ML-Powered AML Checks: Detect Circular Transactions Before Loan Disbursal

A loan applicant submits three months of bank statements. The numbers look good. Regular credits, a healthy average balance, no obvious red flags, nothing a standard review would flag during bank statement analysis. The credit team approves. Disbursement goes through.

A few weeks later, repayment stress sets in. When someone investigates, the pattern emerges: the average balance was the same money rotating between four accounts. Each account credited the next. Each statement appeared clean in isolation. And by the time the funds reached the applicant’s account, no human reviewer could connect the dots.

This is what circular transactions look like in practice. They are a significant fraud risk that, when missed, contributes directly to loan defaults and NPA exposure. Without ML-powered AML checks embedded in the bank statement analysis process, they reach the disbursal stage far too often.

What Circular Transactions Actually Look Like

Circular transactions are a manufactured financial activity. Money moves between accounts, typically through two to five entities, in a loop designed to create the appearance of genuine business income or savings. The accounts involved may belong to family members, shell companies, associates, or other nominated proxies.

Viewed separately, the individual statements show nothing alarming. Each account receives money and sends money out. Balances look stable. But the flow has no economic purpose. Someone engineered it to inflate apparent creditworthiness, nothing more.

The Signals That Surface Before Disbursal

In practice, circular transactions tend to share a few recognisable characteristics:

• Round-figure flows. Large, round amounts moving through multiple accounts within 24 to 72 hours.
• Same-day or next-day outflows. Large credits followed almost immediately by debits of a similar or identical amount.
• Rotating counterparties. The same set of entities appears as both senders and receivers across different accounts.
• Dormant account activation. An account with minimal prior history suddenly receiving high-value transfers in the months before a loan application.

None of these signals is conclusive on its own. Together, they form a pattern that ML-powered AML systems are purpose-built to surface during bank statement analysis.

Why Manual Review Keeps Missing Them

The honest answer is volume. A credit analyst reviewing a single applicant’s three-month statement can catch obvious anomalies, but they cannot simultaneously cross-reference that statement against the counterparties who sent funds into the account.

When an applicant submits statements across two or three accounts, the complexity compounds. Tracing whether Account A’s large credit originated from Account B, which received it from Account C, which the applicant funded themselves, requires systematic cross-referencing that is not feasible at scale under normal processing timelines.

There is also a formatting problem. Bank statements arrive in hundreds of different formats across different institutions. Extracting consistent, comparable transaction data from each takes time and introduces the risk of error at every step. The fraud detection gap widens precisely where the operational pressure is highest.

How ML-Powered AML Detects What Manual Review Misses

ML-powered AML systems approach the problem differently. Rather than reviewing statements sequentially, they process all available accounts simultaneously, building a transaction graph that maps fund movement across every entity in the picture.

Cross-Account Pattern Matching

When multiple bank accounts are uploaded for analysis, an ML system can identify when the same funds appear to move through a chain of entities. It compares transaction amounts, timing intervals, and counterparty identities across every statement. If a large sum is left in Account B on a Tuesday and a near-identical amount arrives in Account A on Wednesday from an entity connected to Account B, the system flags it.

This kind of cross-referencing does not depend on the analyst knowing which accounts to compare. The system surfaces the connection automatically, which is what makes it effective as a fraud detection tool at scale.

Velocity and Timing Analysis

ML-powered AML also analyses the velocity of fund movement. The interval between a large credit and its corresponding debit tells a story. Legitimate business receipts tend to stay in the account or be distributed in ways that reflect genuine operational activity. Money in a rotation scheme moves out quickly, often the same day or within 48 hours.

Precisa’s AML analysis tracks FIFO (first in, first out) patterns across all uploaded accounts, identifies inter-bank transfers, and flags accounts where the ratio of credits to debits within short windows suggests rotation rather than genuine activity. The AML dashboard visualises the money trail, making it far easier for a credit officer or forensic investigator to read the flow without manually reconstructing it from raw data. For credit teams, DSAs, compliance officers, and forensic investigators, this changes what is actionable at the pre-disbursal stage.

What This Means for Lenders, DSAs, Compliance Teams, and Investigators

For lenders and direct selling agents, the practical implication is straightforward. Catching a circular transaction pattern before disbursal addresses a fraud risk at the point where it can still be acted on. After disbursement, the same detection becomes a recovery problem, which is a significantly worse position to be in.

For forensic auditors and AML compliance teams, ML-powered circular transaction detection compresses investigation timelines substantially. A forensic firm using Precisa reduced investigation time from 30–45 days to 25–30 minutes on complex multi-account cases, because the system handles cross-referencing and pattern identification that would otherwise take weeks of manual work.

Precisa’s bank statement analysis platform supports 850+ banks across 1,200+ bank formats and has processed over 15 lakh bank statements covering more than 510 crore transactions, across 1,000+ clients in 25+ countries. That breadth matters because circular transaction schemes often deliberately route funds through less-scrutinised institutions, including smaller co-operative banks that manual processes tend to handle inconsistently.

The Disbursal Window Is the Last Clear Checkpoint

Once funds leave, any ability to act on a fraud detection finding disappears. The loan disbursal risk window closes the moment disbursement is processed. ML-powered bank statement analysis at the pre-disbursal stage gives credit teams a clear answer on whether the income and balance figures in front of them reflect genuine financial activity, or whether the same funds have been moving in a loop.

It does not replace credit judgement; it gives that judgement something reliable to work with.

If your current process depends on a human analyst spotting circular patterns manually, across multiple accounts, in multiple formats, under time pressure, the gap between what is detectable and what is getting detected is probably wider than you think.Precisa’s AML analysis module is part of the bank statement analysis platform, accessible via web portal or API integration with existing lending and origination systems. To see how it works on a real statement, try Precisa for free now.