Associate Director - Analytics

Purplle is two businesses under one roof. A commerce platform — app, web, and ~250 stores serving 15M+ monthly users. And a CPG house of brands that sells on Purplle, on other marketplaces, in quick commerce, and in general and modern trade. Both run on a data stack that is mid-rebuild.

You will own the layer that turns our warehouse into decisions — metric definitions, the semantic layer that agents and decision engines query, and the predictive systems that tell us what is about to happen rather than what already did. Across every function: commerce, category and merchandising, CPG brand P&Ls, supply chain, stores, retail media, marketing, and finance. No function is a second-class consumer.

What you're building

Medallion, with a clear line of ownership. Data Engineering owns bronze and silver — ingestion, contracts, schema validation, freshness SLAs. You own gold: the curated marts, conformed dimensions, and business logic that everything downstream depends on. One definition of a metric, one place it lives, and a hard rule that nothing consumes silver directly.

A large share of what lands in bronze comes from outside our own walls — marketplace and quick-commerce partner feeds, distributor and trade data, store systems, ad platforms. Reconciling that into one trustworthy view of a SKU, a customer, and a brand's true margin is the hardest part of the job and the part with the most upside.

A semantic layer that machines consume, not just humans. Metric and entity definitions as versioned code, compiling into three consumption paths:Natural-language analytics — dynamic resolution for conversational querying, so a category manager, a brand lead, or a store cluster head gets an answer without an analyst in the loopTyped contracts for decision engines and agents — pricing, replenishment, assortment, personalization and spend-allocation systems read metrics through a stable API, not hand-rolled SQL that driftsReference definitions for the ML workbench — features and targets built off the same logic that reports use, so model outputs and dashboards never disagree

This is the highest-leverage asset you will own. Its quality determines whether our agent layer is trustworthy or noisy.

Self-serve as the default, with governance underneath. Business teams build their own views on our internal analytics surface. Your team's job is to make that safe and fast: certified metrics, sensible defaults, row-level access, and clear deprecation of the long tail of stale dashboards. Success is measured by how few questions reach your team, not how many you answer. Metrics and predictive inferencing. Define the metric tree — from GMV and contribution margin down to the operational drivers each pod actually controls, with paired counter-metrics so nothing gets gamed. Every function gets its own branch and its own owner.

Then move it forward in time. Demand and replenishment forecasting. Sell-in versus sell-out reconciliation and stock-out prediction for the brands. Assortment and whitespace identification. Price and promo elasticity. New-store revenue prediction and catchment sizing. Churn and propensity scoring. Anomaly detection with root-cause attribution wired into the workflow that can act on it. Alerts that prescribe an action, not alerts that describe a number.

What you own day to day

Roadmap. Define and execute one analytics roadmap covering commerce and discovery, category and merchandising, the CPG brand portfolio, supply chain and quick commerce, retail stores, retail media, marketing, and finance. Allocate your team's capacity to where the margin is, not to whoever asks loudest.Causal rigour. Own the experimentation standard — power analysis, guardrail metrics, and quasi-experimental methods where clean A/B isn't possible: geo tests, staggered store rollouts, price and promo changes, media holdouts, packaging and formulation changes. Kill naive pre-post reads wherever they appear.AI leverage. Your team uses agents for exploration, SQL generation, and first-pass analysis as a default, not a novelty. You set the norms for where a human must stay in the loop.Executive counsel. Be the person leadership calls before a decision, not after. Translate ambiguity into an analytical question, then into a recommendation with an owner and a number attached.

Year one, measured

Gold layer live across core domains — customer, SKU, store, brand, supply — with certified metric definitions replacing duplicate logicSemantic layer serving both conversational analytics and at least two production decision systemsMajority of recurring business questions answered self-serve, with the legacy dashboard estate materially reducedExperimentation standard adopted org-wide, plus a documented list of decisions it changedAt least two predictive systems in production, in different functions, with tracked business impact rather than model metrics

What we look for

8+ years in analytics, BI, or data science, including 3+ years leading teamsE-commerce, D2C, CPG, retail, or consumer tech — you have carried a number, not just reported oneRange across functions. You can hold a conversation about fill rate and one about ROAS in the same afternoon, and you know which of the two is currently costing more moneyExpert SQL; strong Python for analysis, modelling, and light toolingHands-on with modern warehousing, dbt-style transformation, and a semantic layer or metrics storeTrack record moving a team from reactive reporting to predictive and automated decisioning — including the political work of retiring what people were attached to