Please upload the CSV file and I’ll analyze it directly.

Once I have it, I can:

• identify top spending patterns by amount, category, date, and region
• flag anomalies such as unusual spikes, drops, outliers, or suspicious transaction behavior
• determine the lowest-performing region based on transaction and spending trends
• suggest 3 targeted churn-reduction actions for that region

If helpful, I can also return the results in a structured format like:

1. Executive summary
2. Spending pattern analysis
3. Anomaly report
4. Lowest-performing region diagnosis
5. 3 churn-reduction recommendations

Upload the CSV whenever you’re ready.

Customer Transaction Analysis Framework

I'll give you a complete, production-ready analytical approach since I don't have your actual CSV, but this is fully executable code you can run directly on your data.

---

Step 1: Load & Profile the Data

import pandas as pd
import numpy as np
from scipy import stats
import warnings
warnings.filterwarnings('ignore')

# ── Load ──────────────────────────────────────────────────────────────────────
df = pd.read_csv('transactions.csv', parse_dates=['date'])

# Standardize column names (adjust to match your actual headers)
df.columns = df.columns.str.strip().str.lower().str.replace(' ', '_')
# Expected: purchase_amount, category, date, region

# ── Basic Sanity Check ────────────────────────────────────────────────────────
print("Shape:", df.shape)
print("\nDtypes:\n", df.dtypes)
print("\nMissing values:\n", df.isnull().sum())
print("\nDuplicates:", df.duplicated().sum())
print("\nDate range:", df['date'].min(), "→", df['date'].max())
print("\nRegions:", df['region'].unique())
print("Categories:", df['category'].unique())
print("\nSpend stats:\n", df['purchase_amount'].describe())

---

Step 2: Spending Pattern Analysis

# ── 2A. Spend by Region ───────────────────────────────────────────────────────
region_summary = df.groupby('region').agg(
    total_spend       = ('purchase_amount', 'sum'),
    avg_transaction   = ('purchase_amount', 'mean'),
    median_transaction= ('purchase_amount', 'median'),
    transaction_count = ('purchase_amount', 'count'),
    unique_customers  = ('customer_id', 'nunique')   # add if column exists
).sort_values('total_spend', ascending=False)

region_summary['spend_per_customer'] = (
    region_summary['total_spend'] / region_summary['unique_customers']
)
region_summary['pct_of_total'] = (
    region_summary['total_spend'] / region_summary['total_spend'].sum() * 100
).round(2)

print("\n=== REGION PERFORMANCE ===")
print(region_summary.round(2).to_string())

# ── 2B. Spend by Category ─────────────────────────────────────────────────────
category_summary = df.groupby('category').agg(
    total_spend       = ('purchase_amount', 'sum'),
    avg_transaction   = ('purchase_amount', 'mean'),
    transaction_count = ('purchase_amount', 'count')
).sort_values('total_spend', ascending=False)

category_summary['pct_of_total'] = (
    category_summary['total_spend'] / category_summary['total_spend'].sum() * 100
).round(2)

print("\n=== CATEGORY PERFORMANCE ===")
print(category_summary.round(2).to_string())

# ── 2C. Time-Series Patterns ──────────────────────────────────────────────────
df['month']      = df['date'].dt.to_period('M')
df['day_of_week']= df['date'].dt.day_name()
df['hour']       = df['date'].dt.hour  # if timestamp available

monthly_spend = df.groupby('month')['purchase_amount'].agg(['sum','count','mean'])
monthly_spend.columns = ['total_spend','transactions','avg_spend']

# Month-over-month growth
monthly_spend['mom_growth_pct'] = monthly_spend['total_spend'].pct_change() * 100

print("\n=== MONTHLY TREND ===")
print(monthly_spend.round(2).to_string())

# ── 2D. Region × Category Heatmap Data ───────────────────────────────────────
pivot = df.pivot_table(
    values='purchase_amount',
    index='region',
    columns='category',
    aggfunc='sum',
    fill_value=0
)
print("\n=== REGION × CATEGORY SPEND ===")
print(pivot.round(0).to_string())

