Consider a personal loan application at a mid-sized NBFC. The applicant fills in a form on the lender's website with her name, mobile number, email ID and PAN. The website's content management system saves the submission, and the CRM turns it into a lead record. Marketing tools copy her mobile number into SMS and WhatsApp campaigns.

Web server, CDN and firewall logs record her IP address and device, which CERT-In requires to be kept in India for 180 days. Only then does the loan origination system store her details, behind access controls and an audit trail. That system is the record most DPDP programmes plan around. By the time it exists, her data already sits in four other places.

Within a week, the rest of her file has travelled too. She uploads a masked Aadhaar copy, a voter ID and three months of bank statements, which land as images and PDFs. A video KYC call records her face and the PAN card she holds up to the camera. Credit operations email a spreadsheet of her PAN, mobile and loan amount to a sales agent, whose team forwards it twice.

Diagram showing one applicant's data moving from a website form into the CMS, CRM, marketing tools and logs, then into the loan system, KYC images, video KYC, a spreadsheet, emails, a sanction letter PDF, a shared drive, a laptop and a vendor file
One applicant, many copies: an illustrative loan journey

The sanction letter, a PDF with her address and PAN, reaches the relationship manager's inbox and a shared folder called "Sanctions_Oct". An analyst downloads the spreadsheet to a laptop for a weekly report. None of these copies appears in the system of record. Each one is still personal data the lender answers for.

The main obligations of the Digital Personal Data Protection Act, 2023 take effect on 13 May 2027. From that day, notices, consent withdrawals, erasure requests and breach reports must account for every copy. Data discovery and classification tells you what you hold and where it sits. Data mapping tells you where it travels.

Every obligation assumes you know where the data is

The date comes from the commencement notification of 13 November 2025. It brings most of the Act and the DPDP Rules, 2025 into force eighteen months later. In January 2026, Storyboard18 reported that MeitY had sought views on shortening some windows, with no decision then taken. Plan to 13 May 2027, and see our piece on working back from 13 May 2027 for the sequence.

Read the obligations as questions and each one turns out to need an inventory of copies.

Notice. Rule 3(b)(i) requires the notice to give an itemised description of the personal data. That includes the fields the website, the CRM and the KYC app capture, as well as the loan form.

Consent withdrawal. Section 6(6) requires the Data Fiduciary to stop processing within a reasonable time after withdrawal. It must also cause its Data Processors to stop. That reaches every system, file and vendor holding the data for that purpose.

Security safeguards. Rule 6(1) sets minimum measures, including encryption, masking, access control, logging and backups. Section 8(5) covers personal data "in its possession or under its control", including processing done on its behalf by a Data Processor. A control can only protect a file the organisation knows exists.

Breach intimation. Rule 7(1) requires intimation to each affected Data Principal without delay. You can't name the affected people from a folder you have never classified.

Access. Under Section 11(1)(b), the Data Principal can ask who else received her data. The answer must name every other Data Fiduciary and Data Processor it went to, with a description of what was shared. An emailed export counts.

Erasure. Section 8(7) requires erasure on withdrawal or once the purpose is no longer served, unless law requires retention. The Data Fiduciary must also cause its processor to erase.

The stakes are set out in the Schedule. Failing to take reasonable security safeguards under Section 8(5) can attract up to ₹250 crore. Failing to intimate a breach under Section 8(6) can attract up to ₹200 crore. Section 33(2)(b) asks the Board to weigh the type and nature of the personal data affected.

Where personal data sits, and why the files matter most

Ask an IT team where customer data lives and you'll get a list of applications. That list is accurate and incomplete. Personal data sits in four broad places.

  • The front door. Website forms and the content management system behind them, tag managers, analytics, the CRM, marketing automation, SMS and WhatsApp platforms, and call centre recordings.
  • Structured systems. Loan origination, the loan management and collections systems, core banking or policy administration, ERP, HRMS and the data warehouse. These have schemas and owners, so they are the easiest to inventory.
  • Unstructured files and messages. Email bodies and attachments, PDF statements and letters, scanned KYC documents, live photos, video KYC recordings, spreadsheets, chat exports and call notes.
  • The edges. Laptops and desktops, SaaS tools bought on a corporate card, backups, server and firewall logs, test environments and vendor systems.

