How Data Analytics Is Used in Indian Fintech Companies Today

-- Payment success rate by bank for one day
SELECT
  bank,
  COUNT(*) AS attempts,
  SUM(CASE WHEN status = 'SUCCESS' THEN 1 ELSE 0 END) AS successes,
  ROUND(100.0 * SUM(CASE WHEN status = 'SUCCESS' THEN 1 ELSE 0 END)
        / COUNT(*), 2) AS success_pct
FROM transactions
WHERE txn_date = '2026-11-05'
GROUP BY bank
HAVING COUNT(*) >= 100
ORDER BY success_pct ASC;

import pandas as pd

tx = pd.DataFrame({
    "status": ["SUCCESS", "FAILED", "FAILED", "SUCCESS", "FAILED", "SUCCESS"],
    "error":  [None, "BANK_TIMEOUT", "WRONG_PIN", None, "BANK_TIMEOUT", None],
})
user_errors = {"WRONG_PIN", "INSUFFICIENT_FUNDS"}

failed = tx[tx["status"] == "FAILED"].copy()
failed["type"] = failed["error"].apply(
    lambda e: "user" if e in user_errors else "system")
print(failed.groupby(["type", "error"]).size())

import pandas as pd

steps = pd.DataFrame({
    "step": ["Started", "KYC done", "Offer shown", "Accepted", "Disbursed"],
    "users": [10000, 7200, 5100, 2300, 2050],
})
steps["from_prev_%"] = (steps["users"] / steps["users"].shift(1) * 100).round(1)
steps["from_start_%"] = (steps["users"] / steps["users"].iloc[0] * 100).round(1)
print(steps)
# The biggest drop shows where to investigate first.

-- Share of loans more than 30 days late, by disbursal month
-- and months since disbursal (dummy schema)
SELECT
  strftime('%Y-%m', disbursed_on) AS cohort,
  months_since_disbursal,
  ROUND(100.0 * AVG(CASE WHEN days_past_due > 30 THEN 1 ELSE 0 END), 2)
    AS dpd30_pct
FROM loan_monthly_status
GROUP BY cohort, months_since_disbursal
ORDER BY cohort, months_since_disbursal;

import pandas as pd

tx = pd.DataFrame({
    "device": ["d1"] * 6 + ["d2"] * 2,
    "ts": pd.to_datetime([
        "2026-11-05 10:01", "10:03", "10:04", "10:05", "10:06", "10:07",
        "2026-11-05 11:00", "15:00"], format="mixed"),
})
tx["hour"] = tx["ts"].dt.floor("h")
counts = tx.groupby(["device", "hour"]).size().rename("txns").reset_index()
print(counts[counts["txns"] >= 5])   # candidates for review, not auto-block

import pandas as pd

ev = pd.DataFrame({
    "user": ["u1", "u1", "u2", "u3", "u3", "u3"],
    "week": [0, 1, 0, 0, 1, 2],   # weeks since first payment
})
cohort_size = ev[ev["week"] == 0]["user"].nunique()
retention = (ev.groupby("week")["user"].nunique() / cohort_size * 100).round(1)
print(retention)

transactions(txn_id, user_id, txn_time, amount, bank,
             method, app_version, status, error_code)

loans(loan_id, user_id, channel, disbursed_on, amount)

loan_monthly_status(loan_id, months_since_disbursal,
                    days_past_due)

Generate rows with Python (random + faker) for practice.

Payments dashboard
- Success rate today vs last 7 days
- Success rate by bank and method
- Top 5 error codes (user vs system)
- Hourly attempts and failures

Filters: date, bank, method, app version

Case: Payment success fell on 5 Nov

Finding: Drop concentrated in one bank, 7-9 pm
Evidence: Bank success 88% vs 97% usual; others normal
Error: BANK_TIMEOUT made up most failures
Action: Share with payments team for partner follow-up
Caveat: Based on one day; monitor next 3 days