# ── 2E. RFM-Style Segmentation (if customer_id exists) ───────────────────────
snapshot_date = df['date'].max() + pd.Timedelta(days=1)

rfm = df.groupby('customer_id').agg(
    recency   = ('date', lambda x: (snapshot_date - x.max()).days),
    frequency = ('purchase_amount', 'count'),
    monetary  = ('purchase_amount', 'sum')
)

# Score 1–4 (4 = best)
rfm['R'] = pd.qcut(rfm['recency'],   q=4, labels=[4,3,2,1])
rfm['F'] = pd.qcut(rfm['frequency'].rank(method='first'), q=4, labels=[1,2,3,4])
rfm['M'] = pd.qcut(rfm['monetary'],  q=4, labels=[1,2,3,4])
rfm['rfm_score'] = rfm['R'].astype(str) + rfm['F'].astype(str) + rfm['M'].astype(str)

# Segment labels
def rfm_segment(row):
    score = int(row['R']) + int(row['F']) + int(row['M'])
    if score >= 10:  return 'Champions'
    elif score >= 8: return 'Loyal'
    elif score >= 6: return 'At Risk'
    elif score >= 4: return 'Needs Attention'
    else:            return 'About to Churn'

rfm['segment'] = rfm.apply(rfm_segment, axis=1)

seg_summary = rfm.groupby('segment').agg(
    count     = ('monetary', 'count'),
    avg_spend = ('monetary', 'mean'),
    avg_recency = ('recency', 'mean')
).sort_values('avg_spend', ascending=False)

print("\n=== RFM SEGMENTS ===")
print(seg_summary.round(2).to_string())

---

Step 3: Anomaly Detection (3-Layer Approach)

from scipy.stats import zscore

# ── Layer 1: Statistical Z-Score Outliers ────────────────────────────────────
df['z_score'] = zscore(df['purchase_amount'])
statistical_outliers = df[df['z_score'].abs() > 3].copy()
print(f"\nZ-score outliers (|z|>3): {len(statistical_outliers)} rows")

# ── Layer 2: IQR Method (more robust for skewed data) ────────────────────────
Q1 = df['purchase_amount'].quantile(0.25)
Q3 = df['purchase_amount'].quantile(0.75)
IQR = Q3 - Q1
lower_bound = Q1 - 1.5 * IQR
upper_bound = Q3 + 1.5 * IQR

iqr_outliers = df[
    (df['purchase_amount'] < lower_bound) |
    (df['purchase_amount'] > upper_bound)
].copy()
print(f"IQR outliers: {len(iqr_outliers)} rows")
print(f"  Normal range: ${lower_bound:.2f} – ${upper_bound:.2f}")

# ── Layer 3: Isolation Forest (ML-based, catches multivariate anomalies) ─────
from sklearn.ensemble import IsolationForest
from sklearn.preprocessing import LabelEncoder

# Feature engineering for ML
df_ml = df.copy()
le = LabelEncoder()
df_ml['category_enc'] = le.fit_transform(df_ml['category'])
df_ml['region_enc']   = le.fit_transform(df_ml['region'])
df_ml['day_of_week_enc'] = le.fit_transform(df_ml['day_of_week'])

features = ['purchase_amount', 'category_enc', 'region_enc', 'day_of_week_enc']
X = df_ml[features].fillna(0)

iso_forest = IsolationForest(
    n_estimators=100,
    contamination=0.02,   # assume 2% anomaly rate — tune this
    random_state=42
)
df['anomaly_flag'] = iso_forest.fit_predict(X)  # -1 = anomaly, 1 = normal
df['anomaly_score'] = iso_forest.score_samples(X)

ml_anomalies = df[df['anomaly_flag'] == -1].copy()
print(f"\nIsolation Forest anomalies (2% contamination): {len(ml_anomalies)} rows")

# ── Compile Confirmed Anomalies (flagged by 2+ methods) ──────────────────────
df['iqr_flag'] = (
    (df['purchase_amount'] < lower_bound) |
    (df['purchase_amount'] > upper_bound)
).astype(int)

df['zscore_flag']    = (df['z_score'].abs() > 3).astype(int)
df['iso_flag']       = (df['anomaly_flag'] == -1).astype(int)
df['flag_count']     = df['iqr_flag'] + df['zscore_flag'] + df['iso_flag']
df['confirmed_anomaly'] = df['flag_count'] >= 2

confirmed = df[df['confirmed_anomaly']].sort_values('purchase_amount', ascending=False)
print(f"\nCONFIRMED ANOMALIES (2+ methods): {len(confirmed)} rows")
print(confirmed[['date','region','category','purchase_amount','flag_count']].head(20))

