Fintech & Banking
PSD2 and Open Banking: Roles, APIs, SCA and eIDAS Certificates
Intermediate8 min readUpdated
PSD2 turned the bank account into a platform: licensed third parties can read balances and initiate payments on a customer's behalf, over interfaces banks are obliged to provide. This guide explains the regulated roles, the security machinery, the competing API standards and what actually happens on the wire.
What you will learn
- Distinguish the regulated roles — ASPSP, AISP, PISP and CBPII — and what each may do
- Explain strong customer authentication, dynamic linking and the main exemptions
- Describe the dedicated interface obligations and how eIDAS certificates identify a third party
- Compare the Berlin Group, UK Open Banking and STET approaches to the same regulation
- Design a consent lifecycle and an account information flow that survives real bank behaviour
Before 2018, if you wanted to build an application that showed a customer all their bank accounts in one place, you asked for their online banking password and logged in as them. It worked, it was fragile, and it was indefensible from a security point of view. The revised Payment Services Directive — PSD2 — replaced that arrangement with something better: licensed third parties, explicit customer consent, and interfaces that banks are legally obliged to provide.
The result is a regulated API ecosystem with an unusual property for engineers: the specification of what must be offered is legal text, while how it is offered was left to the market, producing several competing technical standards on top of one obligation. This guide covers both halves — the regulatory model you must satisfy and the API mechanics you will actually code against.
What changed and why
PSD2 pursued three goals at once: more competition in payments, better security, and a legal basis for the account access that was already happening informally. Its most consequential provision is access to accounts: a bank holding a payment account that is reachable online must let a licensed third party, acting with the customer's explicit consent, read that account or initiate a payment from it — without charging for the privilege and without requiring a contract between the bank and the third party.
Two more provisions shape everyday engineering. Strong customer authentication became mandatory for most electronic payments and for online account access. And liability rules moved decisively toward the customer: unauthorised transactions are refunded promptly unless the provider can show fraud or gross negligence, which is why banks invest so heavily in the authentication step.
"Open banking" is the market term for the ecosystem the directive created. Similar frameworks exist elsewhere in the world with different legal bases and comparable technical shapes.
The regulated roles
Four acronyms carry the whole model.
- ASPSP — Account Servicing Payment Service Provider. The institution holding the account, obliged to expose the interface. In practice, the bank.
- AISP — Account Information Service Provider. Reads account data: balances, transactions, account details. Read-only, aggregation, budgeting, credit assessment.
- PISP — Payment Initiation Service Provider. Instructs a payment from the customer's account with their consent. It never touches the funds; the bank executes.
- CBPII — Card Based Payment Instrument Issuer. Asks the bank a single yes/no question: are these funds available right now? Used by card issuers whose cards draw on an account at another institution.
A firm can hold several roles. The customer, in the legislation's language the payment service user, is the only party who can grant consent, and can revoke it at any time.
The PISP role deserves attention because it changes the economics of e-commerce: a merchant can be paid by an account-to-account credit transfer initiated at checkout, with no card scheme, no interchange and — with an instant scheme underneath — irrevocable funds within seconds. That is a genuinely different risk profile from the card flow described in card payments, because a credit transfer has no chargeback mechanism.
Strong customer authentication
SCA requires authentication with at least two independent elements drawn from three categories:
- Knowledge — something only the user knows (a passcode, not a memorable word sent by email).
- Possession — something only the user has (a registered phone with a bound key, a hardware token, a card with a chip).
- Inherence — something the user is (fingerprint, face, or behavioural biometrics).
Independence means the compromise of one element must not compromise the others; a code shown on the same browser that holds the session does not qualify. For remote payments there is an extra requirement, dynamic linking: the authentication code must be cryptographically bound to the specific amount and payee, and any change to either invalidates it. That single rule kills a whole class of attacks in which an approved authentication is replayed against a different beneficiary.