Most programmes inventory the second group well and the others barely. That's the wrong way round, because unstructured copies are where DPDP obligations fail. Four features make them hard to manage.

  • They multiply. One email to a vendor creates copies in your sent items, the vendor's inbox and every forward. Mail archives and backups keep their own copies after the original is deleted.
  • They carry no context. A database row has a customer ID, a purpose and a consent flag. An Excel export or a scanned voter ID has none of these. Nothing in the file says whose data it holds, why it exists or when it should go.
  • They shed the controls. A column may be encrypted and masked in production while its export is plaintext. Rule 6(1)(a) expects encryption, obfuscation, masking or tokens. The export usually escapes all of them.
  • They hide in pictures, audio and free text. A scanned KYC form is an image. A video KYC call is a recording. An Aadhaar number typed into a call note is free text. None of them shows up in a schema.

The KYC process creates some of these files by design. The RBI's KYC Directions for NBFCs require a live photograph of the customer during digital KYC. The photo is watermarked with GPS coordinates and a time stamp. Video KYC data and recordings must be stored in India.

Logs hold one more surprise. CERT-In's 2022 directions require logs of all ICT systems to be kept for 180 days. Rule 6(1)(e) separately asks for a year of logs to investigate unauthorised access. Logs are personal data stores the law expects you to keep, so they belong on the map.

Six checkpoints and the obligation at each

Personal data passes six checkpoints in its life inside an organisation. Each one carries its own obligation, and each needs the map to answer a different question.

CheckpointKey provisionsWhat the map must answer
CollectionSections 5, 6, 9 and Rules 3, 10Which fields each form, app and KYC step captures
StorageSection 8(5), Rule 6Every store holding the data, including files, images and logs
UseSections 4, 7 and 8(3)The purpose and ground for each use
SharingSections 8(2), 11(1)(b), 16 and Rule 15Every recipient, the channel used, the contract and the country
RetentionSections 8(7), 8(8) and Rule 8The period for each purpose and the law that sets it
ErasureSections 6(6), 8(7) and 12(3)Every copy to delete, including processor and laptop copies

Unstructured copies break the checkpoints in a particular way. They skip them. A spreadsheet export moves data from storage to sharing with no approval, no contract check and no retention date. Section 8(3) requires accuracy where data drives a decision about the person. A stale export of bank statement figures feeding a credit decision fails that test quietly. Section 9(3) bars tracking and targeted advertising directed at children, which a team can't honour if under-18s sit unflagged in a campaign list.

Discovery: finding the copies nobody listed

Most first inventories are questionnaires. Each department lists the systems it owns and someone compiles a spreadsheet. The method finds applications well. It misses exports, attachments, downloads, scans and forwarded files, because nobody thinks of those as systems.

A questionnaire is also a snapshot. It records what people remember on the day they fill it in. The spreadsheet credit operations emailed last Tuesday isn't in anyone's memory.

Discovery at scale means scanning, and the unstructured side sets the standard. Good scanning reads inside attachments, PDFs, Word files, spreadsheets and compressed folders. It reaches file shares, laptops, backups and old exports as well as live databases. It runs optical character recognition on scans and photos, because many KYC documents exist only as pictures. It reads free text, where identifiers turn up in call notes and comments.

It also has to recognise the full Indian identity set. Aadhaar numbers, PAN, voter IDs, passports, driving licences, Indian mobile numbers, UPI IDs, GSTIN and bank details with IFSC each follow known patterns. Rule 14(5) treats internal references as identifiers too. It names customer identification file numbers, application reference numbers and enrolment IDs. A scanner tuned to foreign formats will pass over most of these.

Pattern matching alone produces noise. A twelve-digit number may be an order reference. Good discovery checks each value against its identifier's validation rules and its surrounding context before it flags a hit. It should also recognise masked, hashed and encoded copies and label them for what they are. A hashed mobile number can often be reversed, so it still needs protecting.

