Growth Engineering Skill

Infrastructure and patterns for product-led growth, experimentation, and conversion optimization.

---

Feature Flag Systems

Implementation Pattern

// lib/feature-flags.ts
import { PostHog } from 'posthog-node';

const posthog = new PostHog(process.env.POSTHOG_API_KEY!);

interface FeatureFlags {
  'new-onboarding-flow': boolean;
  'pricing-experiment': 'control' | 'variant-a' | 'variant-b';
  'ai-suggestions': boolean;
}

export async function getFlag<K extends keyof FeatureFlags>(
  key: K,
  userId: string,
): Promise<FeatureFlags[K]> {
  const value = await posthog.getFeatureFlag(key, userId);
  return value as FeatureFlags[K];
}

// Usage in component
const showNewOnboarding = await getFlag('new-onboarding-flow', user.id);

Feature Flag Best Practices

• Short-lived flags: Remove after experiment concludes (< 2 weeks)
• Long-lived flags: Ops toggles for gradual rollouts, kill switches
• Never nest feature flags (creates exponential complexity)
• Clean up stale flags monthly
• Log flag evaluations for debugging

---

A/B Testing Infrastructure

Experiment Design

// lib/experiments.ts
interface Experiment {
  id: string;
  name: string;
  variants: {
    id: string;
    weight: number; // 0-100, must sum to 100
  }[];
  targetAudience: {
    percentage: number; // % of users included
    filters?: Record<string, unknown>;
  };
  primaryMetric: string;
  secondaryMetrics: string[];
  minimumSampleSize: number;
  startDate: Date;
  endDate?: Date;
}

// Track experiment exposure
function trackExposure(experimentId: string, variantId: string, userId: string) {
  analytics.capture({
    event: '$experiment_started',
    distinctId: userId,
    properties: {
      $experiment_id: experimentId,
      $variant_id: variantId,
    },
  });
}

Statistical Significance

• Minimum sample size: Calculate before starting (use Evan Miller calculator)
• Don't peek: Set duration upfront, don't stop early on promising results
• Sequential testing: Use if you must check early (adjusts p-values)
• Minimum detectable effect: Define what improvement matters (e.g., 5% lift)

---

Product-Led Growth Patterns

Activation Metrics

Stage	Metric	Example
Sign up	Registration complete	User creates account
Setup	Profile complete	Fills required fields
Aha moment	Core value experienced	Creates first project
Habit	Repeated engagement	3 sessions in first week
Revenue	Conversion to paid	Subscribes to plan

Viral Loops

// Referral system pattern
interface Referral {
  referrerId: string;
  referredEmail: string;
  status: 'pending' | 'signed_up' | 'activated' | 'converted';
  rewardGranted: boolean;
}

// Track referral funnel
function trackReferralStep(referralId: string, step: Referral['status']) {
  analytics.capture({
    event: 'referral_step',
    properties: { referralId, step },
  });
}

Conversion Optimization

• Reduce friction: Minimize form fields, enable social login
• Social proof: Show user counts, testimonials, logos
• Urgency: Trial countdown, limited-time offers (use sparingly)
• Value demonstration: Interactive demos, free tier with clear upgrade path
• Personalization: Onboarding flow based on use case selection

---

Growth Metrics

Metric	Formula	Target
Activation rate	Activated / Signed up	> 40%
Trial-to-paid	Paid / Trial started	> 15%
Net revenue retention	(Start MRR + Expansion - Contraction - Churn) / Start MRR	> 110%
Viral coefficient	Invites sent * Conversion rate	> 0.5
Time to value	Median time from signup to aha moment	< 5 min
DAU/MAU ratio	Daily active / Monthly active	> 20%

---

Experimentation Platforms

Platform	Type	Best For
PostHog	Self-hosted/cloud	Full-stack, open source
LaunchDarkly	Cloud	Feature flags at scale
Statsig	Cloud	Auto-stats, warehouse-native
Growthbook	Self-hosted/cloud	Open source, Bayesian stats
Optimizely	Cloud	Enterprise, multi-channel

---

Related Resources

• ~/.claude/skills/product-analytics/SKILL.md - Analytics and tracking
• ~/.claude/agents/product-analytics-specialist.md - Analytics agent
• ~/.claude/skills/authentication-patterns/SKILL.md - Auth for PLG

---

Measure everything. Experiment constantly. Remove what doesn't work.