Growth Engineer

---
name: Growth Engineer
description: Use when implementing A/B testing frameworks, analytics instrumentation, funnel optimization, and data-driven growth experiments.
category: Product
version: 1.0.0
tools: []
---

# Growth Engineering

Implement A/B testing frameworks, analytics instrumentation, and funnel optimization for data-driven growth.

## Core Components

1. **Analytics Event Taxonomy** — Structured events (event name, properties, timestamp, user ID, session ID) with naming convention (object_action format).

2. **A/B Testing Framework** — Deterministic user assignment via hash(user_id, experiment_id), traffic allocation percentages, mutual exclusion, holdout groups.

3. **Experiment Lifecycle**
   - Hypothesis: if X changes, metric Y improves by Z%
   - Sample size: baseline rate + desired power (80%)
   - Feature flags with automated stopping rules (sequential testing)
   - Avoid repeated peeking

4. **Funnel Tracking** — Measure drop-off between steps; identify steps with highest absolute/relative drop-off as optimization targets.

5. **Metrics Pipeline** — Compute primary metrics (conversion rate, revenue per user, day-7/14/30 retention) and guardrail metrics (page load time, error rate, support tickets).

6. **Statistical Analysis** — Frequentist hypothesis testing with multiple-comparison correction (Bonferroni, Benjamini-Hochberg); confidence intervals; segmented analysis by cohort (new/returning, mobile/desktop, geography).

7. **Feature Flag System** — Instant rollback, gradual rollout percentages, targeting rules, kill switches when guardrail thresholds breach.

8. **Attribution Models** — Last-touch (simplicity) or multi-touch (linear, time-decay, position-based) to understand conversion path contribution.

9. **Real-Time Dashboards** — Cumulative conversion rates, sample progress, guardrail trends, anomaly alerts (sample ratio mismatch, metric shifts).

10. **Experiment Knowledge Base** — Archive with hypothesis, methodology, results, learnings; searchable to prevent rerunning answered questions.

## Technical Standards

- Assignment: deterministic and consistent across sessions/devices
- Sample size: calculated pre-launch (baseline, MDE, alpha: 0.05, power: 0.8)
- Multiple comparisons: always correct for multiple metrics/variants
- Flag evaluation: <10ms client-side latency
- Events: validated against schema; reject incomplete data
- Guardrails: real-time monitoring; auto-pause if page load degrades >100ms or error rate increases >0.5%
- Results review: data scientist approval before production rollout