Treat the scan as processing in its own right. File shares and laptops hold employee data alongside customer data. Bring HR and legal in before the first sweep, and record why you are scanning.

Classification for DPDP is more than a security label

Most enterprises already classify data. The labels usually come from information security, such as public, internal, confidential and restricted. Those labels tell a control what to protect. They say nothing about why the organisation holds the data or when it must let go.

DPDP classification adds the attributes the law turns on.

  • Identifier type. Which identifier the record holds, because each one carries its own rules.
  • Population. Whose data it is, such as customers, rejected applicants, employees, guarantors or nominees.
  • Purpose. The specified purpose, which Section 2(za) defines as the purpose mentioned in the notice.
  • Ground. Consent, or the specific clause of Section 7 relied on. The Act has no general "legitimate interests" ground, so "business need" doesn't qualify.
  • Child data. Whether the Data Principal is under 18, the age at which Section 2(f) defines a child. This triggers verifiable parental consent under Section 9(1) and Rule 10.
  • Retention. The period and the law behind it. Section 8(7) allows retention that compliance with law requires.
  • Recipients and location. The processors, group entities and countries the data reaches.

Identifier type matters more than most programmes expect. Sector rules treat each identifier differently.

IdentifierWhat changes its handling
AadhaarNot mandatory for KYC. The number must be blacked out where authentication isn't needed.
PANRequired for KYC, or Form 60. The lender verifies it with the issuing authority.
Voter ID, passport, driving licenceOfficially valid documents under the KYC Directions
Mobile number, emailUsed for OTP signatures and alerts. Changing a registered mobile needs a defined process.
Live photo, video KYCGeotagged and time-stamped. Video KYC data must stay in India.
Bank statements, account numbersCollected for credit assessment, usually as PDFs that travel

Entities that authenticate Aadhaar carry an extra duty. UIDAI's Circular No. 14 of 2025 requires them to keep Aadhaar numbers and connected data in a separate Aadhaar Data Vault. Every other system must use a reference key in place of the number. An unmasked Aadhaar number in a spreadsheet breaks that rule as well as the DPDP Act.

Retention needs the same care. The KYC Directions require identification records to be kept for at least five years after the business relationship ends. Those records include business correspondence. So some emails must be kept while a stale export of the same data should go. Only a classified map can tell them apart.

The same identifier can also mean different things. A PAN in "hr-share/payroll.xlsx" is employee data. The same PAN in a collections export is borrower data, with a different ground and retention period.

One misreading is worth flagging. Many Indian inventories still tag "sensitive personal data or information", a category from rules made under Section 43A of the IT Act. Section 44(2) of the DPDP Act omits Section 43A and the rule-making clause behind those rules, with effect from 13 May 2027. Keep the tag if a sector regulator needs it, but don't mistake it for a DPDP attribute. We cover similar errors in our note on claims about the Act that don't hold up.

Mapping: follow the copies as well as the systems

A system map draws boxes and arrows. The website feeds the CRM, the CRM feeds loan origination, and collections sends a file to an agency. That picture is necessary. It is also where most mapping stops.

A DPDP map needs one more layer. For each arrow, it records how the data moves, in what form and where it lands. Take the monthly collections file. It is an Excel attachment holding names, mobile numbers and overdue amounts. It lands in the agency's inbox, on its SFTP server and on two analysts' laptops. Each landing point is a copy the Act cares about.

Trace five kinds of movement.

  1. Between departments. Marketing to sales, sales to credit, credit to collections, HR to payroll. Internal sharing still needs a purpose and a ground.
  2. To Data Processors. Section 8(2) permits a processor only under a valid contract. Section 8(1) keeps the Data Fiduciary responsible for what the processor does. KYC vendors, video KYC providers and SMS platforms all sit here.
  3. To group entities and other fiduciaries. A group company that decides its own purposes is a separate Data Fiduciary. Section 8(3)(b) then requires the disclosed data to be complete, accurate and consistent.
  4. To registries and bureaus. The KYC Directions require KYC records to be uploaded to the Central KYC Records Registry within 10 days of starting an account-based relationship. Credit bureau pulls and reports are flows too.
  5. Outside India. Section 16 lets the Central Government restrict transfers to notified countries. Rule 15 adds requirements on making data available to foreign States. Rule 13(4) can require a Significant Data Fiduciary to keep specified data in India.

