2nd/10_Wiki/Topics/AI_and_ML/G-Stack Principles.md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-g-stack-principles	G-Stack Principles	10_Wiki/Topics	verified	self	G-Stack growth stack growth engineering growth marketing	none	B	0.82	applied	growth growth-stack marketing analytics experimentation		2026-05-10	pending	language framework TypeScript / Python GA4 / Mixpanel / GrowthBook / Amplitude

G-Stack Principles (Growth Stack)

매 한 줄

"매 acquisition + activation + retention + revenue + referral 의 의 의 의 modern data + tooling stack". 매 product-led growth (PLG), 매 experimentation, 매 personalization. 매 stack: GA4 + Mixpanel/Amplitude + GrowthBook + Segment + Customer.io.

매 핵심

매 AARRR (Pirate metrics)

• Acquisition.
• Activation: 매 'aha' moment.
• Retention: 매 D1, D7, D30.
• Revenue.
• Referral.

매 stack tier

• Analytics: GA4, Mixpanel, Amplitude, PostHog.
• CDP: Segment, RudderStack.
• Experimentation: GrowthBook, Optimizely, LaunchDarkly.
• Marketing automation: Customer.io, Braze, Iterable.
• CRM: Salesforce, HubSpot.
• Reverse ETL: Hightouch, Census.
• Warehouse: Snowflake, BigQuery.

매 응용

1. Onboarding optimize.
2. Retention email / push.
3. Pricing test.
4. Feature flag rollout.
5. Personalization.

💻 패턴

Event tracking (Segment)

import { Analytics } from '@segment/analytics-next';
const analytics = new Analytics({ writeKey });
analytics.track('Sign Up Completed', {
  plan: 'pro',
  source: 'landing',
});

A/B test (GrowthBook)

import { GrowthBook } from '@growthbook/growthbook';
const gb = new GrowthBook({ apiHost, clientKey });
await gb.loadFeatures();
gb.setAttributes({ id: userId, country });

if (gb.isOn('new_pricing_page')) showNewPricing();
const variation = gb.getValue('hero_headline', 'default');

Funnel analysis (Mixpanel)

import mixpanel
mp = mixpanel.Mixpanel(token)
mp.track(distinct_id, 'Funnel Step', {'step': 1, 'name': 'view_landing'})

Cohort retention

import pandas as pd
def retention_curve(events_df):
    events_df['cohort'] = events_df.groupby('user_id').date.transform('min').dt.to_period('M')
    events_df['period'] = (events_df.date.dt.to_period('M') - events_df.cohort).apply(lambda x: x.n)
    return events_df.groupby(['cohort', 'period']).user_id.nunique().unstack()

Activation prediction

import xgboost as xgb
def predict_activation(user_features):
    """매 user 의 의 의 activate?"""
    return xgb.XGBClassifier().fit(X_train, y_train).predict_proba(user_features)[:, 1]

Segment + reverse ETL

// 매 warehouse → tool sync
// 매 Hightouch config:
{
  source: 'snowflake',
  query: 'SELECT email, ltv FROM users WHERE ltv > 1000',
  destination: 'customer_io',
  mode: 'upsert',
  on: 'email',
}

Email automation (Customer.io)

import requests
requests.post(f'https://api.customer.io/v1/campaigns/{cid}/triggers',
    json={'recipients': {'segment': {'id': 1}}, 'data': {'name': '{{first_name}}'}})

Cohort-based feature flag

gb.setAttributes({
  id: user.id,
  cohort: user.cohort,
  isPro: user.plan === 'pro',
});
if (gb.isOn('new_dashboard')) showNewDashboard();

LTV prediction

def predict_ltv(user_features):
    """매 first 30-day signal → 매 12-month LTV."""
    return xgb.XGBRegressor().fit(X_train, y_train).predict(user_features)

Churn prediction

def churn_probability(user_features):
    return classifier.predict_proba(user_features)[:, 1]

def trigger_winback(users):
    high_churn = users[churn_probability(user_features) > 0.7]
    customer_io.send_campaign('winback', high_churn)

Multi-touch attribution

def first_touch_attribution(touchpoints):
    return touchpoints[0]

def linear_attribution(touchpoints):
    return [(tp, 1/len(touchpoints)) for tp in touchpoints]

def time_decay_attribution(touchpoints, half_life_days=7):
    weights = [0.5 ** ((now - tp.date).days / half_life_days) for tp in touchpoints]
    total = sum(weights)
    return [(tp, w/total) for tp, w in zip(touchpoints, weights)]

Eval metric

def growth_metrics(users, period_days=30):
    return {
        'D1_retention': retention(users, day=1),
        'D7_retention': retention(users, day=7),
        'D30_retention': retention(users, day=30),
        'avg_LTV': mean([u.ltv for u in users]),
        'CAC': sum(u.acquisition_cost for u in users) / len(users),
        'NRR': net_revenue_retention(users, period_days),
    }

Onboarding flow optimize

def onboarding_funnel(events):
    steps = ['signup', 'verify_email', 'create_first_project', 'invite_teammate', 'first_action']
    drop_offs = []
    for i in range(1, len(steps)):
        rate = events[steps[i]].nunique() / events[steps[i-1]].nunique()
        drop_offs.append((steps[i-1], steps[i], 1 - rate))
    return drop_offs

매 결정 기준

상황	Tool
Product analytics	Mixpanel / Amplitude / PostHog
Pipeline	Segment + warehouse + Hightouch
Experimentation	GrowthBook / Optimizely
Email	Customer.io / Braze
Open-source	PostHog + GrowthBook
Enterprise	Amplitude + LaunchDarkly + Braze

기본값: 매 Segment + 매 warehouse + 매 PostHog/Amplitude + 매 GrowthBook + 매 Customer.io + 매 cohort retention focus.

🔗 Graph

• 응용: E-commerce-Optimization · Dynamic-Creative-Optimization
• Adjacent: Feature-Flag

🤖 LLM 활용

언제: 매 SaaS / consumer product. 매 PLG. 언제 X: 매 enterprise sales-led only.

❌ 안티패턴

• Vanity metrics: 매 page views, signups.
• No experimentation discipline: 매 noise as truth.
• Tool sprawl: 매 5+ overlapping.
• No cohort analysis: 매 average misleading.

🧪 검증 / 중복

• Verified (Reforge, GrowthBook docs, Segment docs, Pirate metrics McClure).
• 신뢰도 B.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — AARRR + 매 Segment / GrowthBook / cohort / LTV / churn code