Data Scientist

⌘Role Overview The Data Scientist is the first hire on the team whose job is to put machine learning into production , not just into a notebook. You'll build models that power product and other live flows sometimes surfaced to users in real time, not just reported on in a dashboard a week later. The first flagship project is about personalization across the user journey: a model that decides, per user, what offer to show, wired directly into the product and lifecycle experience rather than sitting in a warehouse table. From there, you'll extend the same muscle: model → API → product surface to other high-leverage moments in the user journey. You'll work hand-in-hand with Product, Engineering and Lifecycle marketing to ship models as features, not as reports. This is also a foundational role for the team's infrastructure: most of what Data does today is batch (dbt, Lightdash, BigQuery); you'll help establish our first real-time low-latency serving patterns on GCP and work closely with engineering on this. ⌘Key Responsibilities 1. Production ML Development Build, validate, and ship ML models (propensity, pricing/discount optimization, personalization, churn/LTV) that go live in the product, not just proof-of-concept notebooks. Own the full lifecycle: problem framing, feature engineering, training, evaluation, deployment, monitoring, retraining. Write production-grade code (tested, versioned, reviewed) — you'll be shipping alongside Engineering, held to their bar. 2. Personalization across the journey: from paywall to lifecycle Design and ship the personalized discounting model: who gets what offer, and why, served at the moment of the paywall decision. Partner with product on machine learning experiment design (A/B, holdouts) to prove causal lift of the models, not just correlation. Build the measurement framework so pricing/discount decisions are defensible to finance and leadership. 3. ML Infrastructure & MLOps (GCP) Stand up our first low-latency model serving pattern on GCP (e.g., Vertex AI endpoints, Cloud Run, or equivalent) Define the feature pipeline pattern: what's precomputed in BigQuery/dbt vs. what needs to be fresh/real-time via Pub/Sub or similar. Set up model monitoring: drift, staleness, prediction quality so a live model doesn't silently degrade. 4. Product & Engineering Partnership Sit close to Product and Engineering, not just Data this role is measured by what ships to actual users, not just notebooks Translate a product problem ("how might we reactive lapsed payers") into a modeling problem, and a model output into an API contract Engineering can build against. Document handoffs clearly enough that Engineering can own the serving layer long-term without you as a bottleneck. 5. Experimentation & Causal Inference Design uplift/causal models where "who responds to a discount" matters more than "who churns" . Run and interpret experiments that isolate the model's actual incremental impact on revenue/retention. Design the experimentation program for improving data science and machine learning models with the same rigour we use across our already ongoing experimentation programs ⌘Expected Outcomes Personalized discounting model live in production , serving real paywall decisions to real users: shipped end-to-end, not a prototype sitting in staging. A documented, reusable low-latency serving pattern established on GCP (Vertex AI endpoints or Cloud Run) — the next model doesn't require rebuilding this from scratch. Proven incremental lift on a core business metric (paywall conversion, discount margin efficiency, or lapsed-payer reactivation — pick the one you want as the flagship KPI), demonstrated through a controlled experiment, not just before/after comparison. Model monitoring in place — drift and staleness alerts mean the team knows within days, not months, if a live model silently degrades. A repeatable model-to-production playbook that others in the team can follow ⌘Skills & Competencies Must have (hard skills) Strong Python for data science and ML (scikit-learn, XGBoost/LightGBM; PyTorch or TensorFlow a plus if deep learning is relevant to future use cases). Proven track record shipping models to production, not just modeling in a notebook. Can talk through at least one model that served live traffic. Hands-on experience with a cloud ML platform, ideally GCP (Vertex AI, BigQuery ML, Cloud Run/Functions) or fast ability to translate equivalent AWS/Azure experience. Solid SQL; comfortable working against a dbt/BigQuery warehouse. Software engineering fundamentals: git, code review, testing, CI/CD: You'll be shipping code Engineering has to trust. Causal inference / uplift modelling or applied experimentation experience: pricing and discounting need "what if we hadn't," not just "who churns." Nice to have Experience with streaming/event pipelines (Pub/Sub, Dataflow, Kafka): useful as we move off pure batch. Experience with pricing, discounting, personalization specifically. Familiarity with feature stores or the DIY equivalent (versioned feature pipelines). Multi-armed bandits or reinforcement learning for pricing/personalization. Startup experience: comfortable being the first person to build something rather than joining an existing ML platform team. Soft skills Genuinely energised by "does this move the metric," not just "is this model accurate." Can hold their own in a room with Engineering and with Business: Speaks commercial as well as the language of engineering Explains modeling tradeoffs in plain business terms to Product/leadership without dumbing it down or using too much jargon Comfortable owning ambiguity — this role is defining the pattern, not following one. ⌘ Required experience 4–7 years in a data scientist / ML engineer role, with at least one model you personally took from prototype to live production serving real users or real traffic. Quantitative background (CS, stats, engineering or equivalent hands-on experience). B2C, subscription, or marketplace experience is a strong plus Fluent English. ⌘ Recruitment process Introduction Call - 30min with Talent Acquisition Lead Business Interview - 30min with VP Data Case Study - Async assessment Panel Interview - Case study Q&A Final interview - Interview with Product Manager