, say, an order for 10 shirts in size L. But they typically miss subtler techniques used by fraudsters, such as an order for two shirts in every size, or a delivery address that’s an unoccupied home for sale.
How do we know card thieves’ tactics?
We asked them.
While big data excels at sketching broad outlines, so far it has been unable to thwart rapidly growing credit-card fraud. Adding human insight to the mix, however, fill in the details.
To explore how the one-two combination of big data and human insights can level the playing field, we conducted a proof of concept and applied the pairing to credit card fraud.
Human insight is the job of so-called thick data, the kind of qualitative, real-world details we uncovered during our conversations with fraudsters. When overlaid on big data’s bold strokes, thick data helped spot telltale signs of card theft — and turned up potential new weapons to use against it. That’s good news for industries that are hard hit by the crime, such as banking and financial services and retail.
When it comes to cataloging what people do, big data is unrivaled in its ability to reveal records (contained in structured formats) of individuals’ financial transactions and online activities, not to mention their tweets, likes, and pins from unstructured and semi-structured containers.
To augment the sea of data, organizations typically turn to focus groups and surveys. That’s a start. But thick data goes one better: Instead of relying on individuals’ opinions and ability to recall, it observes firsthand how people act and experience the world. We refer to it as contextual analytics.
Worse, fraudsters move almost ghost-like through the dark web. Financial services providers and credit-card facilitators see only the traces that perpetrators leave in data, the patterns of their transactions. They know little about what motivates them, or what might scare them away.
We wondered what financial institutions might learn if data scientists could gain a firsthand understanding of the world of credit card fraud looks like. Could they build better fraud detection algorithms?
The question triggered months of study for our proof of concept. We met, conversed with, and observed fraudsters — and garnered surprising results. Credit card thieves see their line of work as hustling to survive, not a get-rich-quick play. Longevity requires they move fast and spend small sums of money to stay below the radar. It’s labor-intensive work.
That reality opens a different window on how to deter them: Make their work more difficult, and suddenly it’s a less attractive option.
By taking into account the human perspective, banks can sharpen their fraud-detection algorithms. They can better spot the types of orders fraudsters place. They can scrape real-estate databases for-sale listings and cross-tab them against credit-card transactions and produce potentially suspicious purchases.
Big data, it seems, benefits from a human touch.
What business solutions can it team up most effectively with thick data to solve? Any application of analytics that looks to examine and take into account underlying behaviors, for example, call-center dynamics and customer acquisition and retention.
Take a mortgage processing company that acquires loans from large banks. To increase customer loyalty and refinancing opportunities, it might pair thick data and big data to understand the drivers of retention. It might also use it for internal purposes, such as determining the characteristics of successful loan officers. What findings can it turn up to hire high-performing loan officers?
One challenge for organizations is putting clearly defined processes in place. That is, which queries are best posed against thick data? Which against big data? It’s common to try to answer the wrong question with the wrong method.
For example, because most organizations have grown accustomed to associating data with numbers, they often expect to analyze big data first, and then move to thick data. But numbers can constrict their vision. Thick data is where you start developing questions and theories, and then move to big data to scale them.
Determine Whether Your Organization is Ready for Thick Data
Determine your organization’s readiness for adding thick data to your toolkit by reviewing the following steps in the interactive figure below: