Staff Analytics Engineer, Data Platform

WHO WE ARE Zefir is building an AI autopilot for home sales in Europe, starting in France: an AI agent runs the entire sale and purchase journey end-to-end, orchestrating local brokers, portals, buyers, and documents. Backed by over $55 million from top-tier investors like Sequoia Capital, we're committed to accelerating life changes for millions of current and future European homeowners. An AI agent that runs a property transaction end to end only works if the data underneath it is reliable, governed and cheap to query. That is the job. WHY THIS ROLE EXISTS This is the first dedicated data hire in years. The foundations run, nobody owns them yet. Ingestion and centralisation into BigQuery already work, so you are not starting from an empty warehouse. What is missing is everything downstream: a canonical structure, one definition per metric, query cost under control, and a self-serve layer so that Finance, Growth and Ops stop routing every question through one person. Today the stack holds because individuals across Ops, Growth, Finance and Engineering compensate locally. They learned the quirks and built workarounds. It works, and it is fragile: KPIs drift between tools, tracking breaks silently, costs escalate, and nobody owns the translation between raw engineering data and decision-ready truth. You will be the single accountable owner of that layer. Not a support function, not a ticketing desk, not a BI factory. WHAT YOU WILL OWN Canonical models and metric definitions. A documented semantic layer with canonical entities (Buyer, Seller, Asset, Agent) and Bronze / Silver / Gold layers. Clear contracts between what Engineering exposes and what BI consumes, so that KPI debates stop being about whose number is right. Self-serve enablement. The submerged part of the iceberg: clean models, consistent BI primitives, row- and column-level security, so Ops, Growth, Finance and Account Managers build their own dashboards without compromising compliance. Analytics and tracking governance. The global event taxonomy and tracking roadmap, a hybrid client-side and server-side event strategy, consistent sync across CRMs and marketing platforms, and GDPR consent flows by design, so acquisition spend runs on attribution we can trust. Platform reliability, safety and cost. Standards set once rather than team by team: tested and versioned transformations, monitoring of freshness, failures and usage, sane ingestion patterns (read replicas, CDC, batch), and no production code path depending on BI tables. One thing worth stating plainly: our AI tooling already queries the warehouse directly, and the cost of it is not under control yet. Designing the guardrails, the schema curation and the authorisation layer is part of the job from week one, not a phase two. WHAT SUCCESS LOOKS LIKE AFTER 12 MONTHS - One documented event taxonomy, actually used by Engineering, Growth and CRM. - One semantic layer where every shared KPI has a single definition, a single owner and a version history. New joiners understand the data model in days, not months. - Published freshness and failure SLAs, an explicit ingestion topology, and no production path depending on BI tables. - Ops, Growth, Finance and AMs build most of their recurring dashboards themselves, and AI agents query the data layer safely through curated MCPs. - Growth attribution is trustworthy enough that annual acquisition spend decisions are defensible end to end. WHAT WE ARE LOOKING FOR - 7+ years as a Data, Analytics or Platform Engineer, ideally including a stint at a fast-moving consumer or marketplace company. Staff or Lead exposure expected. - Hands-on with the modern data stack: BigQuery (or Snowflake, Redshift), dbt or equivalent, advanced SQL and data modeling, Python for pipelines, orchestration (Airflow, Dagster, Prefect). - You have shipped event tracking and instrumentation in production, end to end: taxonomy, client and server-side events, attribution, GDPR-compliant opt-out, propagation…