Recently I had a chance to spend some time with the Director of the fraud detection department in a global insurance company who had been using his enormous budget to combat fraudsters for decades. It has always been a race for him – whenever he identifies and fills one hole, fraudsters find a new way to the vault. Operating in the world of star schema and multiple mature analytics tools, he dreamt about testing any hypothesis ad hoc.
The team he led was full of experienced analysts and data scientists. He and his people had multiple ideas where the frauds might be hidden as well as very often he was tipped by one of his peers about a new fraud scheme. The challenge though was to quickly test their hypotheses against their history and current claims, as this often required connecting distant data from different silos, organizing it in a sensible data model and then analyzing via various tools. If a pattern was confirmed, an alerting rule had to be implemented and sometimes a predictive model was built. How long did the process take? Days for simple questions, weeks for more complex patterns and months for those analyses that required data from multiple, separate sources.
You can only imagine how this influenced the team’s willingness to ask new questions and test their ideas.
But was there any other choice? Naturally, he could employ more and more data scientists and analysts, throw them onto research projects and then try to implement the most valuable results..but while his budget was big, it wasn’t THAT big.
When he saw an opportunity to bring in all of the silos into one hub and query it even with staff that is not trained in SQL, he grabbed the opportunity and that’s how we did our three day project.
Of course many of the hypotheses the team tested proved to be false, many required tuning, but the hundreds of omitted frauds the team found in the first days of using the flexible analytics repository paid for the effort. Some of the ideas were over a year old and finally revealed the hidden fraud patterns! Another quick win came from integrating the alerts generated by other systems into one repository for analysis, whereit turned out that some alerts needed refurbishing, some need to be replaced, etc. Overall the initial project resulted in an increase of fraudulent claims payment prevention by 11% in just three weeks, with the promise of far greater results as use of the system is expanded.
This one week POC reflected three key points:
First is that data quality is not as big of a problem as people think, if you have a technology that enables rapid data modeling and ingestion because the data gaps are quickly visible and adjustable.
Second is that once you bring the silos together and iterate through multiple models you suddenly get a lot of new ideas. Questions emerge that you never even imagined before you saw your data connected and within the business context.
Third and most important: when people see the data and can manipulate it without any barriers such as the speed of computing (joins!) or breathing through the SQL straw, they suddenly understand and trust the results!
Unlocking the teams’ creativity in a few days got us the contract, but more importantly showed the potential. Our unique technology and our technical team of very capable problem solvers enables us to demonstrate in a one week POC how we can blend data, provide new insights, and solve problems. If it all sounds hard to believe, then please challenge us to show what we can do!