Lead Data Scientist - Recommendations

Scribd, Inc. is on a mission to advance human understanding. Our four products — Scribd®, Slideshare®, Everand™, and Fable — help billions of people across the globe move beyond access and into insight, application, and expertise.

Culture at Scribd, Inc.

We support a culture where our employees can be real and be bold; where we debate and commit as we embrace plot twists; and where every employee is empowered to take action as we prioritize the customer.

We believe the best work happens when individual flexibility is balanced with meaningful community connection. Scribd Flex empowers employees to choose the workstyle and location that support their best performance, while committing to intentional in‑person moments that strengthen collaboration and culture. Occasional in‑person attendance is required for all Scribd, Inc. employees, regardless of location.

So what are we looking for in new team members? At Scribd, Inc., we hire for “GRIT.” Traditionally defined as the intersection of passion and perseverance toward long‑term goals, GRIT reflects the mindset we expect from every employee. For us, it also serves as a practical framework for how we work: setting and achieving Goals, delivering Results within your role, contributing Innovative ideas and solutions, and strengthening the broader Team through collaboration and attitude.

This posting reflects an approved, open position within the organization.

Scribd, Inc.’s Data & Analytics team is hiring a Lead Data Scientist to own measurable outcomes across our recommendation surfaces – translating product goals into metrics, leading roadmap bets, and shipping lifts in business results. You’ll define the offline/online contract end‑to‑end, design and run experiments, diagnose why variants win or lose, and build prototype models while partnering with Engineering to product ionize.

You’ll map goals to metrics with clear success criteria, focus on opportunity sizing and measurement, and apply an AI lens (LLMs, embeddings) where it demonstrably improves retrieval, ranking, or understanding—shaping how millions engage with our global content library.

Scribd is a differentiated subscription platform with strong organic reach and a vast catalog—books, audio books, and hundreds of millions of UGC documents and slides. In a landscape reshaped by AI, our opportunity is to help users cut through noise and discover high‑quality, human‑centered content. You’ll set north stars and guardrails, create leading indicators that predict long‑term outcomes, and build the measurement architecture—identity, attribution windows, metric contracts, and drift/leakage checks—that keeps downstream metrics trustworthy.

You’ll also accelerate decision velocity with clear stop/go criteria and power checks, and tell the story through concise decision memos with trade‑offs and risks.

• Opportunity mapping. Size and prioritize new recs surfaces, intents, and cohorts; trace the funnel and analyze by slice (cold items, long‑tail users, platform) to steer the roadmap.
• Own the evaluation framework. Define north star & guardrails (e.g. diversity, novelty, duplication, safety); set threshold and tradeoffs, and publish the Objective & Eval Contract per surface.
• Offline/Online alignment. Quantify correlation between offline IR metrics (e.g., NDCG@K, MAP, MRR, coverage, calibration) and online KPIs by surface/cohort; publish error bounds and monitor metric drift.
• Create leading indicators. Create short‑horizon metrics that predict long‑term outcomes (e.g., trial to bill‑through); backtest and run post‑hoc causal checks, reporting uncertainty.
• Build the measurement architecture. Set identity & attribution standards ( vs. , qualifying events, windows) so downstream metrics (bill‑through, churn) are trustworthy.
• Design and run advanced experiments such as interleaving tests, pre‑register stop/go criteria, and deliver crisp readouts that drive decisions.
• Codify schemas, freshness, leakage, and drift checks with Analytics and Data Engineers, establish high quality datasets for Recs algo.
• Evaluate when LLMs/embeddings (topics, summaries, semantic similarity) measurably improve…