People have high expectations when they interact with their favorite brands. When they are unimpressed or bombarded with irrelevant content, they are more likely to disengage and bounce. To help organizations avoid this outcome, Froomle offers an AI-powered recommendation system that allows companies to provide highly personalized content suggestions to their users. The Froomle team’s core focus is on the algorithms, so they decided to partner with Panenco to build a complete self-service interface for their clients.
Although their AI system has great performance, the recommendation model’s decisions required a user interface for human oversight. Additionally, Froomle needed to be able to customize the platform to meet the demands of each customer in their diverse client portfolio. Our goal was to create a qualitative web application that satisfies these two business requirements in order to increase AI interpretability, auditability and client customization.
In the first phase of our partnership with Froomle, we created a platform that helped Froomle and their clients interpret the AI models used in the recommendation system. The platform allows end-users to generate personalized recommendations based on user-defined personas and sample data. In the second phase, we extended the platform to allow for better management of organizations and their brands, as well as the technical settings associated with fine-tuning the AI model's hyperparameters. This allows Froomle to adjust the model to better meet the specific needs of each client.
As I previously managed our customer success activities, I was very aware of the technical and operational complexities in building the self-service portal. The Panenco team quickly grasped our objectives and was able to deliver on our expectations and way beyond.
To address the first challenge of improving interpretability, we conceptualized an intuitive user interface that allows clients to view and understand the key factors that influence the model's recommendations.
This feature provides transparency into the decision-making process, allowing clients to have more insights in the recommendations made.To tackle the second challenge of customization, we implemented a flexible system that allows clients and customer success to adjust the model's parameters and constraints to better align with the specific business objectives.
This allows Froomle to offer a tailored solution to each client, ensuring that the recommendations generated by the model are optimized for their individual needs.The result of our efforts is a web application that combines interpretability and customization, providing Froomle's clients with a powerful tool for understanding and utilizing their first-party data to generate high-quality recommendations.
To address the issue of explainability, we gave users the ability to input their own data and observe how the model reacts to it. This allows them to gain a better understanding of the model's behavior and the quality of its outputs. We also implemented the ability for users to tune the model's parameters for increased customization. Throughout the development process, we made sure to prioritize scalability to ensure that the tool could be used effectively by a wide range of users.
The self-service portal is actively being rolled out internally at Froomle to support the customer success activities and to all existing customers. We’re assisting the Froomle team in setting up the necessary instruments to monitor success parameters through qualitative and quantitative measurements. We’re curious to see which impact this product will have across the Froomle customer base.