How-to

How to profile a dataset and find personal data

Before you share, publish or model a dataset, find out what is actually in it — including the personal data nobody mentioned.

Request beta access Open the app

What you will do

You will profile a dataset column by column, see the value patterns and top values in each, and identify the columns holding personal information.

Profiling answers the question you should ask before every other question: what is actually in this data?

When this is useful

  • You are about to share or publish a dataset and need to know what is in it first.
  • Someone asked whether a dataset contains personal data and the honest answer is "probably".
  • You have inherited data and the column names are not telling you much.
  • A quality problem is suspected but not located.

Before you start

  • A loaded dataset — Profiling works on datasets in DataLens. Without one the screen tells you to load a dataset via Sources first.

Steps

  1. Open Govern → Quality & Lineage

    Select Govern in the navigation, then Quality & Lineage.

    The screen has tabs across the top: Catalog, Profiling, Keys & PII, Column Naming, Dictionary, Quality Rules, Lineage, Stewardship and Approvals.

    The tab row, with your dataset selected.

  2. Start with the Catalog

    Catalog lists what you have — the datasets available and their basic shape.

    It is the orientation step: how many datasets, how big, and where they came from.

  3. Profile the columns

    Open Profiling. For each column you get its type, how complete it is, and its Top Values.

    Top values are more revealing than a null count. A "status" column whose top values are Active, active, ACTIVE and Actve has just told you what needs standardising.

    A per-column profile with completeness and the most common values.

  4. Find keys and personal data

    Open Keys & PII. DataLens examines the actual values and reports the patterns it recognises: Email, Phone, Date, URL, Postcode/ZIP, Currency, IP Address and Boolean.

    Detection is by content, not by column name. A column called ref_2 holding email addresses is found; a column called email_address holding nothing is not miscounted.

    Detected Value Patterns listed against the columns they were found in.

  5. Act on what you found

    Personal data found where you did not expect it is a decision point, not just a finding. Drop the column in Prepare, restrict who can see the dataset, or exclude it from what you publish.

    The detection tells you where to look. What to do about it is yours.

  6. Record what columns meanPro

    Against each column you can record a Business Label, a Business Meaning, a Domain, an expected Format / Pattern and Quality Notes.

    This is what turns a profiled dataset into a documented one — the difference between knowing a column holds numbers and knowing what those numbers are.

What happens next

Profiling tells you what is wrong; Prepare fixes it. Standardise Values, Cast Type and Fill Nulls all follow directly from what a profile shows.

Where the same quality problem must be caught every time rather than once, Quality Rules turn a finding into a check.

Example

Before publishing a customer extract, an analyst runs Keys & PII and finds email addresses in a column named contact_ref that nobody had flagged. They drop it in Prepare before publishing — a two-minute check that avoided a disclosure.

Tips

  • Profile before you publish, every time. It is the cheapest check available and the one whose absence costs most.
  • Read Top Values, not just null counts. Inconsistent values hide there and nowhere else.
  • Trust content detection over column names. Names lie; values do not.
  • Record business meaning while you are looking at the column. Nobody comes back to do it later.

Limitations

  • Pattern detection covers email, phone, date, URL, postcode/ZIP, currency, IP address and boolean. Personal data that matches none of those patterns — a free-text note containing a name, for instance — is not detected.
  • Detection is a signal, not a compliance judgement. A person decides what the finding means.
  • Column Naming, Dictionary and Quality Rules are Pro capabilities.
  • Profiling runs on datasets loaded into DataLens, not directly against your source system.

Related how-to guides

Set data quality rulesPro

Turn a one-off quality check into a standing rule, so the same problem is caught every time.

Trace where a column came from

Follow a field back from the dataset you are looking at to the source it came from, step by step.

Clean a messy dataset

Drop nulls, fill gaps, remove duplicates and cap outliers — with a live preview before anything is applied.

Publish and export data

Export as CSV, TSV, JSON, JSONL or Excel, publish an API endpoint, or generate a PDF or Word report.

Related questions

More of these on the DataLens FAQ page.

Can DataLens find personal data in my datasets?

Yes. The Keys & PII tab examines actual column values and reports recognised patterns — email addresses, phone numbers, dates, URLs, postcodes, currency, IP addresses and booleans. Detection is based on content rather than column names, so a personal-data column with an innocuous name is still found.

What does data profiling in DataLens show me?

Per column: the type, how complete it is, the most common values, and any value patterns detected. Top values are usually the most informative part — they surface the inconsistent spellings and stray formats that a null count will never reveal.

Try this in DataLens

DataLens is in private beta. Request access and work through this guide on your own data.

Request beta access