The regulatory technical standards define exemptions the bank may apply, at its own risk:
| Exemption | Typical use | Note |
|---|---|---|
| Low-value transaction | Small remote purchases | Cumulative counters force periodic authentication |
| Contactless at point of sale | Tap payments below a limit | Also subject to cumulative limits |
| Recurring transaction | Fixed amount, same payee | SCA on the first, not on subsequent ones |
| Trusted beneficiary | Payee whitelisted by the customer | Whitelist maintained by the bank, with SCA |
| Corporate payment process | Secure dedicated business protocols | Not for consumer flows |
| Transaction risk analysis | Real-time scoring below fraud thresholds | Allowed only while the provider's fraud rate stays low |
Two points engineers routinely get wrong. Exemptions are requested by the payment provider but granted by the issuer, so your code must handle a challenge arriving anyway. And account information access has its own rhythm: SCA on first access, then a period during which the third party may refresh data unattended a limited number of times per day, then re-authentication.
The interface, and proving who you are
Banks may expose a dedicated interface — a purpose-built API — and the vast majority do. The rules impose obligations that read like an SLA written by a regulator: the dedicated interface must offer availability and performance at least equivalent to the bank's own customer channels, publish uptime and error statistics, provide a testing facility with a sandbox, and support wide-scale testing before launch. Where a bank's dedicated interface fails to meet those standards, a fallback mechanism must exist so third parties can still reach the customer channel; banks that demonstrate a compliant interface may be exempted from building one. This is one of the areas the successor legislation intends to simplify.
Identification uses eIDAS certificates, issued by qualified trust service providers and carrying the third party's authorisation number and roles as an extension:
- QWAC — Qualified Website Authentication Certificate, presented in a mutually authenticated TLS handshake. It identifies the caller at the transport layer.
- QSEALC — Qualified Certificate for Electronic Seal, used to sign the HTTP request itself, so the message stays attributable independently of the transport.
Both encode the role attributes (PSP_AI, PSP_PI, PSP_IC), which lets a bank check authorisation without a bilateral contract. Certificates prove identity; they are not authorisation for a specific account. The customer's consent, captured in a redirect to the bank's own authentication, is what grants access — a separation that maps naturally onto the OAuth 2.0 authorisation-code flow that most implementations use, often with OpenID Connect on top.
The standards landscape
One directive, several API standards, because the legislation deliberately specified outcomes rather than schemas.
Berlin Group NextGenPSD2 is the most widely implemented in continental Europe: REST and JSON, an explicit consent resource, and several SCA approaches (redirect, decoupled, embedded). Its flexibility is also its weakness — banks pick different options, so "Berlin Group compliant" does not mean drop-in compatible.
UK Open Banking came from a competition remedy rather than only from PSD2, which allowed a stricter, centrally governed specification with a directory, conformance testing and a mandated OpenID Connect profile. Consequently it is the most uniform of the three.
STET, originating in France, is a compact REST specification closely aligned with ISO 20022 vocabulary for the payment initiation payloads.
Practical consequence: even within one standard, expect per-bank quirks in pagination, transaction history depth, booking-date semantics, currency handling and error codes. Teams either integrate through an aggregator or build a per-bank adapter layer behind a normalised internal model — a textbook use of the anti-corruption layer idea, and of the routing and quota concerns in API management and gateways.
A typical account information flow
The shape is consistent across standards even where the details differ. First, create a consent describing exactly what is being requested:
POST /v1/consents HTTP/1.1
Host: api.examplebank.test
X-Request-ID: 1a2b3c4d-5e6f-4a7b-8c9d-0e1f2a3b4c5d
PSU-IP-Address: 203.0.113.14
TPP-Signature-Certificate: <base64 QSEALC>
Signature: keyId="SN=...,CA=...",algorithm="rsa-sha256",headers="digest x-request-id",signature="..."
