[ Credit Analysis ]

UPI Transaction Analysis in Bank Statements: Narration, RRN and VPAs

UPI transaction analysis for lenders: parse the narration, match self transfers on the RRN, resolve VPAs to counterparties and read merchant income safely.

Zeus Dhanbhoora

10 min read · 12 November 2026

UPI Transaction Analysis in Bank Statements: Narration, RRN and VPAs
Contents
  1. Why the same UPI payment reads differently in every statement
  2. Match self transfers on the RRN, then confirm
  3. A VPA is a routing address, not an identity
  4. P2M versus P2P does not mean business versus personal
  5. Payment aggregator and marketplace settlements hide the counterparties
  6. UPI Autopay moves obligations onto UPI rails
  7. UPI receipts are the best income evidence a micro merchant has had
  8. What competent UPI handling requires
  9. What UPI still cannot tell you
  10. Frequently asked questions

UPI transaction analysis in a bank statement means parsing each UPI narration into its fields (the 12 digit reference number, the counterparty name, the bank, the UPI ID and the remark) so that turnover, obligations and counterparties are computed on what each payment was, not on how it travelled. Skip that step and self transfers inflate turnover, EMIs collected over UPI Autopay drop out of the obligation count, and concentration is measured on payment handles instead of people.

The scale explains why this is now the core of bank statement analysis rather than a detail of it. NPCI recorded 25.47 billion UPI transactions in October 2026, worth ₹31.53 lakh crore, up 23.0% in volume and 15.6% in value on October 2025, or 821.5 million transactions a day.

October 2026Person to merchant (P2M)Person to person (P2P)
Transactions16.09 billion9.37 billion
Share of volume63.2%36.8%
Share of value30.8%69.2%
Average ticketabout ₹603about ₹2,328

Volume is growing faster than value, so the overall average ticket is shrinking: about ₹1,318 a year earlier against ₹1,238 now. For a lender the consequence is structural. A small business account that a credit officer could once read line by line can now carry thousands of UPI rows a month, most of them small. The statement has become a dataset, and whether it yields a credit view depends on how well the narration is parsed.

Why the same UPI payment reads differently in every statement

NPCI did set a common minimum for UPI narration in a 2018 circular. The pattern runs roughly UPI/DR/<12 digit reference>/<shortened name>/<bank>/<UPI ID or account>/<remark>, with CR in place of DR on the credit side. Banks implement it differently, and that is where analysis breaks, in two layers.

Across banks

Field order varies. Some banks keep the payer's remark and some drop it. Delimiters differ. Names are shortened at different points, so the same counterparty survives intact in one bank's statement and is cut off mid string in another's. A parser written for one bank's layout misreads the next bank's.

Across the two sides of one payment

The payer's statement and the payee's statement describe one transfer differently by construction, because each names the other party. When a borrower holds accounts at two banks and moves money between them, the outgoing leg and the incoming leg share almost no text.

That second layer is where the damage happens, because it is how self transfers escape netting. A borrower moving ₹40,000 between their own accounts creates a debit in one and a credit in the other. If the two legs are not matched, the credit is counted as revenue and turnover inflates by the amount moved, every month, at UPI frequency. Consolidating accounts with internal transfers netted was always necessary. UPI made it both more necessary and harder.

Match self transfers on the RRN, then confirm

The reliable key for cross statement matching is not string similarity. It is the Retrieval Reference Number (RRN), the 12 digit reference shown as the UPI Ref No or UTR in payment apps, which the narration template carries on both sides of the payment.

Narration text differs between banks; the reference does not. So the two legs of a self transfer between a borrower's accounts at two different banks, which share no recognisable text, share an RRN.

It is not a perfect key on its own. NPCI revised the way RRNs are built in 2021 and again in 2024 because duplicates occurred, and some banks print the reference in a different position or not at all. The robust rule is to match on RRN plus amount, date and opposite direction. That combination is close to deterministic, where name matching across banks is a guess.

The same resolved identifiers decide circular flow detection. Round tripping that once needed NEFT batches now runs in minutes at no cost to the sender, within the ₹1 lakh general UPI limit, and detecting it depends on recognising the same counterparty on both sides of an account. At UPI frequency that only works on resolved identifiers. The circular transactions glossary entry sets out the patterns.

A VPA is a routing address, not an identity

One person or business routinely holds several Virtual Payment Addresses (VPAs) across different payment apps. A shopkeeper may receive at a handle on one app, another on a second app, and a static QR code linked to a third.

Three handles, one counterparty. Unless they are resolved to a single entity, a concentration analysis will report that the business has three customers when it has one, or three suppliers when it has one.

The reverse error also occurs. A handle need not match the registered name of the business, so a credit from a personal looking handle may well be a trade receipt and a credit from a business sounding handle may be a personal transfer. The handle tells you where the money was routed, not what it was for.

