Staff Data Scientist

About this role

For over 30 years, Angi has powered the future of the home services industry, creating an environment where homeowners and pros benefit from more jobs done well. For homeowners, our platform is a reliable way to find skilled pros. For pros, we're a reliable business partner who helps them find the winnable work they want, when they want. For employees, we're an amazing place to call home. We can't wait to welcome you.

Angi at a glance: - Founded in 1995 as Angie’s List and rebranded in 2021 - Global company with 9 brands in 8 countries and employees worldwide - Homeowners have turned to us for 300 million home projects and counting About the role: Angi is seeking an exceptional Staff Data Scientist to help build our agent suite for Pros, AI software that automates and runs the Pro's business for them. The first product in the suite is an AI voice and SMS agent that contacts homeowners on the Pro's behalf, qualifies the job, and books the appointment.

It is live with hundreds of Pros and growing, and it is the first of several agents we intend to build. This is a data science role on a product whose core behavior is conversational and probabilistic. The work spans shaping how the agent behaves, building the evaluation systems that tell us whether it is getting better, designing the experiments that decide what we ship, and modeling the parts of the lead to job completion journey where a model is the right answer.

Success in the role shows up directly in improving Pro’s revenue, which is the metric that determines whether Pros see a return from Angi. What you’ll do: - Agent Behavior & Development: Lead improvements to how our agents converse, qualify, and book, through conversation design, prompting, and campaign logic. Success in these areas directly impacts contact rate, engagement rate, appointment set rate, and job win rate which are critical metrics for both Pro outcomes and business success.

- Evaluation & Measurement: Build the evaluation systems that make agent quality measurable, spanning automated checks over conversation traces, simulated conversations graded against expected outcomes, and model-based evaluation where human judgment is required. Establish the frameworks and design of experiment practices we use to decide what ships. - Conversation Analysis: Work directly with real conversations to understand where the agent succeeds and where it falls short, translating that understanding into a shared taxonomy of failure modes, measured rates, and a prioritized path to improvement.