Where personal data hides: a discovery checklist for databases, SaaS and backups

Every organization has a list of the systems it knows hold personal data. The risk is in the systems that are not on it, and in the copies the listed systems quietly make.

Key points

  • Personal data discovery is the technical work of finding where personal data actually sits in databases, SaaS applications, files, logs and backups, including the places no inventory lists.
  • Personal data hides in places that were never designed to hold it: free-text fields, support tickets, logs, exports, test copies of production databases, shared drives and backups.
  • Start with the systems you already know hold people’s data, find what feeds them and what they feed, then sweep the places that grow without an owner.
  • A scanning tool finds patterns such as email addresses and identifiers, but it cannot tell you why the data exists or who owns it, so its results are a lead, not a finding, until someone confirms them.

Why discovery comes before the legal work

A record of processing, a request process, a deletion schedule and an impact assessment all depend on knowing where the data is. The ICO’s documentation guidance recommends a data-mapping or information review exercise to find out what you hold and where; the GDPR record it supports is on EUR-Lex. Interviews find what people remember. Discovery finds what the systems contain, and the two lists differ more often than anyone expects.

The deliverable is a list of locations, each with an owner, a kind of data and a confidence level. It feeds the record of processing.

Four layers to search, in order

Start with the systems you already know hold people’s data, find what feeds them and what they feed, then sweep the places that grow without an owner. The layers below follow that order.

The order to search for personal data Four layers, from the places most likely to be known down to the places most likely to be forgotten. First the core systems of record. Then the SaaS applications and their exports. Then logs, tickets and test copies. Last, backups and archives. Core systems of record Production databases, HR and billing. SaaS and their exports CRM, support, marketing, analytics. Logs, tickets and test copies Free text, dumps, shared drives. Backups and archives Snapshots, tapes, cold storage.
The search starts where the owner is known and ends where nobody is responsible.

Layer one: core systems of record

List every production database and the application that writes to it. For each, read the schema, not the documentation. Look for the obvious columns (name, email, phone, address, date of birth, government identifier) and for the less obvious ones: IP address, device identifier, location, free-text notes and any column named “comments” or “data”.

Then follow the data out. Which jobs export it, which integrations replicate it, and which reporting database holds a copy? The review of database settings is covered separately in the database security checklist.

Layer two: SaaS applications and their exports

SaaS holds a great deal of personal data under accounts that IT may not administer. Start from the single sign-on application list and the accounts-payable ledger. For each application ask four questions: what personal data goes in, who can export it, where the vendor hosts it, and whether it is a processor or something else under the contract.

Exports matter as much as applications. A CSV downloaded from a CRM for a campaign, a spreadsheet emailed to a contractor and a dataset loaded into a notebook are all copies, and none appears in the vendor’s own administration console. Cloud storage reviews, such as the one on cloud security review, are a practical way to find them.

Layer three: logs, tickets and test copies

Personal data hides in places that were never designed to hold it: free-text fields, support tickets, logs, exports, test copies of production databases, shared drives and backups.

  • Application and web server logs. IP addresses, user identifiers in URLs, and request bodies logged during debugging.
  • Support tickets and chat transcripts. People paste their details, their documents and sometimes other people’s.
  • Error trackers and analytics events. Payloads sent with the error often include what the user was doing.
  • Test and staging databases refreshed from production without masking.
  • Shared drives, chat history and email. The least structured and the hardest to search; sample them rather than trying to clear them in one pass.

Layer four: backups and archives

Backups are in scope because they hold personal data too, which is why a deletion process needs a written position on them: how long they are kept and how deleted records are handled when a backup is restored.

Find every backup location: database snapshots, file backups, mail archives, the vendor’s own backups and any laptop image kept after an employee left. For each, note the retention and who can restore it. The legal position on deletion from backups is for your counsel; the operational position, written down, is what the discovery supports. Retention is covered in records management and retention.

Using scanning tools

A scanning tool finds patterns such as email addresses and identifiers, but it cannot tell you why the data exists or who owns it, so its results are a lead, not a finding, until someone confirms them. Expect false positives from random strings that look like numbers, and false negatives where data is stored in an image, an attachment or an encoded blob. A sound approach is to scan a sample of each layer, review what was found with the owner, and decide which locations warrant a full scan.

Scanning production data needs access. It is agreed in writing before it starts, run with the least privilege that works and with results held under the same controls as the data. We do not need to copy the data out to find where it is.

What to record for each location

  • The name of the system or store, and where it is hosted.
  • The kinds of personal data, and whether any is sensitive.
  • Whose data it is: customers, staff, applicants, visitors.
  • The owner, and who can access it.
  • How long it is kept, and what removes it.
  • How confident you are, and what was sampled.

The method and its deliverables are described on personal data discovery.

How the work is bounded

The scope is agreed in writing before work starts, and the engagement is quoted in writing with it.

SecHB does not issue certifications, attestations or audit opinions: those come from accredited certification bodies, CPA firms and QSAs. The work here is what an organization does to be ready for them.

Nothing here is legal advice. Where a question turns on the law, the work is done alongside the client’s counsel, not instead of them.

Questions we are asked

What is personal data discovery?

Personal data discovery is the technical work of finding where personal data actually sits in databases, SaaS applications, files, logs and backups, including the places no inventory lists.

Where does personal data hide?

Personal data hides in places that were never designed to hold it: free-text fields, support tickets, logs, exports, test copies of production databases, shared drives and backups.

Where should the search start?

Start with the systems you already know hold people’s data, find what feeds them and what they feed, then sweep the places that grow without an owner.

Can a scanning tool do this for us?

A scanning tool finds patterns such as email addresses and identifiers, but it cannot tell you why the data exists or who owns it, so its results are a lead, not a finding, until someone confirms them.

Do backups count?

Backups are in scope because they hold personal data too, which is why a deletion process needs a written position on them: how long they are kept and how deleted records are handled when a backup is restored.

Find it before someone asks

Describe your main systems and where backups go. The reply says what a first discovery pass would cover. See personal data discovery, and for what to do with the findings data subject requests across jurisdictions.

Discuss a scope