P2M versus P2P does not mean business versus personal

This is the classification error with the largest consequence, and it falls hardest on the borrowers who most need banking based assessment.

NPCI classifies a payment as person to merchant when the payee is set up as a merchant with a merchant category code, and as person to person otherwise. NPCI also recognises an intermediate category, P2PM, for small merchants paid straight into their personal accounts. None of this records what the money was for.

A customer paying a kirana owner at the owner's personal handle is making a commercial payment that is structurally P2P. A proprietor moving money to their own savings account is structurally P2P and economically nothing. A business paying a freelance designer is structurally P2P and commercially an expense.

For sole proprietors, the boundary between personal and business money never existed in the account, and UPI has made it more porous. Separating drawings from business expenditure, and trade receipts from personal transfers, is a classification problem the P2M and P2P flag does not solve and can actively mislead on.

The practical test is counterparty behaviour over time, not the rail classification. A handle that pays ₹400 to ₹900 at irregular intervals, never receives anything back, and sits among dozens of similar handles is a customer. A handle that both sends and receives, in round figures and larger amounts, is something else.

Payment aggregator and marketplace settlements hide the counterparties

A merchant using a payment gateway does not receive customer payments one by one. It receives netted settlement batches from the payment aggregator, usually the next day or the day after, net of the aggregator's charges. Three distortions follow.

The real customers are invisible

Every customer payment arrives as one lump from one entity. Naive concentration then reads as total dependence on a single counterparty, the aggregator. That is true about the cash flow and meaningless about the business.

Banked receipts sit below gross sales

For a gateway merchant the gap is small: merchant UPI payments carry no merchant discount rate as of November 2026, and cards carry a modest fee plus the aggregator's commission. That will change once NPCI's approved 0.4% charge on most merchant UPI payments above ₹2,000, capped at ₹300, takes effect on a date still to be notified. For a marketplace seller the gap is far larger, because marketplace payouts are net of commission, fixed and shipping fees, GST on those fees, returns, GST tax collected at source of 0.5% and income tax deducted at source of 0.1%. Fee schedules change often, so no fixed percentage gap is "normal". When comparing declared GST turnover against banked credits, reconcile against the marketplace's settlement report instead of assuming a range.

Timing shifts

Settlement lag moves receipts out of the day, and occasionally the month, in which the sale happened.

Aggregator and marketplace settlements therefore need to be identified as their own transaction class, not counted as ordinary trade credits from one dominant buyer.

UPI Autopay moves obligations onto UPI rails

UPI Autopay, launched by NPCI in July 2020 with EMIs among its intended uses, lets lenders, insurers and subscription businesses collect recurring payments through a UPI mandate instead of a NACH mandate. Debits up to ₹15,000 run without fresh authentication, and the limit is ₹1 lakh for insurance premiums, mutual fund subscriptions and credit card bills. The customer gets a pre debit notification at least 24 hours before each debit.

In a naive read, those debits look like ordinary UPI payments. Obligation discovery that only scans for ACH and NACH debits will miss every EMI collected over UPI, so the borrower's obligations are understated by whatever runs on UPI rails.

Mandate debits are recognisable by shape: a fixed amount, a stable date, the same payee every month, and a time signature, because since August 2025 NPCI has run mandate executions outside peak hours. The markers in the narration itself vary by bank. None of this is visible unless the narration is parsed into fields rather than skimmed. Obligation mapping from the debit side has to include the UPI rail.

UPI receipts are the best income evidence a micro merchant has had

Everything above is a problem. This part is the opportunity.

For a small merchant, a kirana store, a tea stall, a salon, a small service provider, UPI receipts are the most granular, continuous, third party produced record of revenue that has existed for the segment. Not a declared figure, not a survey estimate: timestamped individual transactions.

What becomes readable:

  • Daily receipt count and its stability, which is footfall
  • Average ticket size and its movement, which is basket size or pricing
  • Day of week and time of day patterns, which show whether activity fits the stated business
  • Seasonality across twelve months
  • Customer repetition: repeat handles against one time payers
  • Growth or decay in receipt count, which tends to move before receipt value when a business starts to struggle

Micro enterprise credit was flat in the year to June 2026 while small and medium exposures grew over 20%, and we have argued that assessment is one of the constraints on micro lending, alongside risk appetite and unit economics. For merchants who accept UPI, income evidence now exists where there was none.

The condition is that thousands of small transactions have to be classified correctly for the pattern to be visible at all. A credit officer can read a few hundred rows; nobody reads several thousand, so the quality of automated classification becomes the assessment.

