Feedzai Fairband: World's most advanced framework for AI fairness

• New patent-pending AutoML algorithm that automatically discovers less biased machine learning models with zero additional model training cost while increasing model fairness by 93%
• Innovation recognized as “Fraud Prevention Innovation of the Year” by Fintech Breakthrough Awards

San Mateo, California, and London, United Kingdom – March 17th, 2021 – Feedzai, the leading cloud-based financial risk management platform, has announced Feedzai Fairband, the world’s most advanced AI fairness framework. The new AutoML algorithm automatically discovers machine learning models with zero additional model training cost while increasing model fairness by up to 93%, on average. With this new technology, Feedzai allows financial institutions across the world to make better and fairer decisions.

Feedzai Research developed the new framework under the premise that protecting financial institutions from financial crime can and should be done in a fair, accountable, and transparent way and that AI should not harm consumers. Feedzai Fairband can be used with any fairness metrics, model metrics, and sensitive attributes such as age, gender, ethnicity, location, and more. In addition, the technology works with any algorithm and model settings and has shown vast applicability besides risk and financial crime management.

“Feedzai Fairband is one of the biggest milestones in the financial services industry as it presents a low-cost, no-friction framework to address one the biggest problems of our era – AI bias,” says Dr. Pedro Bizarro, Chief Science Officer at Feedzai. “By creating the most advanced framework for AI fairness, Feedzai is allowing financial institutions to incorporate a critical piece of technology that addresses a problem under close public scrutiny with proven damaging effects across the globe. Building accurate and fairer models will be less challenging from now on.”

The problem
The world has witnessed widespread reports of AI bias in several sectors in critical areas of society, such as criminal justice, healthcare, or financial services. Real-world examples include racial and gender discrimination in several domains, including facial recognition, job-applicant screening tools, access to credit, or even medical diagnosis. The lack of regulation in AI Fairness, Accountability, Transparency, and Ethics (FATE) has created a worldwide problem that is expected to grow substantially over the next few years if industry leaders and governments don’t act quickly.

Particularly in the fintech industry, there’s a risk that AI systems deny access to financial services disproportionately across people from different groups, based on race, age, place of residence, profession, or employment status. Access to banking services is paramount today, especially during a pandemic in which there has been a rapid transition to digital payments.

Despite recent awareness, using fairness as an objective when developing AI is not standard practice yet. There’s a lack of practical fairness-enhancing methodologies and tools for practitioners, and developing less biased models faces three main challenges:

1. Practitioners are not sure how to measure fairness and assume that there is a costly trade-off (that much less biased models also have to be much less accurate);
2. Practitioners assume that decisions are out of their hand, frequently stating that “models are not biased, what is biased is the data”;
3. Practitioners assume that creating fairer models is complex in terms of model building activities and expensive in terms of time.

The Solution
Feedzai’s new patent-pending AutoML algorithm automatically discovers fairer machine learning models with zero additional model training cost, and near-zero sacrifice in terms of model accuracy while increasing model fairness by up to 93%, on average.

Feedzai Fairband can be used with any fairness metrics, model metrics, and sensitive attributes such as age, gender, ethnicity, location, and more. In addition, the technology works with any ML algorithm, model settings and can be seamlessly integrated into existing machine learning pipelines, allowing organizations to adapt pre-existing business operations to accommodate fairness with residual cost and without significant friction.

—

By not measuring and preventing bias, financial institutions will perpetuate AI behaviors that ultimately have a negative impact on how people use banking products and services. Adopting technology that deals with bias efficiently and flexibly helps mitigate these unwanted consequences while allowing customers to experience less discriminatory access to financial services, fostering their economic well-being and access to services in the digital age.

Fintech Breakthrough has recognized Feedzai Fairband as “Fraud Prevention Innovation of the Year”. James Johnson, Managing Director, FinTech Breakthrough, said, “Feedzai embodies the spirit of the FinTech Breakthrough Awards with an innovative solution addressing a critical issue. We are proud to extend our congratulations to Feedzai for winning such an important award in the 2021 FinTech Breakthrough Awards program.”

Feedzai Fairband will become available to Feedzai’s customers in Q2 2021. 

If you want to learn more about the problems financial institutions face, check out Feedzai’s blog post: Understanding AI Bias in Banking, and the research paper.

About Feedzai:
Feedzai is the world’s first RiskOps platform, protecting people and payments with a comprehensive suite of AI-based solutions designed to stop fraud and financial crime. Feedzai is trusted by leading financial institutions to manage critical risk and compliance processes, safeguarding trillions of dollars of transactions while improving the customer experience and protecting the privacy of everyday users. For more information, visit feedzai.com.

Media Contacts:
Feedzai [email protected]