An indispensable but often overlooked part of creating an AI-based product is its UX design. So, we built a complex model that runs at scale on a massive cluster of GPUs serving thousands of users in real-time. But how do we make it usable for our customers to interact with? Let’s take a simple example. At Gong, we sometimes allow our users to give feedback on model predictions: mark false positives and surface false negatives that are then integrated back into the model. This can be both powerful and terrifying to the user. What happens if I make a mistake? How will it affect the model? When will I see the changes? Can I change my mind? These questions lie at the intersection of research and UX. And indeed, we find that they can best be answered when the two teams work together. It’s important to create tools that allow the users to engage with our models while setting expectations and mitigating the fear of making mistakes. This joint talk will present some real-life challenges we tackled and how we solved them with clever UX backed by research-based benchmarks.