IDBI Bank ESO study materials
IDBI Bank ESO study resources

Rule-based fraud triage: six alerts, one gap in the rulebook

IDBI Bank ESO · Put your knowledge to work in context.

All case studies

Rule-based fraud triage: six alerts, one gap in the rulebook

Your bank's fraud-monitoring cell runs four real-time rules on retail transactions:

  • R1 - amount above ₹2,00,000 AND device first seen within 7 days AND beneficiary account less than 30 days old → auto-block.
  • R2 - amount between ₹50,000 and ₹2,00,000 AND transaction between 23:00 and 06:00 AND transaction location different from registered address → hold and call the customer.
  • R3 - beneficiary appears on the mule-account list → auto-block, irrespective of amount.
  • R4 - KYC verified AND relationship of 3 years or more AND amount ₹1,00,000 or less AND beneficiary used successfully before → auto-approve.

Precedence: R3 > R1 > R2 > R4.

Six alerts arrived on 14 March:

  • A1: ₹2,40,000; device first seen 2 days ago; beneficiary 12 days old; customer has a 5-year relationship.
  • A2: ₹1,60,000 at 02:15; location Pune, registered address Nashik; beneficiary used twice before.
  • A3: ₹40,000; beneficiary on the mule list; device first seen yesterday.
  • A4: ₹85,000 at 19:00; KYC verified; 4-year relationship; beneficiary used 4 times before; known device.
  • A5: ₹2,10,000 at 01:00; location Pune, registered address Nashik; device 45 days old; beneficiary 6 months old; not on the mule list.
  • A6: ₹95,000 at 03:00; location Pune, registered address Nashik; KYC verified; 4-year relationship; beneficiary never used before.

1. Classify each alert with the rule that governs it, applying the precedence order. Identify any alert that the rulebook does not resolve.

Read the worked answer

Test each alert against all four rules before applying precedence.

  • A1 → R1 (auto-block). ₹2.4 lakh > ₹2 lakh, device 2 days old (< 7), beneficiary 12 days old (< 30). All three limbs satisfied. R2 is not triggered because the amount is above ₹2 lakh; R3 is not stated.
  • A2 → R2 (hold and call). ₹1.6 lakh lies within ₹50,000–₹2 lakh, time 02:15 is inside the 23:00–06:00 window, and Pune ≠ Nashik. R4 cannot apply because of the amount and the night-time condition does not matter for R4, but R2 outranks R4 anyway.
  • A3 → R3 (auto-block). The beneficiary is on the mule list, so R3 fires irrespective of amount and outranks everything. R1 also happens to be satisfied (device new), which confirms the block.
  • A4 → R4 (auto-approve). KYC verified, 4-year relationship ≥ 3 years, ₹85,000 ≤ ₹1 lakh, beneficiary used 4 times. R2 fails because 19:00 is outside the night window; no other rule fires. This is the only clean auto-approve.
  • A5 → unresolved. R1 fails (device 45 days old). R2 fails on amount (₹2.1 lakh is above ₹2 lakh). R3 fails (not a mule). R4 fails on amount. No rule covers it → it must fall to a default manual review.
  • A6 → R2 (hold and call). ₹95,000 is in range, 03:00 is in the night window, location differs. R4 fails because the beneficiary has never been used before.

The gap is A5: a high-value, odd-hours, geo-mismatched transaction with a 6-month-old beneficiary is the exact profile of an account-takeover, yet the rulebook passes it because R1 was written narrowly around new devices and R2 was capped at ₹2 lakh.

2. Which single rule should be amended, and how, so that A5-type alerts are caught without blocking genuine customers? Justify the change against the alerts given.

Read the worked answer

Amend R2, not R1. The defect is the upper bound: risk does not fall when the amount rises above ₹2 lakh, so capping R2 at ₹2,00,000 creates a blind spot exactly where the loss would be largest. Suggested form:

R2 (revised) - amount above ₹50,000 AND transaction between 23:00 and 06:00 AND transaction location different from registered address AND beneficiary not previously used successfully → hold and call the customer.

Effect on the six alerts: A5 now becomes a hold-and-call (₹2.1 lakh, 01:00, Pune vs Nashik, beneficiary never used). A2 still holds, A6 still holds. A4 is untouched because 19:00 is outside the night window. A1 is still blocked first by R1, so precedence is unchanged. Nothing that was auto-approved becomes blocked - the change only converts silent passes into a call, which is a low-friction control.

Why not widen R1 instead? Lowering the ₹2 lakh threshold or broadening "new device" would block routine salary-day transactions and cause more false positives than R2's night-plus-geo-mismatch combination, which is already narrow. A second, structural fix worth recommending is a default rule: any transaction above ₹2 lakh that satisfies no rule goes to manual review within 30 minutes, so the rulebook is closed rather than open-ended.

3. An alert was auto-approved on 14 March. Which conditions must necessarily be true, and which rule conditions can you conclude were not fully satisfied? Show the inference chain.

Read the worked answer

Step 1 - which rule can produce an auto-approve? Only R4. So the four R4 limbs must all be true: KYC verified; relationship ≥ 3 years; amount ≤ ₹1,00,000; beneficiary used successfully before.

Step 2 - because precedence is R3 > R1 > R2 > R4, an auto-approve means no higher rule could have fired.

  • R3 not satisfied → beneficiary is not on the mule list (this one is certain and absolute).
  • R1 not satisfied → but amount ≤ ₹1 lakh already makes the "above ₹2 lakh" limb false, so R1 is excluded automatically and tells us nothing further.
  • R2 not satisfied → amount ≤ ₹1 lakh means the amount limb can be satisfied (₹50,000–₹1,00,000 is inside R2's band), so at least one of the other two limbs must be false: either the transaction was not between 23:00 and 06:00, or the transaction location equals the registered address (or both).

Step 3 - a common trap. It is tempting to conclude "auto-approved, therefore it was a daytime transaction". That does not follow: a ₹70,000 transaction at 02:00 from the registered address would also be auto-approved, because the geo limb fails. The only logically forced conclusions are the four R4 limbs plus "not a mule beneficiary" plus "not (night-time AND different location)".

Answer: certain - KYC verified, ≥ 3-year relationship, amount ≤ ₹1 lakh, beneficiary previously used, beneficiary not on the mule list, and not both night-time and geo-mismatched; not certain - the specific time, the specific location, or the device age.