Home · About · Services · Industries · Projects · Products · Process · Blog · Glossary · FAQ · Contact
India’s Account Aggregator framework: what developers need to know How FIUs, FIPs and AAs interact, what a consent artefact holds, how the encrypted data fetch works, and how long you may keep the data you receive.
A small lender wants to decide on a business loan within a day. Today the applicant uploads six months of bank statements as PDFs. An analyst runs them through a parser, fixes the rows it misreads, and still worries about edited files. The product team has heard that Account Aggregator can replace all of this with a consented data pull straight from the bank.
It can, but the framework is not a single API you call. It is a network of regulated parties with strict roles, signed consents and encrypted payloads. Developers who treat it like a statement-scraping service end up rebuilding their integration. The work is mostly about consent design, key handling and data retention, not about parsing.
The three roles The Account Aggregator system rests on an RBI framework for a new category of NBFC, the NBFC-AA. Three parties matter to a developer.
Financial Information Provider (FIP): the institution that holds the customer’s data, such as a bank, an insurer, a mutual fund registrar or a GST source. Financial Information User (FIU): the regulated entity that wants the data, such as a lender, an insurer or a wealth platform. If you are building a lending app, your client is the FIU. Account Aggregator (AA): the consent manager. It shows the customer what is being requested, records their approval and moves encrypted data from FIP to FIU. The AA is designed to be data blind. It forwards data it cannot read, because the payload is encrypted end to end between FIP and FIU. That single design choice shapes most of your engineering. Your system, not the AA, holds the private keys and does the decryption.
The technical standards come from ReBIT, and Sahamati, the industry alliance, runs onboarding and certification for network participants. Many FIUs connect through a technology service provider rather than building every API themselves. Either way, someone in your stack must understand the flow below.
What a consent artefact contains Everything starts with a consent request that your FIU system sends to the customer’s chosen AA. The customer approves or rejects it inside the AA’s app. An approved request becomes a consent artefact, digitally signed by the AA. That artefact is the legal and technical permission for every data fetch that follows.
The fields deserve real product thought, because the customer sees them and the FIP enforces them.
Purpose: a coded reason, such as account monitoring or a one-off loan assessment. FI types: which kinds of data you want, for example deposits, term deposits or mutual funds. Consent mode: whether you may only view the data, store it, query it or receive it as a stream. Fetch type and frequency: a one-time pull or periodic fetches, with a maximum number per period. FI data range: the period of transactions covered, for example the last twelve months. Consent start and expiry: how long the permission itself stays valid. Data life: how long you may keep the data after you receive it. An abridged consent request looks roughly like this. Field names follow the ReBIT specification, but always build against the current version your AA certifies.
{
"ver": "2.0.0",
"txnid": "4a1c2f7e-0d3b-4c55-9b2e-6f1a2c3d4e5f",
"ConsentDetail": {
"consentStart": "2024-07-18T10:00:00.000Z",
"consentExpiry": "2024-10-18T10:00:00.000Z",
"consentMode": "STORE",
"fetchType": "ONETIME",
"consentTypes": ["TRANSACTIONS", "PROFILE", "SUMMARY"],
"fiTypes": ["DEPOSIT"],
"Customer": { "id": "98XXXXXX10@aa-handle" },
"Purpose": { "code": "105", "text": "Explicit one-time consent for accessing data from the accounts" },
"FIDataRange": { "from": "2024-01-01T00:00:00.000Z", "to": "2024-06-30T23:59:59.000Z" },
"DataLife": { "unit": "MONTH", "value": 3 },
"Frequency": { "unit": "MONTH", "value": 1 }
}
}Ask only for what the decision needs. A lender assessing cash flow rarely needs mutual fund holdings. Narrow consents get approved more often, and they reduce what you must protect later.
The fetch flow, step by step The sequence has more steps than a typical REST integration. Model it as a state machine from the start.
Your FIU sends the consent request to the AA and receives a consent handle. The customer approves in the AA app. The AA sends a consent notification to your callback endpoint. You fetch the signed consent artefact using the consent ID and verify the AA’s signature. You create a data request against that consent. It includes your public key material for this session and a nonce. The AA returns a session ID and passes the request to the FIP. The FIP prepares the data, encrypts it with a key derived from both sides’ key material, and notifies when it is ready. You fetch the data with the session ID and decrypt it using your private key and the FIP’s public key material. The key exchange uses elliptic-curve Diffie-Hellman, with fresh key pairs per session. Do not reuse one key pair for every request. Generate it, keep the private half in memory or a protected store for the life of the session, and discard it afterwards.
