Seven challenges for harmonizing explainability requirements

One-size-fits-all explainability won't work: The paper argues that XAI techniques require "a highly contextualized approach" tailored to specific stakeholders and business applications, rather than applying the same explanation methods across all financial services use cases.

Regulators are pushing for XAI adoption: Financial regulators have signaled interest in using explainable AI for model governance, operational servicing, and compliance, indicating that explainability is becoming a regulatory expectation rather than optional.

Current XAI techniques have limitations: The authors review recent technical literature and identify seven specific challenges in the field, suggesting that existing XAI methods may not fully meet the diverse governance and compliance needs that regulators expect.

Different stakeholders need different explanations: The paper emphasizes that various stakeholders (regulators, business teams, customers, etc.) have different needs for understanding AI models, requiring customized explanation approaches rather than generic solutions.

Implementation requires careful planning: Moving XAI from theory to practice in financial services demands consideration of specific business contexts and stakeholder requirements, implying that organizations need strategic approaches to explainability rather than simply adopting available tools.