Explain life story, project leadership, and negotiation

# Structured answer (example for a Data Scientist HR screen) 1) Life story with two inflection points (3–4 minutes) - Undergraduate: BS in Statistics, University of Michigan (May 2016). I led a capstone on A/B testing for a local retailer, which got me into applied experimentation. - First role: Business/Data Analyst at Acme Analytics (Jun 2016–Aug 2018). Built dashboards and did pricing analyses for retail clients. - Inflection point #1 (Aug 2018): I moved from consulting into product data science to own outcomes end-to-end. I joined Nimbus Ads as a Data Scientist (Sep 2018–Dec 2020) to work on ranking and causal lift measurement. - Second role: Senior Data Scientist at Aurora Marketplaces (Jan 2021–Mar 2023). Focused on marketplace quality, conversion funnels, and experimentation platform. - Inflection point #2 (Apr 2023): I shifted from pure marketplace growth to commerce enablement at Finlink Commerce as a Lead Data Scientist (Apr 2023–present) to work closer to checkout, payments, and merchant success, aligning my work with small-business impact. 2) From your brochure: three insights + one critique, and my 90-day contributions Assumption for context completion: Your latest brochure (2025) emphasizes a merchant-first mission, unified on/offline commerce, an accelerated one-click checkout, and AI-driven merchant tooling. I’ll validate these in the call. - Insight 1: Merchant-first mission with an emphasis on helping new merchants activate quickly. - 90-day contribution: Map the merchant activation journey, define a north-star like “Activated Retained Merchants @90 days,” and ship an event taxonomy v1. Target: reduce time-to-first-sale median from 14 days to 11 days by day 90 via cohort-specific nudges (welcome flows, setup checklists). - Insight 2: Unified checkout with an emphasis on frictionless payments and trust. - 90-day contribution: Baseline checkout drop-off by step; launch one experiment to cut authentication-related abandonment. Target: +1.0–1.5 pp absolute checkout conversion on mobile for returning buyers by day 90. Partner with Eng (Samir Gupta) and PM (Lena Ortiz) to instrument p95 latency and SCA step completion. - Insight 3: AI tools to help merchants create content and listings. - 90-day contribution: Establish offline/online evaluation for AI content: quality rubric, hallucination rate, and brand-safety guardrails. Target: increase listing creation speed by 25% for new merchants while keeping policy-flag rate under 0.5%. - Respectful critique: The brochure is feature-forward but light on outcomes by segment (e.g., new vs. established merchants) and on experiment rigor. - 90-day contribution: Publish an internal metrics rubric tying each feature to a causal KPI (e.g., uplift in weekly active merchants, CAC payback), plus a quarterly readout template. Pilot with two features and document guardrail metrics (latency, support tickets, policy flags). 3) Consequential mistake in the last 3 years - What I misjudged (Sep 2023): We shipped a recommender model update (new embeddings) to 20% traffic based on strong offline AUC. I underweighted online guardrails and assumed latency impact would be negligible. - Earliest leading indicator I missed: Within 24 hours, p95 API latency rose +150 ms and the add-to-cart rate for returning users dipped 0.6 pp on treated traffic. I dismissed it as weekend noise. - Measurable impact: Over 9 days, impacted cohorts saw −1.9 pp checkout conversion, leading to an estimated −$420k gross merchandise value (GMV) delta before we rolled back on Oct 3, 2023. - Systemic change implemented: - Introduced a “must-pass” guardrail bundle (latency, add-to-cart, repeat-purchase) with sequential tests (SPRT) and a 24-hour tripwire for automatic rollback. - Required shadow-mode canaries for 7 days for models affecting ranking, with real-time p95 latency alerts in Slack. - Added a pre-mortem checklist to RFCs (owner: me; approvers: PM/Eng). Post-change, no model rollbacks in 2024 were due to latency regressions. 4) Proudest work achievement - North-star metric: 90-day retained paying merchants (MRR-contributing accounts still active at day 90). - Context (Jan–Nov 2022 at Aurora Marketplaces): New-merchant retention was 54%. Activation friction and unclear setup drove early churn. - Work: Built a propensity-to-churn model, redesigned the onboarding checklist with progressive disclosure, and launched experiment-backed nudges (email + in-product). Paired with PM (Ava Reynolds) and Lifecycle Marketing (Jordan Park). - Trade-offs: We paused two low-ROI onboarding features and diverted 1.5 Eng FTEs to instrumentation and experimentation for 2 sprints. - Risks retired: Compliance risk from UGC during onboarding (added automated checks and human review for risky categories). - Before/after metrics: 90-day retention rose from 54% to 61% (Q4 2022), a +7 pp lift; LTV +11%; activation time (first sale) median dropped from 16 to 12 days. Attribution via CUPED-adjusted experiments and matched-market holdouts. 