AI PRODUCT MANAGER / PRODUCT OWNER — Yoga Studio Management Platform (studioos.ai)

AI PRODUCT MANAGER / PRODUCT OWNER — STUDIOOS (studioos.ai) Long-term contract, remote, overlap with CET or ICT. THE ROLE IN ONE LINE You own the product, the roadmap, and the number. Nobody above you is going to design your experiments for you. WHAT STUDIOOS IS An AI-native operating system for fitness and yoga studios. Small team, no layers, no committees, no roadmap theatre. This seat is the product function, working directly with engineering, design, and the founder. We are not looking for a backlog groomer. We want someone who treats product decisions as empirical questions and knows how to answer them properly. WHAT YOU WILL ACTUALLY DO - Own growth KPIs. Activation, retention, expansion, and the leading indicators underneath them. You define the metric tree, you defend it, you move it. - Run a continuous experimentation program. Hypothesis, design, power, pre-registration, readout, decision. Weekly cadence, not quarterly. - Do real research. Studio owner interviews, jobs-to-be-done, log analysis, funnel forensics. Know which method answers which question. - Handle post-hoc analysis honestly. Segment cuts, heterogeneous treatment effects, observational reads on things you cannot randomize. You are the person who knows when a result is real and when it is noise wearing a suit. - Own the AI product surface. Model choice, prompt and retrieval iteration, evals, latency and cost tradeoffs, failure modes, and what happens when the model is wrong in front of a customer. - Build the measurement infrastructure you need. Event schema, instrumentation specs, dashboards. If it does not exist, you spec it and ship it. - Write. Decision memos, experiment readouts, quarterly narratives. WHAT WE MEAN BY PROPER METHODOLOGY This is the filter. You should be fluent in most of this and honest about the rest: - Hypothesis design and pre-registration. Primary metric, guardrails, minimum detectable effect, decided before launch. - Power analysis and sample sizing. You know why an underpowered test that "won" is worse than no test. - Sequential testing and peeking correction. Always-valid inference, alpha spending, or the discipline not to look. - Variance reduction. CUPED, stratification, covariate adjustment. - Multiple comparisons and false discovery rate control when you slice. - Quasi-experimental methods for when randomization is impossible. Difference-in-differences, synthetic control, geo tests, switchback, regression discontinuity, interrupted time series. - Novelty and primacy effects, network interference, sample ratio mismatch, survivorship bias in cohort reads. - Evaluation for AI features. Offline eval sets, LLM-as-judge with human calibration, online quality metrics, regression suites, and the limits of all four. - Knowing when the honest answer is "this test cannot tell us that." Bonus if you have run experiments in low-traffic B2B SaaS, where sample size is scarce and clean A/B tests are often unavailable. That is the reality here. No PhD required. You need to have done this on a real product with real users and real consequences. WHO THIS IS FOR - 5+ years in product, growth, or data roles where you personally owned outcomes, not artifacts. - Shipped AI or ML-driven features to production users, not internal demos. - Comfortable in SQL. Comfortable enough in Python or R to run your own analysis rather than queue for someone else's. - You go from customer interview to statistical readout in the same week without switching personalities. - Low ego about being wrong, high standards about how you find out. WHO THIS IS NOT FOR - People who need a large org to be effective. - People whose experimentation experience is "we used Optimizely." - People who present dashboards instead of decisions. HOW TO APPLY Send three things in your proposal. Short is fine. Substance is not optional. 1. REFERENCE PROJECTS. Two or three you are genuinely proud of. For each: what the product was and what number you owned, the most important experiment or research effort you ran and how you designed it, what the result was including the null and failed ones, and what you decided afterwards. We care more about the reasoning than the outcome. A well-designed test that killed your favourite idea is a stronger signal than a win you cannot explain. 2. METHODOLOGY NOTE. One page, your own words, no template. A real situation where you could not run a clean A/B test. What you did instead, the threats to validity, and how much you trusted the answer. 3. TEARDOWN. Half a page. Look at studioos.ai or any product in a category you know well. One hypothesis you would test first, the metric you would move, and how you would size and design the test. Proposals that skip these three go unread. Start your proposal with the word EXPERIMENT so we know you read this.