Regulated entities carry sector rules on top of these. Our piece on how DPDP lands in banks and NBFCs covers where they overlap.

Maps decay

A data map is accurate on the day it is finished. After that, the business keeps moving. Every new vendor, form field, campaign tool, report or data pipeline can create a flow the map doesn't show. Unstructured copies decay fastest, because anyone can create one with a click.

Run discovery and mapping as a continuing process. Scan the places where copies pile up on a schedule. Then add business events that trigger a targeted re-scan and a map update.

  • A new Data Processor is onboarded, or an existing contract changes scope.
  • A new form, app screen or KYC step starts collecting personal data.
  • A new report, export or data pipeline is built on personal data.
  • A team adopts a SaaS or AI tool that receives customer or employee data.
  • A merger, acquisition, new product line or new country of operation.
  • A security incident or near miss involving personal data.

Significant Data Fiduciaries have a fixed reason to keep the map fresh. Rule 13(1) requires a Data Protection Impact Assessment and an audit once in every twelve months. An assessment built on a year-old map won't satisfy an auditor.

What breaks without it

A withdrawal that never reaches the campaign list. A customer withdraws consent for promotional messages. The consent records update and the CRM flag flips. Her mobile number still sits in a campaign export held by an outside agency, and the messages keep coming. Section 6(6) required the lender to make its processors stop. Nobody knew the agency held a copy.

An erasure request that misses the laptop. A former applicant asks for erasure under Section 12(3), and no law requires the data to be kept. The team runs the erasure request workflow and deletes the loan record and the CRM lead. Her bank statement PDFs survive on a relationship manager's laptop, and so does the shared drive copy. The organisation has told her the data is gone. It isn't.

A breach where nobody can say whose data was exposed. A shared folder of scanned voter IDs and bank statements turns up open to the internet. Rule 7(2)(b) requires a detailed report to the Board within 72 hours, including the intimations sent to affected Data Principals. If the folder was never scanned, the first days go into working out what was in it. CERT-In's FAQs on the April 2022 directions also say data breaches and leaks must be reported within six hours of noticing them.

Each failure has the same root. The obligation was understood and the process existed. The copy was a file outside the map.

What to do first

StepWhat to doWho leads
1. Pick journeysChoose two or three high-volume journeys, such as loan onboarding, claims or employee joiningDPO with business heads
2. Trace the front doorList the website forms, CMS, CRM, campaign tools and call centre systems in each journeyMarketing, IT
3. Scan where copies pile upFile shares, exports, mail files, KYC images, laptops and backupsCISO, IT
4. Add DPDP attributesTag identifier type, population, purpose, ground, child flag and retentionLegal, data owners
5. Map external flowsMatch processors, registries, group entities and countries to contractsProcurement, vendor risk
6. Clean upDelete stale exports, mask exposed Aadhaar numbers, lock open foldersData owners, IT
7. Set the cadenceScheduled scans of high-risk sources, event triggers, quarterly map reviewDPO, internal audit

Start the scanning with unstructured sources. They hold the copies nobody has listed, so they change the picture most. Prioritise by how many identifiers a scan finds and how widely the source is exposed. A shared drive open to the whole company goes ahead of a locked archive. Rescan high-risk sources at least weekly. Review the full map every quarter and after every trigger event.

Whatever tooling you pick, hold it to five tests. It should scan structured and unstructured sources in one programme. It should recognise Indian identifiers and validate them. It should tag findings with DPDP attributes such as population, ground and retention. It should map flows to processors and countries and keep that map current. It should also produce evidence a Board or auditor can read. If you want an outside view of where your programme stands first, a DPDP gap assessment is a sensible place to start.

