Part II — Payments

Chapter 9 · UPI Architecture

Twenty-four billion times a month, money moves between two bank accounts in seconds, and almost nobody asks how. This chapter asks. The cast — PSP banks, third-party apps, the NPCI switch. The address — the virtual payment address. The journey — one payment traced hop by hop, on-us against off-us, ending in net settlement under NPCI's guarantee. And the instruction that made subscriptions painless: the mandate.

📖 Chapter 9 of 35 ⏱ ~60 min read 🎓 Session 9 · CILO 1 Core 75% Stable 25% Volatile 0% 📝 Quiz · 30 MCQs

🎯 Learning Objectives

1. What UPI Actually Is

Say it precisely, because imprecision costs marks and money: UPI — the Unified Payments Interface — is an open API specification plus a central switch, operated by NPCI on the rails of IMPS, the instant banking transfer system that predates it. UPI is not an app, not a wallet, and not a company. It is a set of messages any bank can implement, a switch those banks connect to, and a rulebook NPCI enforces. By September 2026, 756 banks were live on it.

The cast has four roles. The PSP bank (payment service provider) is a bank connected to the UPI switch; it offers UPI service through its own app or through third parties. The TPAP — the third-party application provider you met in Chapter 6 — builds the customer-facing app and rides one or more sponsor PSP banks; it never touches funds or settlement. The NPCI switch routes every message between PSP banks and resolves addresses. And the accounts — always bank accounts, at both ends; UPI moves money account-to-account, which is why there is no "wallet balance" step in a standard UPI payment.

Notice what the design achieves. Any app can pay any account at any bank, because interoperability lives in the switch and the spec, not in bilateral deals. A two-person startup's app and a trillion-dollar bank's app send identical messages to the same rail. That is Chapter 5's public-rail thesis, expressed as software.

2. The Virtual Payment Address

UPI's signature invention is an alias. A virtual payment address (VPA) — a UPI ID — looks like name@bank: asha@oksbi, kirana@ybl, fees@icici. Behind it sits a mapping to a real account; the switch resolves the alias when a payment is addressed to it. The payer never sees the account number or the IFSC code.

Three properties made the VPA behaviour-changing. Safe to share: an alias can only receive money — handing it over reveals nothing that could debit you, which killed the fear that stopped account-number sharing. Easy to say: a shopkeeper reads her VPA across the counter or prints it in a QR code. Layered on demand: the same account can wear many VPAs across many apps, so identities stay separate — salary, side business, family collection.

The VPA is not the only addressing mode. UPI also accepts account number + IFSC directly, mobile numbers linked to accounts, Aadhaar-linked addressing, and QR codes — a printed static QR simply encodes the payee's VPA, while a dynamic QR is generated per invoice and carries the amount, so the customer approves rather than types. The alias family is the point: address the person or the shop, not the ledger entry.

💡
Pro Tip: The part after the @ usually names the PSP or app family, not the destination bank. A payment to asha@ybl can still land in Asha's account at any bank — the handle names the door, not the house.

3. A Transaction, Traced

This is the session's heart. Watch a ₹150 payment move — a customer (Ashok, banked at Bank A, using a TPAP sponsored by Bank A) scans a tea stall's QR, whose owner banks at Bank B.

  1. Address. The QR resolves to the stall's VPA — chaistall@bankb. Ashok's app shows ₹150 and asks for his UPI-PIN.
  2. App to PSP. The app sends the signed payment instruction over secure APIs to its sponsor PSP — Bank A.
  3. PSP checks and debits. Bank A verifies Ashok's identity binding and balance, debits ₹150 from his account, and forwards the message to the switch.
  4. Switch resolves and routes. The NPCI switch resolves chaistall@bankb to the stall's account at Bank B and routes the credit message to Bank B's PSP systems.
  5. Beneficiary bank credits. Bank B validates the beneficiary, credits the stall's account, and returns confirmation.
  6. Response races back. Switch → Bank A → app: "Paid ₹150". The whole round trip completes in seconds, all day, every day.
  7. Settlement follows. The money has moved — the customer experience is complete — and between the banks, the interbank position settles on a net basis, as Section 4 explains.

Two details reward attention. The payer authenticates with two factors: the transaction is initiated from a device bound to the account (first factor) and confirmed with the UPI-PIN (second factor) — set up using debit-card details or an Aadhaar OTP. And the flow is bidirectional: UPI also supports the collect request, where the payee pushes a request to the payer's app, and the payment happens only when the payer approves it with a PIN. Cards are push-only; UPI asks as well as sends — the quiet feature that made bill payment and doorstep commerce work.

📝
Trace it yourself: The next time you pay by UPI, watch the screen's language — "Verifying", "Processing", "Success" are the hops of steps 3–6 made visible. When a payment "fails but money left my account", you now know exactly which hop broke: the credit leg returned an error, and the debit will be reversed under the system's auto-reversal rules.