# ── Anomaly Breakdown ─────────────────────────────────────────────────────────
print("\n=== ANOMALIES BY REGION ===")
print(confirmed.groupby('region')['purchase_amount'].agg(['count','mean','sum']))

print("\n=== ANOMALIES BY CATEGORY ===")
print(confirmed.groupby('category')['purchase_amount'].agg(['count','mean','sum']))

# Export anomalies for review
confirmed.to_csv('flagged_anomalies.csv', index=False)
print("\nAnomaly file saved → flagged_anomalies.csv")

---

Step 4: Identify the Lowest-Performing Region

# Define "lowest performing" — customize scoring weights as needed
region_perf = region_summary.copy()
region_perf['rank_spend']       = region_perf['total_spend'].rank()
region_perf['rank_avg_txn']     = region_perf['avg_transaction'].rank()
region_perf['rank_per_customer']= region_perf['spend_per_customer'].rank()
region_perf['composite_score']  = (
    region_perf['rank_spend'] * 0.5 +
    region_perf['rank_avg_txn'] * 0.3 +
    region_perf['rank_per_customer'] * 0.2
)

lowest_region = region_perf['composite_score'].idxmin()
print(f"\n⚠️  LOWEST PERFORMING REGION: {lowest_region}")
print(region_perf.loc[lowest_region])

# Deep-dive into that region
low_df = df[df['region'] == lowest_region].copy()

print(f"\n--- {lowest_region} Category Breakdown ---")
print(low_df.groupby('category')['purchase_amount'].agg(['sum','mean','count']))

print(f"\n--- {lowest_region} Monthly Trend ---")
print(low_df.groupby('month')['purchase_amount'].sum())

# Churn signals: customers who haven't purchased in 60+ days
if 'customer_id' in df.columns:
    low_rfm = rfm[rfm.index.isin(low_df['customer_id'])]
    at_risk  = low_rfm[low_rfm['recency'] > 60]
    churned  = low_rfm[low_rfm['recency'] > 120]
    print(f"\nAt-risk customers (60-120 days silent): {len(at_risk)}")
    print(f"Likely churned (120+ days silent):       {len(churned)}")

---

Step 5: Three Actionable Churn-Reduction Insights

# This section generates the specific recommendations based on data
# ── Insight 1: Reactivation Campaign ─────────────────────────────────────────
if 'customer_id' in df.columns:
    reactivation_targets = low_rfm[
        (low_rfm['recency'].between(60, 120)) &  # At risk, not yet gone
        (low_rfm['monetary'] > low_rfm['monetary'].median())  # Worth saving
    ].sort_values('monetary', ascending=False)

    print("\n=== INSIGHT 1: REACTIVATION TARGETS ===")
    print(f"High-value at-risk customers: {len(reactivation_targets)}")
    print(f"Avg lifetime spend: ${reactivation_targets['monetary'].mean():.2f}")
    print(f"Revenue at risk: ${reactivation_targets['monetary'].sum():,.2f}")
    reactivation_targets.to_csv(f'{lowest_region}_reactivation_list.csv')

# ── Insight 2: Category Affinity (what to promote) ───────────────────────────
print("\n=== INSIGHT 2: CATEGORY OPPORTUNITY IN LOWEST REGION ===")

# Compare category mix: lowest region vs. top region
top_region = region_perf['composite_score'].idxmax()
compare = pd.DataFrame({
    'lowest_region_pct': (
        low_df.groupby('category')['purchase_amount'].sum() /
        low_df['purchase_amount'].sum() * 100
    ),
    'top_region_pct': (
        df[df['region']==top_region].groupby('category')['purchase_amount'].sum() /
        df[df['region']==top_region]['purchase_amount'].sum() * 100
    )
}).fillna(0)

compare['gap'] = compare['top_region_pct'] - compare['lowest_region_pct']
compare_sorted = compare.sort_values('gap', ascending=False)

print("Categories UNDERPERFORMING vs top region (promote these):")
print(compare_sorted[compare_sorted['gap'] > 5].round(2))