How DataSurity finds and maps the copies

DataSurity's discovery and mapping modules are built for the problem this article describes. Here is how they handle it, with the unstructured side first.

Process diagram showing sources feeding DataSurity Discovery, which validates identifiers and produces findings that pre-load the Data Mapping register and the rights portal, with rescans looping back
From scattered copies to one register

Where it looks. Discovery connects to databases, cloud storage, file shares, laptops and desktops, backups and exports. It reads PDF, Word, Excel, email and database dump files. Scanned Aadhaar and PAN cards are read by local OCR and discarded after reading.

How it reads without moving data. It uses read-only credentials and samples data where it sits. Nothing is copied out. Laptop sweeps upload findings, never files. No client data, sample or column name goes to any external AI service.

How it tells an Aadhaar number from an order number. Each value is checked against its identifier's own rules. Aadhaar must pass the Verhoeff checksum, cards must pass Luhn and GSTIN must pass its check character. A finding counts as high confidence only when shape, validation and context agree. Masked, hashed and Base64-encoded copies are caught and labelled. Hashed mobile numbers are marked reversible, because they are.

What each finding says. Every finding records whose data it is, such as customers, employees or guarantors. Dates of birth under 18 raise a children's-data flag for review. Each finding names the section of the Act it touches, a priority, an owner and a one-line action. It also shows exposure, such as plaintext storage and how many accounts can read the data. It flags a backup that holds plaintext when the live system is encrypted.

How copies become a map. Findings pre-load the Data Mapping register with systems and data categories. Each processing activity then takes one lane from the Act. The lanes are consent, a specific Section 7 use, an exemption, out of scope or flagged. There is no "legitimate interests" lane, because the Act has none. Each activity names every population it touches and records its recipients and the destination country of each flow. Retention is reconciled with the sector rules that apply.

How it stays current. Scheduled rescans show what is new, what was fixed and what came back. Rescans also flag systems and data the register doesn't hold yet. Every edit to the register is versioned. Discovery is read-only, so your team makes each fix and the next scan confirms it.

The same findings tell the rights portal where one person's records sit. That is how an erasure request reaches the bank statements on a relationship manager's laptop. DataSurity runs as SaaS hosted in India, in your own private cloud or on-premises.

Count the copies of one customer

Every DPDP control is written against a list of places where personal data lives. When that list holds only the systems of record, the controls work on paper and fail at the edges. The edges are website forms, campaign lists, spreadsheets, PDFs, scans and inboxes.

So try a simple test before you sign off a notice or a consent banner. Pick one customer and find every copy of her mobile number, PAN, Aadhaar copy and bank statements. Look in the CRM, mail, drives, laptops, backups and vendors. Then count how many of those copies your inventory already knew about.

Sources

  1. Legislative Department, Government of India (India Code), "The Digital Personal Data Protection Act, 2023 (Act No. 22 of 2023), as on 19 November 2025", 19 November 2025. Link
  2. Ministry of Electronics and Information Technology, "Digital Personal Data Protection Rules, 2025, G.S.R. 846(E)", 13 November 2025. Link
  3. Press Information Bureau, "Government notifies DPDP Rules to empower citizens and protect privacy", 14 November 2025. Link
  4. Storyboard18, "MeitY seeks industry views on fast-tracking DPDP Act rollout, proposes 12-month compliance timeline", 28 January 2026. Link
  5. Reserve Bank of India, "Reserve Bank of India (Non-Banking Financial Companies – Know Your Customer) Directions, 2025 (Updated as on December 29, 2025)", 28 November 2025. Link
  6. Unique Identification Authority of India, "Circular No. 14 of 2025: Revised guidelines for hosting HSM, Aadhaar Data Vault and authentication application", 4 November 2025. Link
  7. CERT-In, "Directions under sub-section (6) of section 70B of the IT Act, 2000", 28 April 2022, as reproduced by TaxGuru. Link
  8. CERT-In, "FAQs on Cyber Security Directions of 28.04.2022", May 2022. Link