4. On-Us, Off-Us and Settlement

Every UPI transaction is one of two kinds, and the kind decides the plumbing. On-us: payer and payee hold accounts at the same bank — both legs settle inside that bank's own ledger, and the transaction never needs an interbank leg. Off-us: the accounts sit at different banks — the switch routes between them, and the banks' net positions must settle between institutions.

Off-us settlement is where the grown-up machinery lives. Thousands of payments criss-cross every second; rather than moving money payment by payment, the system nets each bank's position against every other and settles the net across their settlement accounts. And because a payer's bank must credit the experience in real time even before interbank money has moved, the system runs on settlement guarantee — NPCI's framework (with settlement accounts and risk controls) ensures a paying bank's obligations are covered even if a counterparty fails mid-flight. You met the principle in Chapter 5 — shared infrastructure needs a guardian — and here is where the guardian holds the ring.

On-us has economics worth noticing. Both legs inside one bank means no interbank settlement leg at all — cheaper and operationally simpler for the bank, and one reason large ecosystems love keeping both sides of a payment. Regulators watch the same property with the opposite interest: too much on-us inside one closed ecosystem begins to resemble a private rail wearing UPI's clothes. The tension returns in Chapter 11's competition story.

📋 Stable content — Reviewed: October 2026 · Next review: April 2027

5. Mandates and Autopay

Every payment so far needed a human in the loop — a PIN per transaction. Subscriptions, EMIs and insurance premiums need the opposite: standing permission. UPI's answer, from the 2.0 generation of the specification, is the mandate — a pre-approved instruction the customer signs once, authorising a payee to collect specified amounts on a specified rhythm.

UPI Autopay, launched by NPCI in 2020, productised the mandate on the RBI's e-mandate framework. The customer authenticates the mandate with a UPI-PIN at set-up; thereafter, recurring debits proceed automatically up to the mandate's limits. The framework's best-known line: recurring debits of up to ₹15,000 per transaction need no additional authentication — above that, or at intervals the mandate does not cover, the payer receives an authentication request again. Per-transaction caps on UPI itself sit at ₹1 lakh for general payments, with higher limits (₹5 lakh) for categories such as hospitals, education and tax payments.

Two protections carry the conduct story. Revocation is the customer's: a mandate can be cancelled at any time through the bank or the app — an instruction signed once is not a right granted forever. And pre-debit notice rules require the payee to tell the payer before a collection lands, so a "silent" debit is a compliance failure, not a design feature. Chapter 20 returns to both when it audits dark patterns in recurring collections.

6. Design Choices and Their Consequences

Close the chapter the way an architect would — by reading the consequences out of the choices.

Choice: accounts, not wallets. Consequence: no stored value at the app layer, no float business for TPAPs (Chapter 8), and direct-deposit merchants at near-zero cost — but also irrevocability by default: an authorised account-to-account payment does not charge back like a card. Mistaken and fraudulent payments are hard to unwind, which is exactly the fraud-and-mule problem of Chapter 14. Choice: open spec, central switch. Consequence: radical interoperability and app-layer competition — and radical concentration risk if the apps themselves fail to diversify. The system learned this the hard way in March 2020, when a moratorium on one sponsor bank took its single-PSP-dependent app dark for a day; the response — TPAPs must distribute across multiple PSP banks — is now rulebook.

Choice: PIN on every payment, device-bound. Consequence: strong customer authentication at population scale — and a fraud industry that pivoted to social engineering, tricking the human instead of breaking the cryptography. Choice: inclusion add-ons — UPI 123​Pay for feature phones, Aadhaar-linked addressing, rising caps. Consequence: the rail keeps absorbing users other systems price out. And the next choice is live now: credit lines and pre-paid instruments riding UPI flows — Chapter 19's subject — extend the architecture from moving deposits to moving credit.

⚠
Warning: Do not leave this chapter able to name the parts but unable to trace the flow. In assessments — and in every payments job interview — the failing answer describes the cast; the passing answer walks the money through the hops, names where on-us changes the path, and states who bears what until settlement completes.

You can now see inside a payment. Part II's next question is the analyst's: what does the system's data say? Chapter 10 teaches you to read payment-system statistics the way NPCI publishes them — volume, value, ticket size and frequency — and, in the spirit of Chapter 3, to state precisely what each series cannot prove.

🛠 Hands-On: Transaction Flow Tracing

This is the session's classroom activity, made self-paced. Four payments follow. For each, list every hop in order (app, PSP, switch, counter-PSP, accounts), classify on-us or off-us, name who debits and credits and when the interbank position settles, and state where the two authentication factors appear. Answers in writing first.