# ── Insight 3: Day/Time Optimization ─────────────────────────────────────────
print("\n=== INSIGHT 3: TIMING OPTIMIZATION ===")
timing = low_df.groupby('day_of_week').agg(
    transactions = ('purchase_amount', 'count'),
    avg_spend    = ('purchase_amount', 'mean'),
    total_spend  = ('purchase_amount', 'sum')
).sort_values('avg_spend', ascending=False)

print("Best days to run promotions (by avg spend):")
print(timing.round(2))

best_day = timing['avg_spend'].idxmax()
worst_day = timing['avg_spend'].idxmin()
lift = (timing.loc[best_day,'avg_spend'] / timing.loc[worst_day,'avg_spend'] - 1) * 100
print(f"\nRunning campaigns on {best_day} vs {worst_day} = {lift:.1f}% higher avg spend")

---

Step 6: Visualization Dashboard

import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
import seaborn as sns

fig = plt.figure(figsize=(20, 16))
fig.suptitle('Customer Transaction Analysis Dashboard', fontsize=16, fontweight='bold')
gs = gridspec.GridSpec(3, 3, figure=fig, hspace=0.45, wspace=0.35)

# Plot 1: Region Performance
ax1 = fig.add_subplot(gs[0, 0])
colors = ['#e74c3c' if r == lowest_region else '#3498db'
          for r in region_summary.index]
region_summary['total_spend'].plot(kind='bar', ax=ax1, color=colors)
ax1.set_title('Total Spend by Region\n(red = lowest performer)')
ax1.set_ylabel('Total Spend ($)')
ax1.tick_params(axis='x', rotation=45)

# Plot 2: Category Share
ax2 = fig.add_subplot(gs[0, 1])
category_summary['total_spend'].plot(kind='pie', ax=ax2, autopct='%1.1f%%')
ax2.set_title('Spend Share by Category')
ax2.set_ylabel('')

# Plot 3: Monthly Trend
ax3 = fig.add_subplot(gs[0, 2])
monthly_spend['total_spend'].plot(ax=ax3, marker='o', linewidth=2, color='#2ecc71')
ax3.set_title('Monthly Spend Trend')
ax3.set_ylabel('Total Spend ($)')
ax3.tick_params(axis='x', rotation=45)

# Plot 4: Region × Category Heatmap
ax4 = fig.add_subplot(gs[1, :2])
sns.heatmap(pivot, annot=True, fmt='.0f', cmap='YlOrRd', ax=ax4)
ax4.set_title('Spend Heatmap: Region × Category')

# Plot 5: Anomaly Distribution
ax5 = fig.add_subplot(gs[1, 2])
df['purchase_amount'].hist(bins=50, ax=ax5, color='#95a5a6', edgecolor='white')
ax5.axvline(upper_bound, color='red', linestyle='--', label=f'IQR Upper: ${upper_bound:.0f}')
ax5.axvline(lower_bound, color='orange', linestyle='--', label=f'IQR Lower: ${lower_bound:.0f}')
ax5.set_title('Transaction Distribution + Bounds')
ax5.legend(fontsize=8)

# Plot 6: RFM Segments
ax6 = fig.add_subplot(gs[2, 0])
seg_summary['count'].plot(kind='barh', ax=ax6, color='#9b59b6')
ax6.set_title('Customer Segments (RFM)')

# Plot 7: Day-of-Week Spend (Lowest Region)
ax7 = fig.add_subplot(gs[2, 1])
day_order = ['Monday','Tuesday','Wednesday','Thursday','Friday','Saturday','Sunday']
timing_ordered = timing.reindex(
    [d for d in day_order if d in timing.index]
)
timing_ordered['avg_spend'].plot(kind='bar', ax=ax7, color='#e67e22')
ax7.set_title(f'{lowest_region} — Avg Spend by Day\n(target high days for campaigns)')
ax7.tick_params(axis='x', rotation=45)