What competent UPI handling requires

Six things, in rough order of importance:

  1. Parse the narration into fields: RRN, VPA, counterparty name, bank, remark, with per bank layout handling.
  2. Match self transfers across accounts and banks on RRN plus amount, date and direction, and net both legs instead of counting either.
  3. Resolve multiple VPAs to single entities, so concentration and circularity are computed on counterparties, not handles.
  4. Identify aggregator and marketplace settlements as a distinct class, with the gross to net gap reconciled instead of read as a shortfall.
  5. Separate UPI Autopay mandate debits from discretionary payments and fold them into obligation discovery.
  6. Classify by counterparty behaviour over time, not by the P2M and P2P flag or by amount thresholds alone.

A tool that does none of this still produces an output. It produces inflated turnover, understated obligations, invented concentration and invisible circularity, confidently and with no error raised. Our categorisation benchmark shows how far apart two tools can sit on the same statements, and UPI heavy accounts are where those errors pile up, because the volume is high and the failures are silent.

What UPI still cannot tell you

Cash is still invisible

For many micro merchants a meaningful share of sales is still in cash. A rising UPI share can look like revenue growth when part of it is customers switching away from cash.

Some UPI spending never reaches the savings statement

UPI Lite payments, up to ₹1,000 each from an on device balance of up to ₹5,000, do not appear in the bank statement; only the top ups do. Payments made on a credit line or a RuPay credit card linked to UPI are charged to the credit account, so the savings statement shows only the repayments, which are themselves an obligation to find. Payments made by a secondary user under UPI Circle are debited from the primary account holder's account, not the person who made them.

Receipt volume can be manufactured

A VPA says nothing about the payer's identity or legitimacy, and at UPI's cost structure receipt volume is cheaper to manufacture than any other form of turnover. Patterns that do not fit the stated business, such as uniform amounts, uniform intervals and handles that pay once and never recur, belong with the behavioural fraud checks, not in the income line.

Nothing on the rail says what was sold

Goods, services, a loan from a friend and a repayment all look alike at the transaction level.

Frequently asked questions

Why is UPI difficult to analyse in bank statements? Volume and narration inconsistency. A business account can carry thousands of UPI transactions a month, banks implement NPCI's narration pattern differently, and the payer's and payee's statements describe the same transfer differently. Self transfer netting, counterparty resolution and obligation discovery all depend on parsing each narration into structured fields first.

How do you match the two sides of a UPI self transfer across different banks? Use the Retrieval Reference Number, the 12 digit UPI reference carried on both sides of the payment, together with amount, date and opposite direction. Narration text and counterparty names differ between banks, so name matching is approximate. RRN alone is not perfect, because duplicates have occurred and some banks print it differently.

Does P2M or P2P tell you whether a UPI payment was business or personal? No. The classification follows whether the payee is set up as a merchant, not the purpose of the payment. A customer paying a sole proprietor's personal handle is a commercial receipt classified as P2P, and a proprietor moving money to their own savings account is also P2P. Purpose has to come from counterparty behaviour over time.

Why do payment aggregator and marketplace settlements distort credit analysis? They arrive as netted batches, so the real customers are invisible and concentration appears to be total dependence on one aggregator. Marketplace payouts also sit below gross sales by commission, fees, returns, 0.5% GST TCS and 0.1% income tax TDS. Reconcile against the settlement report instead of assuming a fixed gap.

Can UPI data be used to verify a small merchant's income? Yes. Daily receipt counts, average ticket size, day of week patterns, seasonality and customer repetition make it the most granular revenue evidence available for micro merchants. It needs thousands of small transactions classified correctly first, it does not capture cash sales, and UPI Lite payments do not appear in the statement.

How do UPI Autopay EMIs show up in a bank statement? As UPI debits, not NACH debits, so obligation discovery that scans only for ACH and NACH entries misses them. They are recognisable by a fixed amount, a stable date, the same payee every month and execution outside peak hours. Narration markers vary by bank, which is why the narration has to be parsed into fields.

Can UPI receipts be faked to inflate turnover? Yes, and more cheaply than most other forms of turnover, which makes behavioural checks necessary. Uniform amounts, uniform intervals, counterparties that never recur and receipt patterns that do not fit the stated business are the markers worth testing before any UPI receipt is counted as income.


Fiscus parses UPI narration into structured fields across bank layouts, matches self transfers across accounts, resolves handles to counterparties, and separates mandate debits and aggregator settlements from ordinary flows. Book a parallel evaluation on UPI heavy accounts your team has already assessed.

Written by

Zeus Dhanbhoora

Zeus Dhanbhoora is the CEO of BridgeUp Tech, the company behind Fiscus. He previously co-founded Bacferim Technologies and was an associate at the law firm Bharucha & Partners. He writes the Fiscus credit desk blog on benchmarks, fraud detection and credit underwriting methods.

See it on your own cases.

Run a parallel evaluation. Bring statements your team has already analysed and compare output, depth and time.

Book a demo