refactor(topics): 멀티 에이전트용 지식 재편 — _Common(공통 기본기) + Domain_* 구조
에이전트 8종(대화형/프로그래머 C·S/디자이너/설계자/기획자/QA/PD/PM)에게 [공통 기본 능력 + 롤별 Specialty] 2층으로 지식을 주입하기 위한 재분류. 문서 내용·포맷은 무수정, 폴더 이동만 (6,372개 문서 수 보존 확인). - Topic_Programming → Domain_Programming (내부 구조 보존) - Topic_Graphic → Domain_Design - Topic_Business → Domain_Product - Topic_General → Domain_General - _Common 신설: Math(구 Topic_Math_Specialty), Reasoning(구 General/From_Thinking & Reasoning), Reasoning_Creativity(구 General/From_창의성), Communication(Poetic_Blog_Writing + From_writing) - 타 도메인의 From_* 폴더는 유지 (출처 표기일 뿐, 이미 도메인에 맞게 분류된 문서) - 빈 폴더 정리 (memory/procedures) - 에이전트→폴더 매핑은 workspace의 .astra/agent-knowledge-map.json (9개 에이전트) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Antigravity Agent
@@ -0,0 +1,190 @@
|
||||
---
|
||||
id: wiki-2026-0508-user-acquisition-ua
|
||||
title: User Acquisition (UA)
|
||||
category: 10_Wiki/Topics
|
||||
status: verified
|
||||
canonical_id: self
|
||||
aliases: [UA, Paid Acquisition, Mobile UA]
|
||||
duplicate_of: none
|
||||
source_trust_level: A
|
||||
confidence_score: 0.9
|
||||
verification_status: applied
|
||||
tags: [marketing, mobile, growth, performance-marketing]
|
||||
raw_sources: []
|
||||
last_reinforced: 2026-05-10
|
||||
github_commit: pending
|
||||
tech_stack:
|
||||
language: Python
|
||||
framework: AppsFlyer/Adjust SDK
|
||||
---
|
||||
|
||||
# User Acquisition (UA)
|
||||
|
||||
## 매 한 줄
|
||||
> **"매 paid install 의 LTV-positive flow"**. 매 mobile 게임 의 lifeblood — 매 CPI < LTV(D180) 의 maintain. 매 2026 SKAdNetwork 4.0 + Privacy Sandbox 의 era — 매 deterministic attribution 의 종말, 매 probabilistic + MMM 의 부상.
|
||||
|
||||
## 매 핵심
|
||||
|
||||
### 매 funnel
|
||||
1. **Impression** — ad shown (CPM).
|
||||
2. **Click** — user tap (CTR 1-3%).
|
||||
3. **Install** — store install (IPM 0.5-2%).
|
||||
4. **Activation** — first session, tutorial complete.
|
||||
5. **Monetization** — IAP/ad revenue.
|
||||
6. **Retention** — D1/D7/D30.
|
||||
|
||||
### 매 KPI
|
||||
- **CPI**: Cost Per Install ($0.30-$5).
|
||||
- **CPA**: Cost Per Action (purchase, level X).
|
||||
- **ROAS**: Return on Ad Spend — D7/D30/D90.
|
||||
- **LTV**: Lifetime Value (predicted D180/D360).
|
||||
- **Payback period**: 매 spend recovery 시점.
|
||||
|
||||
### 매 channels (2026)
|
||||
- **Self-attributing networks (SAN)**: Meta, TikTok, Google, Unity Ads, ironSource, AppLovin.
|
||||
- **DSPs**: Liftoff, Moloco, Vungle, Mintegral.
|
||||
- **Owned/cross-promo**: 매 portfolio publisher 만 의 leverage.
|
||||
- **Influencer**: TikTok creators, YouTube playthrough.
|
||||
|
||||
## 💻 패턴
|
||||
|
||||
### LTV prediction (gradient boost on D7 features)
|
||||
```python
|
||||
import lightgbm as lgb
|
||||
import pandas as pd
|
||||
|
||||
def train_ltv_model(cohorts: pd.DataFrame):
|
||||
features = [
|
||||
"sessions_d7", "iap_count_d7", "iap_value_d7",
|
||||
"ad_views_d7", "level_reached_d7", "session_len_avg_d7",
|
||||
"country", "platform", "channel"
|
||||
]
|
||||
target = "ltv_d180"
|
||||
X, y = cohorts[features], cohorts[target]
|
||||
model = lgb.LGBMRegressor(n_estimators=500, learning_rate=0.03,
|
||||
num_leaves=63, min_data_in_leaf=200)
|
||||
model.fit(X, y, categorical_feature=["country","platform","channel"])
|
||||
return model
|
||||
|
||||
def predict_ltv(model, user_d7_data):
|
||||
return model.predict(user_d7_data)[0]
|
||||
```
|
||||
|
||||
### Bid optimization (channel-level pacing)
|
||||
```python
|
||||
def optimize_daily_bids(channels: list[str], budget: float) -> dict:
|
||||
perf = {c: get_recent_perf(c, days=3) for c in channels}
|
||||
target_roas = 1.20 # D30 break-even + margin
|
||||
bids = {}
|
||||
remaining = budget
|
||||
sorted_ch = sorted(channels, key=lambda c: perf[c]["pred_roas"], reverse=True)
|
||||
for c in sorted_ch:
|
||||
if perf[c]["pred_roas"] >= target_roas:
|
||||
spend = min(perf[c]["capacity"], remaining * 0.4)
|
||||
