SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Dwelly is building an AI operating system for residential lettings, operating over 15,000 properties and $470M in GMV across the UK. You will be the first dedicated data scientist in the Growth function, owning the analytical agenda alongside commercial and product leads. This is a thought-partner role where you bring hypotheses, challenge priorities, and build analyses that drive decisions—not a service seat.
The role spans multiple high-impact areas: churn early-warning that identifies root causes (arrears, SLA breaches, sentiment from call recordings) and separates recoverable customers from those to upsell; share-of-wallet expansion by estimating landlords' off-platform portfolios and mining call recordings for product insights; pricing elasticity analysis across acquired agencies to recommend defensible fee increases; rent guarantee underwriting using proprietary loss data; designing readable growth experiments with proper holdout groups and uplift estimates; and owning the Growth data layer including pipelines from platform databases, payments, comms, and call transcripts.
You will work with years of unstructured data—conversations, property management jobs, payments, arrears, and call recordings—that competitors cannot access. Recent LLM advances make this data actionable. You need 5+ years as a data analyst or scientist with full-stack capabilities: Python, SQL, and pipeline building (dbt or equivalent). Statistical depth is essential: experiment design, causal inference (diff-in-diff, matching, synthetic control), survival analysis, and elasticity estimation. You should apply pragmatic ML (churn propensity, uplift modeling, pricing, forecasting) and be fluent with LLMs on messy text and audio, including building eval sets to validate output quality. Data engineering is part of the job, not adjacent. You operate with high autonomy, volunteer caveats before others find them, and prefer well-calibrated production models over experimental notebooks.
