FinMonitorSmall scripts.
Clearer signals.

One batch. Three deliberate steps.

Each script writes the file the next script reads. Run order matters; orchestration is a future extension.

1. INPUTtransactions.csvSimulated records
2. 01 / CLEANclean_data.pyPrepare the dataset
3. 02 / DETECTanalyze_data.pyCalculate and flag
4. 03 / REPORTreport.pySummarise for review

The Python behind the pipeline.

The three code excerpts supplied for this demo, with formatting restored for readability. They use pandas as pd; the excerpts omit the import and assume the input files and output folder already exist.

Original comments and output labels are retained. “Company logic” labels this illustrative decision; it does not establish Deloitte ownership. “High-risk” merchant labels and “amount at risk” are demo terminology, not verified risk assessments or losses.

# 1. Load raw data
df = pd.read_csv('../data/transactions.csv')

# 2. Standardize column names
df.columns = df.columns.str.lower().str.strip()

# 3. Remove exact duplicate rows
df = df.drop_duplicates()

# 4. Convert date column to datetime
df['date'] = pd.to_datetime(df['date'], errors='coerce')

# 5. Handle missing or invalid amounts
df['amount'] = pd.to_numeric(df['amount'], errors='coerce')
df['amount_was_missing'] = df['amount'].isna()
median_amount = df['amount'].median()
df['amount'] = df['amount'].fillna(median_amount)

# 6. Remove transactions with missing critical identifiers
df = df.dropna(subset=['transaction_id', 'customer_id', 'merchant'])

# 7. Remove negative or zero-value transactions
df = df[df['amount'] > 0]

# 8. Sort for consistency (audit-friendly)
df = df.sort_values(by=['customer_id', 'date'])

# 9. Save cleaned data
df.to_csv('../output/cleaned_data.csv', index=False)

# 1. Load cleaned data
df = pd.read_csv('../output/cleaned_data.csv')

# 2. Customer-level baseline
avg_spend = df.groupby('customer_id')['amount'].mean()
df['avg_customer_spend'] = df['customer_id'].map(avg_spend)

# 3. Rule 1: Spend spike anomaly
df['spend_anomaly'] = df['amount'] > (2 * df['avg_customer_spend'])

# 4. Rule 2: High absolute transaction value
df['high_value_anomaly'] = df['amount'] > 20000

# 5. Rule 3: High-risk merchant
high_risk_merchants = ['Apple', 'Dell']
df['merchant_risk'] = df['merchant'].isin(high_risk_merchants)

# 6. Final anomaly flag (company logic)
df['final_anomaly'] = (
    df['spend_anomaly'] |
    (df['high_value_anomaly'] & df['merchant_risk'])
)

# 7. Save outputs
df[df['final_anomaly']].to_csv('../output/anomalies.csv', index=False)

# 1. Load anomaly data
df = pd.read_csv('../output/anomalies.csv')

# 2. Handle case where no anomalies exist
if df.empty:
    print("No anomalies detected for this period.")
else:
    # 3. Summary metrics
    total_anomalies = len(df)
    unique_customers = df['customer_id'].nunique()
    total_amount = df['amount'].sum()
    avg_anomaly_amount = df['amount'].mean()

    print(f"Total flagged transactions : {total_anomalies}")
    print(f"Unique customers impacted : {unique_customers}")
    print(f"Total amount at risk (₹)  : {total_amount:,.2f}")
    print(f"Average anomaly amount    : {avg_anomaly_amount:,.2f}")

    # 4. Top risky merchants
    print("\nTop merchants by anomaly count:")
    print(df['merchant'].value_counts().head())

    # 5. Top customers by total anomaly spend
    print("\nTop customers by anomaly spend:")
    print(
        df.groupby('customer_id')['amount']
          .sum()
          .sort_values(ascending=False)
          .head()
    )

    # 6. Save daily summary
    summary_df = pd.DataFrame([{
        'total_anomalies': total_anomalies,
        'unique_customers': unique_customers,
        'total_amount_at_risk': total_amount,
        'average_anomaly_amount': avg_anomaly_amount
    }])
    summary_df.to_csv('../output/daily_anomaly_summary.csv', index=False)

A useful handoff, not a black box.

＋ Console output adds the summary and top merchants/customers. This is a reporting channel, not a fourth output file. Flagged amounts are review totals, not confirmed financial losses.