Círculo de Crédito - How to prevent fraud with machine learning

From the outset, Featurespace demonstrated extensive experience in fraud prevention

Círculo de Crédito


Círculo de Crédito began operations in 2005 as a Credit Information Society (SIC) focused on promoting and supporting sectors that did not participate in other societies, such as the middle and lower socioeconomic levels, which has consolidated them as the Credit Information Society of Financial inclusion in Mexico.

“We have been working for over 20 years to develop Círculo de Crédito with a clear goal: to be the credit information society of inclusion. And we have achieved it, as we currently manage the information of more than 80 million Mexicans and actively contribute to the generation of solutions so that more people have access to credit,” said Ruiz Palmieri after receiving the Most Valuable Player (MVP) award granted by Featurespace.


Download the Círculo de Crédito Hero Story

The challenge

Icon – Tools

No specialized tools to mitigate fraud

Users had few alternatives to perform manual fraud assessments, which represented a process limited by the knowledge of each user and product and/or limited by the number of analysts, as all assessments required at least one analyst for a certain time.

Icon – Press Button

All fraud prevention procedures were manual

Analysts could generate home visits, phone calls, or requests for references, but in addition to the high costs involved, they could not guarantee the applicant’s intentions.

Icon – Dashboard

A generic score was the only alternative

The other option presented was to request a generic score that meant the anaylsts would lose focus on its market (personal loans), and resulted in lower efficiency levels than expected, and generated some uncertainty about the benefits of the model.

Why Círculo de Crédito chose Featurespace

A solution different from all our scoring systems

  • Integrated Machine Learning techniques
  • Extensive experience in fraud prevention
  • At the forefront with attractive performances
  • A combination of experience and results
  • Adaptability to change

How Círculo de Crédito uses ARIC™ Risk Hub

The integration and design of ARIC Risk Hub was always focused on first-person fraud because we wanted to maintain the essence and objective for which ARIC Risk Hub was designed, which is focused on detecting patterns with a low presence rate in the total set of events.

Additionally, fraud has always been a behavior that changes continuously; it never stabilizes or becomes a set of static characteristics. This is because fraudsters look for new ways to exploit vulnerabilities in any process and attack them. And when these vulnerabilities are removed, the fraudster will look for the next open door to attack.

This is where Adaptive Behavioral Analytics helps us to identify the common attack patterns and continuously reviews and updates to identify new patterns that could be exploited and that was not previously known.

The Results

Icon – Chart

ARIC Risk Hub surpasses other scores

In terms of fraud detection, ARIC Risk Hub surpasses other scores when applied to the value of fraud savings.

Icon – Chart pointing down

ARIC Risk Hub is sensitive to the amount of losses

ARIC Risk Hub shows a big difference in the amount of fraud savings, and delivers more accuracy in cases where the loss can potentially be larger. The results has a great impact on fraud mitigation as it can double the amount of fraud saved compared to generic scores.

Icon – Magnifying Glass

51% increase in fraud detection

We have specific users who, when comparing a generic score and taking an equal cut-off point, had an increase in fraud detection of up to 51% compared to one decile of the population – this generated savings of more than one million pesos to increase the business’s profitability.

Icon – Chart with bars

62% difference between lowest and highest-scored fraud losses

It was found that these levels of fraud detected is due to the accuracy and prioritization achieved by ARIC Risk Hub, achieving a difference of up to 62% between fraud losses of the lowest-scored applications and the highest-scored fraud losses.