Content-Type: application/json
{
"access": { "balances": [ { "iban": "GB11ABCD12345678901234" } ],
"transactions": [ { "iban": "GB11ABCD12345678901234" } ] },
"recurringIndicator": true,
"validUntil": "2026-12-07",
"frequencyPerDay": 4,
"combinedServiceIndicator": false
}
The bank replies with a consent identifier and a link to its authentication page. The customer is redirected there, authenticates with SCA against their own bank, and is returned to your redirect URI. The consent then moves from received to valid, and data calls become possible:
GET /v1/accounts/3dc3d5b3-7023-4848-9853-f5400a64e80f/transactions?bookingStatus=booked&dateFrom=2026-08-01 HTTP/1.1
Consent-ID: 4a1e7a2c-88f1-4f2f-9a1e-8b7c6d5e4f30
Authorization: Bearer eyJhbGciOiJQUzI1NiIsInR5cCI6IkpXVCJ9...
A payment initiation follows the same pattern: create the payment resource, redirect for authentication, then poll the status — RCVD, ACTC, ACSC and friends, taken straight from the ISO 20022 vocabulary described in payment systems fundamentals.
Design notes that save weeks:
- The consent is a first-class entity in your domain. Persist its id, scope, expiry, status and remaining daily frequency; drive a re-consent journey from it.
- Never store customer banking credentials. Beyond being prohibited, receiving them at all makes your service a phishing target.
- Treat every bank as flaky. Timeouts, maintenance windows and partial data are normal; use circuit breakers, cache with clear staleness signalling, and degrade to "last updated at" rather than an error page.
- Log for audit, not for content. Regulators and disputes need proof of consent and of what was requested; account contents in application logs are a liability.
- Minimise data. Requesting ninety days of every account when you need a current balance is both a compliance risk and a slower product.
Where the rules are heading
The European Commission has proposed a successor package: a revised directive (commonly PSD3) alongside a directly applicable payment services regulation, plus a separate framework for wider financial data access. The direction of travel is toward stricter interface performance obligations with the fallback requirement removed for banks that meet them, clearer rules on authentication and on liability for impersonation fraud, more consistent supervision, and a move of technical detail from national transpositions into directly applicable text. The texts are still moving; build for the current rules and keep the consent and authentication parts of your architecture replaceable.
Key takeaways
- PSD2 obliges banks to open payment accounts to licensed third parties acting on explicit customer consent, replacing credential-sharing screen scraping.
- Four roles matter: the account-servicing bank, the account information provider, the payment initiation provider, and the funds-check role for card issuers.
- Strong customer authentication needs two independent factors plus dynamic linking of amount and payee; exemptions are requested by the provider but granted by the bank.
- eIDAS certificates identify the third party at transport and message level, while OAuth-style redirects carry the customer's consent.
- Berlin Group, UK Open Banking and STET solve the same obligation differently, and per-bank variation means an adapter layer or an aggregator.
- Model consent explicitly, expect expiries and flaky interfaces, minimise the data you request, and keep the authentication layer replaceable as PSD3 and the payment services regulation land.
Sources and further reading
Frequently asked questions
- Do I need a licence to build an open banking application?
- To call the regulated interfaces yourself, yes: you must be authorised or registered as a payment institution in the relevant role by a national competent authority, and hold the certificates that prove it. Many products avoid this by working through a licensed aggregator that holds the permissions and exposes a single commercial API, accepting the extra dependency in exchange.
- Is open banking the same thing as screen scraping?
- No, and PSD2 was written partly to end scraping. Before the directive, aggregators logged into online banking with the customer's own credentials and parsed the HTML, which meant sharing full credentials and breaking on every redesign. PSD2 requires banks to publish a dedicated interface and forbids third parties from asking for those credentials once it works.
- Why do consents expire and need re-authentication?
- An account information consent is time-boxed by the rules, and the customer must re-authenticate periodically for continued access, with an allowance for a limited number of unattended refreshes between authentications. Build for it: store the consent status, track its expiry, and design a re-consent journey rather than discovering the expiry when a nightly job starts failing.
- What changes with PSD3 and the payment services regulation?
- The proposals under negotiation aim to tighten interface quality and availability, remove the fallback-interface obligation for banks meeting performance requirements, refine the authentication rules, and put much of the detail in a directly applicable regulation rather than a directive transposed differently in each country. Treat specifics as provisional until the final texts are published.