Fraud with Amanda Bernard – New Perspectives on Accounting and Finance


(upbeat music) – So when we’re looking for
fraud and identifying fraud data analytics takes us pretty far. It identifies what we refer to as red flags or anomalies in the data. But you can’t really stop there. You do have to pursue those
red flags with human analysis because not every red
flag signifies fraud. Or it could be a really common reason for the red flag or anomaly. And the only way you can really drill down and find the reason is
through the human followup. It’s interesting right now. What we see is a lot
of the larger companies are really thoroughly
using data analytics. I work more with the small
and medium-sized businesses. And some of ’em have yet
to really dive into it. A lot of it takes a business investment into the technology needed
to do data analytics. Or maybe it’s just a skillset the employees don’t currently have. There has been an
identification a long time ago of the fraud triangle. Donald Cressey came out
with it a long time ago and it still holds true today. And there are three components. First, there’s usually some type of financial pressure
involved where the individual may have a gambling problem,
drug or alcohol addiction. Or sometimes it’s some
type of medical expenditure that they incur, some
reason where they just need some extra money and they can’t find that through a legitimate source. The next piece is rationalization, because most of these people
aren’t necessarily criminals, or at least they wouldn’t
identify themselves as a criminal, at least as we think of one. So they need to look at
themselves in the mirror at the end of the day, and that’s where the rationalization comes into play. So sometimes it’s as simple as, my boss doesn’t treat me fairly. They deserve it. I’ll pay it back, is a really common rationalization for fraud. And those are two pieces we don’t have a lot of control over as
management of a company. What we do have control over is the opportunity side of the triangle. And that involves having
decent controls in place to prevent somebody from
being able to steal money. A lot of times we start just
by gathering a lot of data. It can be a little overwhelming. So when we’re planning our data analytics, the first step is usually
to take a step back and decide what we want to do and what we want to look for. Part of that, at least how I
find it easiest to approach, is to think about it as
a brainstorming session and say, okay, what could go wrong? So some examples are, if an employee was going to add themselves
to the vendor system and pay themselves, how would they do it? If management wanted to mis-state the financial statements,
how would they do it? And that really allows us
to identify the patterns that we would be looking for
in the data analytics analysis. I work with small and
medium-sized businesses mostly, so we haven’t delved into
the AI and machine learning capabilities of computers in analytics. So we’ve more focused on
some of the basic techniques, things that can be done
in Microsoft Excel, or very simple tasks that would allow you to look for fraud in a more simpler form. In the long run, I think
that is where we’re going, the AI and machine learning. We’re just not there yet, at least at the small company scale. Some of the large companies
are already delving into that though and
it’s quite fascinating. One of the more fascinating
data analytics techniques we’ve discovered is called
Benford’s Law analysis. So on the face of things,
if you think about a group of numbers, the
dataset, you would assume that the first digit of all
numbers would occur equally. In other words, about
one in nine times or 11%. It’s actually not true. And it was discovered a long time ago by looking at, if you remember
the days before calculators, to do logarithms and calculus
you had big logarithm books. They discovered that when
they turned the pages, the one pages were a lot more worn than the twos, threes, and up to nine. So once they analyzed these patterns and performed the analysis
across multiple datasets what they discovered was that in reality a one occurs as the
first digit of a number about 30% of the time, a two, about 18%, and so on down to a nine. It actually represents a curve. So what we can do is plot
a set of data numbers against this expectation of a curve, and things tend to spike when
somebody’s committing fraud because most people don’t
know about Benford’s Law. And it’s actually human nature if you’re making up
numbers to commit fraud to select numbers like a five, a six, an eight as your first digit. So it’s actually, I’ve always
found it quite fascinating that we can use something like that that’s been around for 100 years and use it as fraud detection. They’ve actually found it works for lengths of rivers,
population statistics, basically any dataset of natural numbers. So that’s one area I find
absolutely fascinating.

Leave a Reply

Your email address will not be published. Required fields are marked *