How Fraud Detection Is Evolving: Graphs, Shared Data and UPI
Fraud detection is moving in three directions. First, from single-transaction rules to models that look at behaviour over time. Second, from looking at one account to graph analytics that map links between accounts, devices and phone numbers, which is how mule networks are found.
Topic: AI in Cybersecurity in Cybersecurity & Ethical Hacking · Explainer
Also covers: Cyber Threats, Scams & Online Safety
Quick answer: Fraud detection is moving in three directions. First, from single-transaction rules to models that look at behaviour over time. Second, from looking at one account to graph analytics that map links between accounts, devices and phone numbers, which is how mule networks are found. Third, from each bank working alone to shared intelligence, such as the Department of Telecommunications Financial Fraud Risk Indicator that the RBI asked banks to integrate in June 2025. Real-time payments like UPI drive all three, because money moves in seconds and must be checked before it leaves, not after.
Key takeaways
- Instant payments leave almost no time to review after the fact.
- Graph analytics finds networks that single-account models miss.
- Shared risk signals help a bank see what other institutions have seen.
- Behavioural signals matter more as scam victims authorise payments themselves.
- Privacy law and explainability limit how data can be pooled and used.
Table of Contents
- How Has Fraud Detection Changed Over Time?
- Why Do Real-Time Payments Change Everything?
- What Is Graph Analytics in Fraud Detection?
- What Is Shared or Consortium Fraud Data?
- How Do You Detect Scams the Victim Authorises?
- What Limits These New Methods?
- What Should Fraud Teams Do Next?
- References
How Has Fraud Detection Changed Over Time?
| Approach | How it works | Weakness |
|---|---|---|
| Static rules | Block if amount or country matches a rule | Easy to learn and avoid |
| Supervised models | Learn from labelled past fraud | Slow to spot new scams |
| Behavioural analytics | Compare activity with the customer's normal pattern | Needs history; new customers are hard |
| Graph analytics | Analyse links between accounts, devices and numbers | Needs clean, connected data |
| Shared intelligence | Use risk signals from other institutions and agencies | Privacy, governance and data quality |
The FATF report on new technologies describes this shift and its challenges.
Why Do Real-Time Payments Change Everything?
UPI settles transfers almost instantly, around the clock. Once funds leave, criminals often move them through several mule accounts within minutes. Detection must therefore score the payment before it completes, using only signals available in that moment: device, location, payee history and recent account changes.
What Is Graph Analytics in Fraud Detection?
A graph treats accounts, phone numbers, devices and IP addresses as points and transactions or shared attributes as links. Patterns that look normal one account at a time become obvious as a network:
- Many new accounts sharing one device or phone number.
- Money fanning in from many victims and quickly fanning out.
- Short chains of accounts that each hold funds for minutes.
The RBI Innovation Hub's MuleHunter.AI, announced in December 2024, targets this mule account problem.
What Is Shared or Consortium Fraud Data?
No single bank sees the whole scam. Shared intelligence lets institutions use signals reported elsewhere. In India, the RBI advised banks on 30 June 2025 to integrate the Financial Fraud Risk Indicator from the Department of Telecommunications. It rates mobile numbers as medium, high or very high risk using inputs including cybercrime reports, so banks can delay or decline risky payments.
How Do You Detect Scams the Victim Authorises?
In investment scams, digital arrest calls and fake customer support, the real customer logs in and approves the payment. Passwords and OTPs do not help. Useful signals include:
| Signal | Why it helps |
|---|---|
| Long active phone call during a payment | Suggests coaching by a scammer |
| First payment to a high-risk payee | Mule accounts are often new |
| Sudden large transfers from savings | Unusual for the customer |
| Payee or number flagged in shared data | Already linked to reports |
For card payments, strong customer authentication through EMV 3-D Secure adds risk data to each transaction. See also UPI and card transaction security controls and how banks catch mule accounts.
What Limits These New Methods?
- Privacy: pooling customer data must respect the DPDP Act 2023.
- False positives: delaying genuine urgent payments harms customers.
- Explainability: graph scores can be hard to explain to customers and regulators.
- Adaptation: criminals rotate numbers, devices and mule accounts.
What Should Fraud Teams Do Next?
- Map which signals are available before a payment completes.
- Connect fraud, AML and cyber data so graphs can be built.
- Join available shared intelligence schemes and document governance.
- Track model drift and false positives using the NIST AI RMF.
- Tell customers how to report: 1930 and cybercrime.gov.in.
WebAsha offers cybersecurity and fraud analytics training for teams building these capabilities.
References
Frequently Asked Questions
What's Your Reaction?
Like
0
Dislike
0
Love
0
Funny
0
Wow
0
Sad
0
Angry
0