Developed a user-centric dashboard for Adyen, enabling dynamic monitoring of payment authenticity.
The development of Adyen's Sherlock, an innovative machine learning feature designed to enhance risk assessment capabilities, presented an intricate challenge: making the complex, data-driven decision-making process of the model comprehensible and accessible to users with minimal machine learning expertise.
To bridge this gap, our approach was both immersive and collaborative, ensuring we fully grasped the nuances of risk assessment from the perspectives of those who knew it best—Risk Officers and Data Scientists.
Our initial step involved a deep dive into the world of risk assessment. By engaging directly with Risk Officers and Data Scientists, we sought to uncover the essence of their work: the criteria they consider, the data they rely on, and the jargon that permeates their discussions. This phase was crucial for gathering the insights needed to demystify the model's operations for a non-specialist audience.
Armed with a comprehensive understanding of risk assessment processes, we moved to identify the specific language, terminology, and data types that would be essential in explaining the model's "thinking" to users. This involved translating complex machine learning concepts into more accessible terms, ensuring clarity and simplicity without sacrificing accuracy.
With a clear set of requirements in hand, the design phase focused on creating an interface that was not only user-friendly but also educational. The goal was to design interactions that would guide users through the model's risk assessment process, illustrating how data inputs lead to risk evaluations. This approach aimed to foster a deeper understanding and trust in the model's capabilities.
American Airlines served as the ideal candidate for testing our design concepts. By applying our solutions in a real-world context, we were able to gather valuable feedback from a large-scale user with complex risk assessment needs. This phase was instrumental in refining our approach, ensuring it met the needs of diverse users.
The insights gained from testing with American Airlines facilitated the final development of Sherlock. We were able to implement a solution that not only met the sophisticated requirements of a major airline but also scaled to empower smaller customers. By enabling these customers to opt into ML-based risk assessment, we provided them with access to sophisticated risk mitigation tools without the need for a dedicated team of risk officers.
Explainability was a widget that shows weights of the machine learning processing.
Some handy sidebars in use for different user actions.
A complete revamp of the risk rule builder. Field collections and types. Client-based rule testing.
Each rule has to have an outcome.
Modal with all the available fields to create bespoke rules.
