Huge number of false alerts triggered from AML monitoring systems is one of the biggest challenge being faced by Banks and Financial institutions today. Due to inefficiencies of transactions monitoring systems (TMS), 90 to 95 percent of alerts are false. Not only does this translate into operational overhead, it may lead to missing real alerts hiding under enormous number of false alerts

The False -ve financial Exposure Problem:

What is the financial exposure as a result of AML?

3-5% 0f Global GDP is laundered through the global systems. Penalties, alone, was $12 trillion and expected increased by $400 billion in 2020.

The false +ve “We can’t stop the fraud unless we see it” problem

Given the ever-growing sophistication of contra parties, 96% of system generated alerts are “false positives”, requiring investigations(costly and time-intensive).

False positive cost billions of dollars in wasted investigation time each year but more importantly, expose banks to steep fines and reputational damages for failing to identify bad actors involved in organized crime sanctions evasion, or terrorism.

Our learning algorithms take advantage of the large pools of data and heightened computing power available to
detect patterns that might go unnoticed by data scientists.

Challenges with many of the existing AML products


Linear rule based techniques become brittle and are hampered by the multiple data privacy regulations.


Consistent occurrence of false negatives.


Very high false positives.

Our differentiators

Differentiating Machine Learning

Al models often lack learning corpus which makes implementation very difficult. Solving this issue using SME defined taxonomies is key.

Applying algo techniques (random forest / neural network) on overall data makes them brittle, so forcing Taxonomy led segmentation using CART methods, to define segments for decision trees makes the leaning models a lot more flexible without loss of accuracy.

Non-Linear Regression Models with Synthetic data

Effectiveness of graph / regression models can be enhanced by generating synthetic data to seed pattern generations using various techniques.

Synthetic data represents banks historical behavior coupled with industry intelligence enabling more bank-relevant patterns.

Various Reporting Views:

Because patterns can run into thousands, understanding anomalies (F-ve) and conforming transactions (F+ve) is aided with various visualizing techniques to help defend the final reports with regulators.

Extensive logs to help audit the findings, including excel based outputs to help customize and integrate reports with exiting regulatory SARs reporting infrastructure.