Database Reference
In-Depth Information
one of the authors of this topic. He had an issue with a company that was incor-
rectly billing to his credit card, but the bank told him that he had to resolve the
issue with the vendor. The bank would not stop the billing because it was a pre-
authorized charge. He emailed, then chatted, then called the bank over the
course of a couple days, all with increasing levels of frustration. After no luck
with this resolution method, he walked into a local branch that promptly tried to
upsell him on a new card feature. This clearly wasn't the right time to offer him
a promotion—he wasn't in the right customer state—but the poor CSR had no
idea this was the wrong time to do this. This bank experience is far from unique
(and our author wants you to know they are generally pretty darn good to deal
with), but it's a fact: most firms don't pick up the signals customers send them.
Fraud Detection:
“Who Buys an Engagement Ring at 4
a
.
m
.?”
When was the last time you bought an engagement ring at 4 a.m.? (We're
excluding any purchases made in Las Vegas.) Not very common, is it? This is
a good example of the concept of
outliers
, which are key to finding and trying
to predict fraud. You search for outliers all the time. Consider your smart
phone's detailed bill—you certainly don't look at hundreds of calls when
trying to figure out how you blew 2,000 minutes; but we're willing to bet you
spot that 70-minute phone call when the others are all measured in single-
digit minutes. Fraud and risk are part of cross-industry Big Data use cases
that are not restricted to the FSS. Big Data platforms, as we will see, are very
well suited to finding outliers. What's more, highly dynamic environments
commonly have cyclical fraud patterns that come and go in hours, days, or
weeks. If the data that's used to identify new fraud detection models isn't
available with low latency, by the time you discover these new patterns, it's
too late and some damage has already been done.
Several challenges to fraud detection are directly attributable to the sole
utilization of conventional technologies. The most common theme you'll see
across all Big Data patterns are limits on what (amount and type) can be
stored, and what compute resources are available to process your intentions.
In other words, models to predict fraud often are either too laser focused due
to compute constraints and miss things, or they are not as refined as they
could (or should) be because the dimensions of the model have to be
