top of page
bg_.jpg
Automatically calculate your accounting fraud risk
(Wirecard AG would have scored a 99.90%!)
Save 160 hours/audit or money back
wirecard.png
Service
Picture1.png
brain.png

Image 1: AI mimics the way that biological neurons signal to each other

mad_wix.png

Image 3: Simulated profits for Madoff Investments [AuditedAI research]

adv_wix.png

Image 4: Faking legitimate accounting transactions [AuditedAI research]


Introduction

Imagine it's the year 2012. 


When you start auditing, you tell an AI to scan the accounting transactions for all your companies.

One of your companies is Wirecard
.

For Wirecard, the AI returns a risk of 94% for accounting fraud.

Unfortunately, AI didn't have the capability back then.

But today, after three years of research, AuditedAI can do it.

 


Dr. Franco Arda
Consultant | Founder
Frankfurt am Main, Germany

 

 

 

Teaching an AI to detect accounting fraud:

 

AI is loosely based on how humans learn. 

 

An AI learns similarly to a human brain by

inputs (1), outputs (2), and an objective function

that guides the learning process.

In our case, the objective function of the AI
is to minimize the number of errors in
predicting accounting fraud
.

Essentially, we fed the AI an enormous amount of

legitimate and fraudulent accounting transactions
(100 billion accounting transactions) and trained it to predict fraud. As we will see in the following three use cases, the AI learned to detect even very complicated fraud patterns. For more details, please refer to "How AI learns to detect accounting fraud."

Wirecard:

Wirecard was the biggest fraud in German history.

The Financial Times published some of the
Wirecard revenues publicely.


We can see gaps (1) and suspicious

patterns (2) in Wirecard's sales, and

the AI predicts a high risk of fraud (3).

 

For a fraud at this scale, I’m surprised
that the data doesn’t look more similar
to Madoff Investments (next),
suggesting that Wirecard booked
a considerable amount of
legitimate transactions.

Madoff Investments:

Madoff was a $64.8 billion Ponzi scheme. 

In a Ponzi scheme, all transactions are
invented, making visualizations (1) look flat.

 

Among all accounting fraud types,
Ponzi schemes are the easiest to
detect
.

Hence, the high level of risk (2)
predicted by the AI.

 

Why was Madoff not caught earlier?

One crucial mistake was analyzing Madoff’s
monthly returns (i.e., aggregated data).
Based on our research, accounting fraud can (almost)
only be detected at a transactional level (i.e., granular data).

Advanced Accounting Fraud:


With the popularity of Benford’s Law,
some advanced fraudsters might try to
fake legitimate accounting transactions.

 

It’s not easy, but tutorials are available
that shows how to mimic Benford’s
distribution (e.g., Beat Benford).

However, our AI can detect such
mimicking patterns (1) where the
observed distribution (blue bars)
is unnaturally close to the
expected distribution (red line).

 

During the training, the AI learned
probabilistic patterns where a fraudster
mimics Benford’s distribution.


 

Image 2: Wirecard data published by the Financial Times (15 October, 2019

wirecard_wix.png
Use Cases
bottom of page