Editing Ai Marketing (section)

== <span style="color: #FFFFFF;">Applying</span> ==
'''Customer LTV and churn prediction pipeline:'''
<syntaxhighlight lang="python">
import pandas as pd
import numpy as np
from sklearn.ensemble import GradientBoostingClassifier, GradientBoostingRegressor
from sklearn.model_selection import train_test_split
from sklearn.metrics import roc_auc_score, mean_absolute_percentage_error
import shap

# Load customer transaction data
df = pd.read_csv("customer_transactions.csv")
df['order_date'] = pd.to_datetime(df['order_date'])

# RFM features (Recency, Frequency, Monetary)
cutoff_date = df['order_date'].max()
rfm = df.groupby('customer_id').agg(
    recency=('order_date', lambda x: (cutoff_date - x.max()).days),
    frequency=('order_id', 'nunique'),
    monetary=('order_value', 'sum'),
    avg_order_value=('order_value', 'mean'),
    days_since_first=('order_date', lambda x: (cutoff_date - x.min()).days),
    order_gap_std=('order_date', lambda x: x.sort_values().diff().dt.days.std()),
).reset_index()

# Churn model (target: no purchase in last 90 days)
rfm['churned'] = (rfm['recency'] > 90).astype(int)
features = ['recency', 'frequency', 'monetary', 'avg_order_value',
            'days_since_first', 'order_gap_std']
X, y = rfm[features].fillna(0), rfm['churned']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, stratify=y)

churn_model = GradientBoostingClassifier(n_estimators=200, max_depth=4)
churn_model.fit(X_train, y_train)
print(f"Churn AUC: {roc_auc_score(y_test, churn_model.predict_proba(X_test)[:,1]):.3f}")

# LTV model (target: next 12 months revenue)
rfm_ltv = rfm[rfm['churned'] == 0].copy()  # Only active customers
X_ltv = rfm_ltv[features].fillna(0)
y_ltv = rfm_ltv['next_12m_revenue']  # Requires additional data preparation
ltv_model = GradientBoostingRegressor(n_estimators=200).fit(X_ltv, y_ltv)

# SHAP explanations for campaign targeting decisions
explainer = shap.TreeExplainer(churn_model)
shap_values = explainer.shap_values(X_test)
shap.summary_plot(shap_values, X_test, feature_names=features)
</syntaxhighlight>

; Marketing AI tool landscape
: '''Ad platform AI''' → Google Performance Max, Meta Advantage+, Amazon DSP
: '''MMM''' → Meta Robyn, Google Meridian (open source), Analytic Partners
: '''Attribution''' → Northbeam, Triple Whale, Rockerbox (post-iOS 14)
: '''Email personalization''' → Klaviyo AI, Braze, Iterable predictive sending
: '''Audience platforms''' → LiveRamp, The Trade Desk AI, Salesforce Marketing Cloud
: '''Experimentation''' → Optimizely, VWO, LaunchDarkly (feature flags + stats)
</div>

<div style="background-color: #8B4500; color: #FFFFFF; padding: 20px; border-radius: 8px; margin-bottom: 15px;">