Data scientists, software engineers, and other experts are creating business solutions that rely on algorithms, automated decision-making tools, artificial intelligence (AI), and machine learning models. Their work has developed faster, more efficient business processes, but it has also raised some concerns. 

• Washington state took measures to prevent abuse by banning the use of credit information for setting insurance rates.
• Lenders in New York need to provide fair and equal access to lending options and ensure that their practices are not discriminatory.
• In Colorado, insurers who develop new insurance products must be prepared to explain the protections that they have in place for avoiding unintended bias.

Regulators and consumers expect companies to demonstrate how they have mitigated bias and discriminatory decisions in their advanced decision-making models. Enterprises will need to explain how their work attempts to minimize these risks.

Data itself is neither good nor evil

In some models, data points like age or gender might be defined as “bad” and excluded to avoid biased outcomes. In other models, the same data points could be defined as “good” and integrated to produce unbiased outcomes. This ambiguity has been the source of public criticism, and faulty assumptions expose enterprises to serious consequences.

The advanced application of data offers a mix of threats and opportunities. Models and algorithms can reduce human error by interpreting data objectively, but they can also deploy human and societal biases at scale. In these cases, it is frequently the underlying data, not the algorithms themselves, that cause problems. 

Biased data not only presents risks to consumers, but also to the companies that serve them. Flawed pricing models can lead to financial losses, and regulators like the US Federal Trade Commission have warned that companies will face regulatory action if they fail to hold themselves accountable for the fairness of their algorithms. 

However, organizations can integrate clean, equitable, and actively monitored external data into their workflows to improve digital onboarding, online fraud prevention, and credit decisioning processes — in addition to reducing model bias — to meet regulatory needs and customer concerns. 

Discernment and Expertise

Market-leading enterprises overcome the limits of their own internal data by working with external data products. The key is to select expert partners who curate high-quality data that is relevant, cost-effective, and as close to real-time as possible.

Organizations like FinRegLab are exploring how external data can be applied to credit underwriting decisions. CGAP is researching emerging business models to determine which business models have the clearest links to financial inclusion. And Demyst has worked with TruEra to develop a solution for mitigating bias in AI-based credit decisioning models.

Enterprises that commit to making a difference have seen instant value with immediate positive results. At the core level, these organizations are working with unbiased data and minimizing distortions so that their models account for a broader, more diverse population. The effective integration of external data is a key driver of digital transformation and external data platforms such as Demyst continue to work in partnership with innovative enterprises to help to make that difference.