5) Most recent large project (end-to-end) - Project: Merchant Trust & Risk Scoring for onboarding and early transaction protection. - Timeline: Jan 8, 2024–Oct 18, 2024. - Scope: Ingest KYC/KYB signals, build a trust score (gradient-boosted trees), set tiered thresholds, and implement a human-in-the-loop review workflow with SLAs. - Cross-team coordination: PM (Lena Ortiz), Eng (Samir Gupta and Mei Lin), Legal/Compliance (Alex Cho), Risk Ops (Priya Nair), Sales/Support (Diego Martinez), Finance (Li Wang). - Conflict and resolution: Sales wanted lenient thresholds to maximize onboarding; Risk wanted conservative thresholds to minimize chargebacks. I built a scenario analysis showing expected GMV vs. bad-loss. We agreed on dual-path onboarding: a “fast lane” with caps and rolling limits and a “review lane” with expedited SLAs (<6 hours). We reviewed outcomes weekly. - Key risks and mitigations: - False positives blocking good merchants → Appeal flow with 24-hour SLA; feature importance explanations in the reviewer UI. - Regulatory compliance (KYC consent) → Explicit consent capture and audit trail; monthly Legal audits. - Model drift → Weekly PSI monitoring; auto-retrain with human approval. - Midstream change and what we cut/added: - Change (May 2024): New EU guidance required explicit consent logging and data retention limits. - Added: Consent event schema, redaction jobs, and audit dashboards. - Cut: Real-time retraining MVP and a secondary device-graph feature to hit compliance deadlines. We documented cuts in the decision log and committed to a Q1 2025 follow-up. - Outcomes (Nov 2024): Bad-loss rate −32% year-over-year for new merchants; onboarding conversion −2.1 pp (acceptable within target band of −3 pp). Net GMV +$6.8M with 95% CI excluding zero based on matched-market analysis. 6) Delivered without manager support - Situation (Jun–Aug 2023): My manager was on parental leave during a push to unify revenue attribution for growth budgeting. - Alignment and sponsorship: I wrote a 6-page PRD and a 3-page RFC outlining a phased approach (last-touch baseline → incrementality tests → MMM). I secured sponsorship from the Growth Director (Nadia Patel) and Finance (Li Wang) in a steering review on Jun 20, 2023. - Blocker I couldn’t resolve: Warehouse access to finance-grade bookings tables. I raised it to the Data Governance Council, proposed a read-only view with row-level security, and got approval on Jul 3. - Protecting scope: Marketing asked for real-time influencer tracking; I parked it in “Phase 2,” protecting the Phase-1 deliverable (weekly budget reallocation report by channel). - Artifacts produced: PRD, RFC, decision log, data contract schema, and a living QA checklist. We launched on Aug 14, 2023; impact: −12% CAC for paid social in Sep–Oct via budget shifts, with +4.2% overall conversions at flat spend. 7) Compensation role-play (if range isn’t shared) - Strategy: Keep it collaborative, seek signals (level, total comp structure), anchor to market data, and offer a range contingent on scope/level while avoiding a low anchor. - Exact phrasing I would use: - “I understand if you can’t share the full range yet. To make sure we’re aligned on level and scope, could you share the internal leveling for this role and the typical mix of base/bonus/equity?” - “Based on current market data for similar roles in [city/remote], for this scope I’m targeting a total compensation in the range of $X–$Y, depending on level and equity mix. If that’s materially outside your band, I’d love to calibrate early so we don’t waste anyone’s time.” - “If sharing a range isn’t possible today, no problem—once we confirm level, I’m confident we can find a number that works for both sides.” — How to adapt this structure for your own story - Use STAR per section: Situation → Task → Action → Result with dates, names, and numbers. - Pre-commit targets in 90-day plans (e.g., +1–2 pp conversion, −20% time-to-insight) and name collaborators. - Always include guardrails (latency, churn, policy flags) for experiments and ML changes. - Keep a decision log to show scope protection and traceability. - Validate any brochure-derived insights during the call if you’re inferring details.