Python Customer Analytics, Churn Prediction & Behavioral Clustering

import pandas as pd
from sklearn.preprocessing import StandardScaler
from sklearn.cluster import KMeans
from sklearn.metrics import silhouette_score

# 1. Ingest customer behavior telemetry
df = pd.read_csv("customer_behavior.csv")
features = ["monthly_sessions", "feature_adoption_pct", "support_tickets", "monthly_mrr"]

# 2. Scale features to prevent high-dollar values from dominating Euclidean distance
scaler = StandardScaler()
X_scaled = scaler.fit_transform(df[features])

# 3. Fit K-Means with k=4 clusters
kmeans = KMeans(n_clusters=4, random_state=42, n_init=10)
df["cluster"] = kmeans.fit_predict(X_scaled)

# 4. Measure cluster separation (Ranges from -1.0 to +1.0)
sil_score = silhouette_score(X_scaled, df["cluster"])
print(f"Optimal Silhouette Score: {sil_score:.3f}")

# 5. Centroid Profiling for Executive Marketing Strategy
centroids = df.groupby("cluster")[features].mean()
print(centroids)

import pandas as pd
from lifetimes import BetaGeoFitter, GammaGammaFitter
from lifetimes.utils import summary_data_from_transaction_data

# 1. Structure transactional logs into RFM summary format
# frequency = number of repeat purchases
# recency = customer age at last observed purchase (weeks)
# T = total customer tenure since first purchase (weeks)
rfm = summary_data_from_transaction_data(df, "cust_id", "date", monetary_value_col="amount")

# 2. Fit BG/NBD model for repeat transaction dynamics
bgf = BetaGeoFitter(penalizer_coef=0.01)
bgf.fit(rfm["frequency"], rfm["recency"], rfm["T"])

# 3. Predict probability that each customer is still active (P(Alive))
rfm["p_alive"] = bgf.conditional_probability_alive(rfm["frequency"], rfm["recency"], rfm["T"])

# 4. Predict expected transactions over the next 52 weeks
rfm["pred_txs_52w"] = bgf.conditional_expected_number_of_purchases_up_to_time(
    52, rfm["frequency"], rfm["recency"], rfm["T"]
)

# 5. Fit Gamma-Gamma model for conditional expected order value
returning = rfm[rfm["frequency"] > 0]
ggf = GammaGammaFitter(penalizer_coef=0.01)
ggf.fit(returning["frequency"], returning["monetary_value"])

# 6. Calculate 12-Month Net Discounted Residual CLV
rfm["clv_12m"] = ggf.customer_lifetime_value(
    bgf, rfm["frequency"], rfm["recency"], rfm["T"], rfm["monetary_value"],
    time=12, discount_rate=0.01
)

import pandas as pd
from lifelines import KaplanMeierFitter, CoxPHFitter

# 1. Load subscription tenure logs
# churned = 1 if customer terminated; 0 if active today (right-censored)
df = pd.read_csv("subscriptions.csv")

# 2. Fit non-parametric Kaplan-Meier survival estimator
kmf = KaplanMeierFitter()
kmf.fit(durations=df["tenure_months"], event_observed=df["churned"], label="Enterprise Cohorts")

# 3. Calculate median time until 50% of the cohort has churned (Half-Life)
print(f"Median Customer Half-Life: {kmf.median_survival_time_:.1f} months")

# 4. Predict cohort survival probability at key milestone months
milestones = [1, 3, 6, 12, 24]
print("Milestone Survival Probabilities:")
print(kmf.predict(milestones))

# 5. Fit Cox Proportional Hazards to quantify risk multipliers (Hazard Ratios)
cph = CoxPHFitter()
cph.fit(df[["tenure_months", "churned", "onboarding_score", "support_wait_hrs"]], "tenure_months", "churned")
print(cph.summary[["exp(coef)", "p"]])
# exp(coef) > 1.0 indicates heightened hazard of churn

from mlxtend.frequent_patterns import apriori, association_rules

# 1. Transform transaction basket into boolean occurrence matrix
# (1 if customer purchased SKU, 0 otherwise)
frequent_itemsets = apriori(df_basket, min_support=0.02, use_colnames=True)

# 2. Extract actionable cross-sell rules with Lift > 1.5
rules = association_rules(frequent_itemsets, metric="lift", min_threshold=1.5)

# 3. Filter for high-confidence Next Best Action triggers
nba_rules = rules[rules["confidence"] > 0.60].sort_values("lift", ascending=False)
print(nba_rules[["antecedents", "consequents", "support", "confidence", "lift"]])

import pandas as pd

# Ingest paid seat allocations vs active telemetry over rolling 30 days
df = pd.read_csv("enterprise_accounts.csv")

# Compute instantaneous seat utilization ratio
df["seat_utilization"] = df["active_30d_users"] / df["paid_seats"]

# Flag silent churn accounts before annual renewal contract negotiations
silent_churn_risk = df[
    (df["seat_utilization"] < 0.35) & 
    (df["paid_seats"] >= 50)
].sort_values("paid_seats", ascending=False)

print(f"CRITICAL ALERT: {len(silent_churn_risk)} enterprise accounts flagged with ghost seats.")
# Automated webhook dispatched to CSM platform (e.g. Gainsight / Salesforce)