A high-performing fraud model doesn't necessarily mean a healthier marketplace. Fraud teams often measure success through precision, recall, and detection rates. But those metrics can't answer the questions that matter most: - Did harmful behavior actually decrease? - Were legitimate members protected? - Did the marketplace become safer? Join Gabrielė Mongirdaitė of Vinted and Elisa Ahern of Incognia as they explore how fraud teams can move from measuring model performance to measuring real-world impact. They'll unpack the path from offline validation and experimentation to ongoing monitoring, examine the hidden costs of false positives and false negatives, and discuss why different interventions require different confidence thresholds. Attendees will leave with a practical framework for connecting fraud decisions to downstream outcomes, from harmful activity and member trust to marketplace health and business impact.

In the AI era, the oldest question in renting — is this person who they say they are? — has become the hardest to answer. Payslips can be generated, references fabricated, and forged documents now sail past checks that a letting agent could once catch by eye. Identity verification has become core infrastructure for the rental market, yet every check is also a moment of friction in a business built on conversion.

Trust & Safety teams already red-team feature launches, score risk against structured rubrics, map abuse surfaces, and interrogate how product mechanics can incentivize harm. But what happens to that discipline as the software development lifecycle becomes AI-native - as AI accelerates how fast features get designed, built, and shipped? This session examines how to keep rigorous, human-led safety evaluation embedded in a faster pipeline: where AI extends coverage and speed, where expert human judgment stays irreplaceable against novel features and novel abuse vectors, and how to measure prevention with credible leading indicators.

AI agents are quickly becoming one of the biggest talking points in commerce, promising a future where software can search, shop, book, negotiate and even make payments on a consumer’s behalf. But how much of that future is actually here today, and how much is still hype? This panel will separate the reality of agentic commerce from the speculation. Industry leaders will break down how AI agents are beginning to interact with marketplaces and merchants, what changes when the customer at the keyboard may no longer be human, and the new questions this creates around identity, authorization, payments, fraud and trust. From distinguishing legitimate agents from malicious automation to understanding whether an agent is actually authorized to act on someone’s behalf, the conversation will give attendees a practical foundation for navigating the emerging agentic economy.

Most platforms treat SMS one-time passcodes as a fixed cost of doing business: expensive, occasionally frustrating, but necessary. Bumble's engineering team decided to test that assumption. Jim Dolan, VP of Engineering, will discuss the technical and organizational case for replacing SMS OTPs with passive, key-based authentication across Bumble's global platform, and what the migration has revealed so far — from reduced fraud exposure and lower operational cost to a smoother sign-up experience for new members. A candid session for engineering, fraud, product and risk leaders evaluating whether their own authentication stack is quietly working against them, and what to expect when you start pulling the thread.