Every hop is asynchronous. The customer may take minutes to approve, or never. The FIP may be slow or temporarily down. Your system must persist each state and resume cleanly, rather than holding an HTTP request open. A typical status table looks like this.
CREATE TABLE aa_consent (
id uuid PRIMARY KEY,
applicant_id uuid NOT NULL,
aa_handle text NOT NULL,
consent_handle text,
consent_id text,
status text NOT NULL CHECK (status IN
('REQUESTED','ACTIVE','REJECTED','PAUSED','REVOKED','EXPIRED')),
consent_expiry timestamptz,
data_life_until timestamptz,
created_at timestamptz NOT NULL DEFAULT now()
);
CREATE TABLE aa_fetch_session (
id uuid PRIMARY KEY,
consent_id uuid NOT NULL REFERENCES aa_consent(id),
session_id text NOT NULL,
status text NOT NULL, -- PENDING, READY, FETCHED, FAILED
fips_ready int NOT NULL DEFAULT 0,
fips_total int NOT NULL,
requested_at timestamptz NOT NULL DEFAULT now(),
fetched_at timestamptz
);One consent can cover accounts at several FIPs. Data may arrive from one bank while another is still pending. Decide in advance whether your credit logic can run on partial data, and show the applicant which accounts came through.
Handling consent status changes A consent does not stay active forever. The customer can pause or revoke it from their AA app at any time. It also expires on its own date. The AA notifies your callback when status changes, and you must act on it.
Revocation means no further fetches. Your scheduler for periodic pulls should check status before every run, not trust a cached flag from last week. For monitoring use cases, such as watching a borrower’s account after disbursal, build alerts for revocations, because the credit team usually wants to know.
Callbacks need the usual hardening. Verify the signature on each notification. Make the handler idempotent, because the same notification may arrive twice. Log the raw payload with a request ID, since disputes about who consented to what will come back to these records.
Storing and expiring the data The data life field is not a suggestion. If the consent says three months, the decrypted financial data must be gone after three months. India’s data protection law, the Digital Personal Data Protection Act 2023, adds its own expectations around purpose and retention, so the same discipline serves both.
A sensible design separates three kinds of record.
Raw FI data: the decrypted account statements. Store encrypted at rest, in a dedicated table or bucket, with the expiry date stamped on every row or object. Derived features: the numbers your model or analyst produced, such as average monthly balance or bounce count. Check with your compliance team whether these fall under the same data life. Audit records: the consent artefact, status history and fetch logs. These prove the data was obtained lawfully and are usually kept longer. A nightly job then enforces expiry. Keep it boring and observable.
-- run nightly; alert if rows remain past expiry
DELETE FROM aa_fi_data
WHERE data_life_until < now();
SELECT count(*) AS overdue
FROM aa_fi_data
WHERE data_life_until < now() - interval '1 day';Backups are the trap. If your database backups run for ninety days and data life is thirty, expired data lives on in snapshots. Either shorten retention for that store or keep the raw data in a separate system with its own backup policy.
Testing and going live Sahamati and the AAs provide sandbox environments with test FIPs. Use them to rehearse failure paths, not just the happy path. Reject a consent, let one expire, revoke mid-session, and make an FIP time out. Your state machine should handle each without manual database edits.
Watch the user experience on mobile. The customer leaves your app for the AA’s flow and must come back. Deep links and redirect URLs need testing on real devices, with slow networks, because that is where applications get abandoned.
Finally, parse defensively. FI data follows published schemas, but banks differ in narration text, transaction categories and how they report balances. Keep the raw payload, write your parser for the schema, and log anything unexpected rather than dropping it silently.
A checklist before you build Confirm your client is, or will partner with, a regulated FIU, and which AAs you will support. Write down the purpose, FI types, data range, frequency and data life for each product, and justify each field. Model consent and fetch sessions as persisted state machines with idempotent callbacks. Generate fresh key pairs per session and keep private keys out of logs. Stamp an expiry on every stored data row and enforce it nightly, including backups. Rehearse rejection, revocation, expiry and FIP failure in the sandbox before go-live. Get these right and the parsing work becomes the easy part.
Infigini IT Services · GSTIN 05AAIFI1118B1Z3 · 11 Shree Ram Puram Lane No.2, Kanwali Road, Dehradun, Uttarakhand 248001, India · info@infigini.com · +91 95684 87789 · LinkedIn