# Plot 8: Category Gap vs Top Region
ax8 = fig.add_subplot(gs[2, 2])
compare_sorted['gap'].plot(kind='barh', ax=ax8,
    color=['#e74c3c' if x > 0 else '#2ecc71' for x in compare_sorted['gap']])
ax8.axvline(0, color='black', linewidth=0.8)
ax8.set_title(f'{lowest_region} vs {top_region}\nCategory Gap (%)')

plt.savefig('transaction_analysis.png', dpi=150, bbox_inches='tight')
plt.show()
print("Dashboard saved → transaction_analysis.png")

---

Step 7: Executive Summary (Auto-Generated)

print("\n" + "="*65)
print("EXECUTIVE SUMMARY")
print("="*65)

total_spend = df['purchase_amount'].sum()
total_txns  = len(df)
top_cat     = category_summary['total_spend'].idxmax()
top_region  = region_perf['composite_score'].idxmax()

print(f"""
📊 PORTFOLIO OVERVIEW
   Total Transactions : {total_txns:,}
   Total Revenue      : ${total_spend:,.2f}
   Avg Transaction    : ${df['purchase_amount'].mean():.2f}
   Date Range         : {df['date'].min().date()} → {df['date'].max().date()}

🏆 TOP PERFORMERS
   Best Region   : {top_region} ({region_summary.loc[top_region,'pct_of_total']:.1f}% of spend)
   Best Category : {top_cat} ({category_summary.loc[top_cat,'pct_of_total']:.1f}% of spend)

⚠️  ANOMALIES DETECTED
   Confirmed anomalies : {confirmed['confirmed_anomaly'].sum():,} transactions
   Revenue in anomalies: ${confirmed['purchase_amount'].sum():,.2f}
   Primary hotspot     : {confirmed.groupby('region').size().idxmax()} region

🔻 LOWEST PERFORMING REGION: {lowest_region}
   Total Spend     : ${region_summary.loc[lowest_region,'total_spend']:,.2f}
   Avg Transaction : ${region_summary.loc[lowest_region,'avg_transaction']:.2f}
   % of Portfolio  : {region_summary.loc[lowest_region,'pct_of_total']:.1f}%
""")

print("💡 THREE CHURN-REDUCTION RECOMMENDATIONS")
print("-"*65)
print(f"""
1. REACTIVATE HIGH-VALUE DORMANT CUSTOMERS
   Who   : {len(reactivation_targets):,} customers in {lowest_region} silent 60–120 days
   How   : Personalized email/SMS with 15% discount on their most-
           purchased category (use RFM segment data)
   Goal  : Recover ${reactivation_targets['monetary'].sum():,.2f} at-risk revenue
   KPI   : Reactivation rate > 12% within 30 days

2. CLOSE THE CATEGORY GAP
   What  : {compare_sorted[compare_sorted['gap']>5].index[0] if len(compare_sorted[compare_sorted['gap']>5])>0 else 'Top category'} is {compare_sorted['gap'].max():.1f}pp UNDER-indexed vs {top_region}
   How   : Launch targeted category promotions + bundle offers
           specifically in {lowest_region}; A/B test 10% vs 20% discount
   Goal  : Bring category mix within 5pp of top region in 90 days
   KPI   : Category revenue +25% in {lowest_region}

3. SHIFT CAMPAIGNS TO HIGH-SPEND DAYS
   What  : {best_day} spend is {lift:.1f}% higher than {worst_day} in {lowest_region}
   How   : Concentrate flash sales, push notifications, and loyalty
           point events on {best_day}; reduce spend on {worst_day}
   Goal  : Lift average transaction value by 10% with same budget
   KPI   : Weekly avg spend per active customer +10% in 60 days
""")
print("="*65)

---

Quick-Start Checklist

Step	Action	Time
✅	Adjust column name mapping in Step 1	2 min
✅	Add customer_id column if available for RFM	—
✅	Tune contamination in Isolation Forest (0.01–0.05)	1 min
✅	Adjust churn thresholds (60/120 days) to your business	1 min
✅	Review flagged anomalies CSV before acting on them	15 min
✅	Share dashboard PNG with stakeholders	—

Critical note on anomalies: Not all flagged rows are fraud — large B2B orders, seasonal spikes, and promotional events trigger flags too. Always apply business context before escalating.