Steps

  1. The QR chai payment. Ashok (Bank A account, Bank A-sponsored TPAP) pays ₹150 to a stall QR resolving to chaistall@bankb. Trace it.
  2. The same-bank split. Two colleagues, both at Bank C, one paying the other via a Bank C app after a collect request. Trace it and classify.
  3. The collect request. A tutor raises a request for ₹800 to a parent's VPA; the parent approves. Who initiates, who authenticates, and in what order?
  4. The subscription. A ₹499 monthly streaming mandate, set up via UPI Autopay, executes on the 5th of each month. What happened at set-up, what happens monthly, and what if the fee rose to ₹999?
  5. Diagnose the failure. A customer's screen shows "money debited, merchant says not received". Using the hop model, name the leg that failed and what should happen next.
View Solution / Walkthrough

1. App → Bank A (PSP: verifies binding and balance, debits ₹150, factor 2 = UPI-PIN captured in the app; factor 1 = device binding) → NPCI switch (resolves chaistall@bankb, routes credit) → Bank B credits the stall → confirmation returns. Off-us: Bank A and Bank B's net position settles interbank under NPCI's settlement guarantee; the customer experience completed at credit.

2. App → Bank C PSP → (the switch may still be in the message path, but both legs post on Bank C's own ledger) → payee credited. On-us: no interbank settlement leg at all — cheaper for the bank, and the reason ecosystems favour keeping both sides. The collect request simply reverses who initiates: the payee's side raises it, and the payer's UPI-PIN still authorises the debit.

3. Initiation: the tutor's PSP sends a collect request to the switch, which routes it to the parent's PSP and app. Authentication: the parent alone — the request is only an invitation until their PIN converts it into a payment. The debit and credit then follow the standard hops.

4. Set-up: one authenticated mandate creation — payee, amount range, frequency — signed with the customer's UPI-PIN. Monthly: a pre-debit notice, then an automatic ₹499 debit within the mandate — under ₹15,000, so no additional authentication. At ₹999 the same applies, still within cap; if the amount exceeded the mandate's or the framework's limits, the payer would re-authenticate. And throughout, the customer may revoke the mandate at the bank or in the app.

5. The debit leg (payer's bank) completed; the credit leg (beneficiary's side) failed or timed out — hop 5 of Section 3. Auto-reversal rules return the debit, typically within the system's timelines; the merchant's "not received" is consistent until the trace resolves. This exact window — debited but uncredited — is where Chapter 14's dispute routes begin.

Key Takeaways

1

UPI is an open API specification plus NPCI's switch on IMPS rails — not an app, not a wallet. Interoperability lives in the spec.

2

PSP banks connect to the switch; TPAPs build apps on their rails and never touch funds. 756 banks live on UPI (September 2026).

3

The VPA — name@bank — made addresses safe to share; the switch resolves the alias to the account. QR codes simply carry VPAs.

4

Trace the hops; classify the payment. On-us settles inside one bank; off-us crosses the switch and nets under NPCI's settlement guarantee.

5

Mandates (Autopay, 2020, on the e-mandate framework) pre-authorise recurring debits — ₹15,000 per transaction without re-authentication, revocable by the customer anytime.

Test Your Understanding

Choose one best answer per question. Select Check Answers to score yourself. An explanation appears after each question once checked. Aim for 24/30 before moving to Chapter 10.

1. UPI is most precisely described as:

2. Within UPI, a PSP bank's defining role is:

3. A third-party application provider:

4. A virtual payment address looks like:

5. The property that made VPAs behaviour-changing is that they are:

6. The QR code printed on a tea stall's counter encodes, at minimum:

7. The correct order of hops in an off-us UPI payment is:

8. When a payment is addressed to chaistall@bankb, the alias is resolved by:

9. A transaction is "on-us" when:

10. An off-us UPI payment requires:

11. Between PSP banks, UPI interbank obligations settle by:

12. A PhonePe user whose sponsor PSP is Bank X pays a merchant banking with Bank Y. The payment is:

13. In a UPI collect request:

14. The two authentication factors in a UPI payment are:

15. A customer sets her UPI-PIN using:

16. The difference a dynamic QR makes over a static one is that it:

17. The rule that TPAPs must distribute across multiple PSP banks followed:

18. UPI Autopay was launched in 2020, built on:

19. Under the e-mandate framework, recurring debits proceed without additional authentication up to:

20. UPI's general per-transaction limit, with higher caps for categories like hospitals and tax payments, is:

21. A customer who no longer wants a merchant's autopay to run may:

22. UPI 123​Pay exists to serve:

23. The number of banks live on UPI as of September 2026 was about:

24. The legacy rail whose infrastructure UPI builds on is:

25. A UPI payment is instant around the clock because:

26. A single bank account can:

27. Which of these is a TPAP expressly not permitted to do?

28. The design consequence of account-to-account (no-wallet) architecture is:

29. Why do large payment ecosystems like on-us transactions?

30. The next chapter in Part II teaches you to: