Column classification
- Personal data: 0 columns, 0 rows.
- Not personal: 0 columns.
AutoDiscover
Understand a database, read only
Business data lives in systems whose structure only a few people understand.
Demonstration: Wexcombe Retail Group (fictional) · Retail. Fictional data and a scripted answer.
The situation
- Organisation
- Wexcombe Retail Group (fictional)
- Who is asking
- Data governance lead
- Industry
- Retail
- What happened
- The company is connecting a read-only copy of its sales database and must know which columns hold personal data before anything is analysed.
The question asked
Which columns in this database hold personal data, and which classifications are we sure of?
What it reads, and who can see the result
- Schema and a small profile databaseTable and column names, types and statistical profiles from a read-only replica; no row content leaves.
- Reads
- The schema and aggregate profiles (counts, null rates, pattern types) through a read-only account on a replica.
- Does not read
- Individual rows' values at this step; nothing is written.
- Permission
- A read-only database role created by the data owner for the replica.
- Who sees it
- The data governance lead and the database owner.
Try it
Review each flagged record: confirm it or dismiss it. The totals and the answer below follow your decisions.
| Column | Type | Pattern found | Rows | Finding | Decision |
|---|---|---|---|---|---|
| orders.total | decimal | Amounts | 4,800,000 | Matched | None needed |
| customers.email | text | Email address pattern in 97% of rows | 610,000 | Looks like an email address | |
| products.sku | text | Codes | 18,000 | Matched | None needed |
| customers.postcode | text | Postcode pattern | 610,000 | Postcode may identify a household | |
| support.notes | free text | Names and phone numbers appear in a sample | 240,000 | Free text: personal data appears in places and cannot be ruled out elsewhere | |
| stores.manager_id | integer | Links to the staff table | 410 | A key that points to a person | |
| orders.channel | text | Five values | 4,800,000 | Matched | None needed |
- Flagged
- 4 of 7 · 1,460,410
- Personal data
- 0 · 0
- Not personal
- 0 · 0
- Undecided
- 4 · 1,460,410
The answer
Simulated answer. Written for this fictional case, not produced by a live model. Only the figures in it are computed, from what you decided above.
4 of 7 columns may hold personal data, covering 1,460,410 rows.
Discovery reads the shape of the data, not the values: pattern types and null rates are enough to flag an email-like column [S1]. A postcode, a person-linked key and a free-text field are less obvious, and the definitions in the policy decide how they are treated [S2].
You have classed 0 as personal data (0 rows) and 0 as not personal (0); 4 are undecided. A free-text column cannot be cleared by a sample alone.
Relies on: S1 · Replica schema and profileS2 · Personal-data definitions
Review and approval
The data governance lead decides. Approving records the classification used for every later analysis. It does not read any row values and does not change the database.
Not ready to approve. 4 flagged columns are still undecided. Confirm or dismiss each one first.
In the product this is a different person with the right permission. Here you play that role to see the rule.
What it produces
Limits of this example
- A pattern is a signal, not proof; a numeric column may still identify someone.
- The replica must stay read-only; the connection can be revoked by the database owner.
- This is a scripted demonstration on fictional data. No model was called and no real organisation, person or result is shown.
More in AutoDiscover
- AutoDiscover from a folder you approveNobody has time to organise years of files before they can benefit from any tool.
- One model of the organisationEach tool keeps its own idea of who, what and which system, so they disagree.
- The ways your information can reach the platformOrganisations differ in what they will share, with whom, and how.
The AutoDiscover story · This capability in the full register