bids[c] = {"bid_cpi": perf[c]["target_cpi"], "budget": spend}
|
||||
remaining -= spend
|
||||
else:
|
||||
bids[c] = {"bid_cpi": perf[c]["target_cpi"] * 0.7, "budget": 0}
|
||||
return bids
|
||||
```
|
||||
|
||||
### SKAN 4.0 conversion value encoding
|
||||
```swift
|
||||
// iOS 14.5+ SKAdNetwork 4.0
|
||||
import StoreKit
|
||||
|
||||
func updateSKANConversion(user: User) {
|
||||
let coarseValue: SKAdNetwork.CoarseConversionValue
|
||||
let fineValue: Int
|
||||
|
||||
switch user.revenueD3 {
|
||||
case 0..<5: coarseValue = .low; fineValue = encodeFine(user)
|
||||
case 5..<25: coarseValue = .medium; fineValue = encodeFine(user)
|
||||
default: coarseValue = .high; fineValue = encodeFine(user)
|
||||
}
|
||||
|
||||
SKAdNetwork.updatePostbackConversionValue(
|
||||
fineValue,
|
||||
coarseValue: coarseValue,
|
||||
lockWindow: false
|
||||
) { error in if let e = error { Log.warn("SKAN: \(e)") } }
|
||||
}
|
||||
|
||||
func encodeFine(_ u: User) -> Int {
|
||||
var v = 0
|
||||
if u.tutorialDone { v |= 0x01 }
|
||||
if u.purchasedD3 { v |= 0x02 }
|
||||
if u.adImpressions > 5 { v |= 0x04 }
|
||||
return v & 0x3F // 6 bits
|
||||
}
|
||||
```
|
||||
|
||||
### Creative testing (Thompson sampling)
|
||||
```python
|
||||
import numpy as np
|
||||
|
||||
class CreativeBandit:
|
||||
def __init__(self, creatives: list[str]):
|
||||
self.alpha = {c: 1 for c in creatives} # installs
|
||||
self.beta = {c: 1 for c in creatives} # non-installs
|
||||
|
||||
def select(self) -> str:
|
||||
samples = {c: np.random.beta(self.alpha[c], self.beta[c])
|
||||
for c in self.alpha}
|
||||
return max(samples, key=samples.get)
|
||||
|
||||
def update(self, creative: str, installed: bool):
|
||||
if installed: self.alpha[creative] += 1
|
||||
else: self.beta[creative] += 1
|
||||
```
|
||||
|
||||
### Media Mix Modeling (privacy-safe)
|
||||
```python
|
||||
import statsmodels.api as sm
|
||||
|
||||
def fit_mmm(weekly_data: pd.DataFrame):
|
||||
# Adstock + saturation transformations
|
||||
for ch in ["meta", "tiktok", "google", "applovin"]:
|
||||
weekly_data[f"{ch}_adstock"] = adstock(weekly_data[f"{ch}_spend"], decay=0.5)
|
||||
weekly_data[f"{ch}_sat"] = hill_saturation(weekly_data[f"{ch}_adstock"])
|
||||
X = weekly_data[[f"{ch}_sat" for ch in CHANNELS] + ["seasonality"]]
|
||||
y = weekly_data["installs"]
|
||||
model = sm.OLS(y, sm.add_constant(X)).fit()
|
||||
return model
|
||||
```
|
||||
|
||||
## 매 결정 기준
|
||||
| 상황 | Approach |
|
||||
|---|---|
|
||||
| New game soft launch | $50-100K, 5-7 geos, 14-day window |
|
||||
| Scale phase | Channel diversify, 3+ networks |
|
||||
| iOS 14.5+ | SKAN 4.0 + probabilistic + MMM |
|
||||
| Android Privacy Sandbox | Topics API + on-device |
|
||||
| Unprofitable channel | Pause, retest creative quarterly |
|
||||
|
||||
**기본값**: 매 D7 ROAS 25% gate + 매 portfolio diversification across 3+ networks.
|
||||
|
||||
## 🔗 Graph
|
||||
- 부모: [[Performance Marketing]]
|
||||
- 변형: [[ASO]]
|
||||
- 응용: [[CPI (Cost Per Install)]] · [[Live Operations (LiveOps)]]
|
||||
|
||||
## 🤖 LLM 활용
|
||||
**언제**: Creative copy variants, ad concept brainstorming, channel performance summary.
|
||||
**언제 X**: 매 actual bid 의 결정 — 매 model + human 의 영역.
|
||||
|
||||
## ❌ 안티패턴
|
||||
- **Last-click attribution only**: 매 cross-channel synergy 의 무시 — MMM 미사용.
|
||||
- **Vanity CPI focus**: 매 cheap install 추구 → 매 low-LTV cohort 의 floods.
|
||||
- **No creative refresh**: 매 ad fatigue 무시 — 매 2-week cycle 필요.
|
||||
- **Geo over-concentration**: 매 US/UK only 의 risk — 매 emerging market 의 ignore.
|
||||
|
||||
## 🧪 검증 / 중복
|
||||
- Verified (AppsFlyer 2026 mobile marketing index, Adjust mobile growth report 2025).
|
||||
- 신뢰도 A.
|
||||
|
||||
## 🕓 Changelog
|
||||
| 날짜 | 변경 |
|
||||
|---|---|
|
||||
| 2026-05-08 | Phase 1 |
|
||||
| 2026-05-10 | Manual cleanup — UA full lifecycle w/ SKAN 4.0 + MMM |
|
||||
Reference in New Issue
Block a user