PII detection

The Detection (PII) menu entry gathers two tabs: “Word lists” (/system/pii/wordlists) and “Learning suggestions” (/system/pii/review-queue). Both take effect across the installation, for all tenants.

The word list overview

PII detection (finding personal data automatically) draws on word lists per language — first names, street suffixes, salutations, closing phrases and so on. File-based defaults are shipped with the product; your changes are layered over them as a DB override.

For each language the table shows: language, source, entries, updated, actions. The source badges mean:

Badge	Meaning
File only	The shipped default applies; no customisations.
DB + file	Your customisations are layered over the default.
DB only	The list exists only as a customisation (there is no file default).

The actions per row: Edit (opens the editor), Download JSON, Upload JSON and Reset to default (removes the customisation; a two-step action with a confirmation).

Important for file uploads

A maximum of 8 MB, and the encoding must be strictly UTF-8. Edit exported files with a UTF-8-capable editor (VS Code, Notepad++) — Windows Notepad and older PowerShell versions save in other encodings by default and destroy accented characters.

The word list editor

Edit opens the editor (/system/pii/wordlists/{language}):

At the top is the action bar: Validate (a dry run without saving), Save (active as soon as there are changes — an “Unsaved changes” badge reminds you) and Reload (discards changes after a prompt). Saved changes take effect immediately for all tenants (hot reload, no restart needed).

The content is divided into sections (the left-hand navigation shows entry counts):

Section	Content
Personal names	First and last name lists, plus NameExclusions (words that are never recognised as part of a name).
Addresses	Street suffixes, cities (stand-alone place detection), address vocabulary.
Salutations	Word sequences such as “Dear Mr” that are recognised as one contiguous salutation.
Closing phrases	Phrases that introduce a signature block (“kind regards”).
Email headers	Header keywords (“From:”, “Von:”, “To:”) for detection in printed emails.
Post-processing	The post-processing vocabulary used by detection runs.
Detector vocabularies	The vocabularies of the name, amount and ID detectors; entries extend the built-in default additively.
Institutions & companies	Markers and boundaries for detecting companies and public bodies.
Advanced	Custom regex patterns and key-value rules (see below).

Every individual list offers an input field (“New entry, press Enter to add”), a filter, a sort switch (A→Z / order of entry) and removal by clicking the × on an entry. The JSON tab shows the raw content read-only; the Diff tab (where a customisation exists) contrasts additions and removals against the default.

Example — adding regional first names: a tenant base with many Turkish names? Open the Personal names section → the FirstNames list, type the names in (Enter adds them) and click Save — from the next detection run those names are found across the installation.

Custom regex patterns

In the Advanced section you define regular expressions of your own for PII detection to apply in addition — for staff numbers or internal file references, for instance. Per pattern: a name, the regex (with live syntax checking), a score (0–1, default 0.80 — the confidence of the matches), a category and the languages (empty = all).

Example: to detect file references of the form AZ-2026/0815: name File reference, regex \bAZ-\d{4}/\d{4}\b, category CaseNumber, languages de.

Key-value rules

Also under Advanced: key-value rules detect labelled values in forms (“Customer number: 12345”). Per rule: a label, a category, a value regex, a direction (right/left/above/below — where the value sits relative to the label), label synonyms, soft separators (“=”, “is”, for example), plus limits for the token count and the spatial distance.

Learning suggestions (the review queue)

The Learning suggestions tab collects place names that appeared in detection runs but are not yet in any word list:

• Only non-personal geographic terms are collected — no names, and no tenant identity (only the number of tenants in which the term occurred).
• Filters by status (open/approved/rejected) and language.
• Approve adopts the term into the language’s global city word list (taking effect for all tenants); Reject parks it permanently. For entries that have been decided, the reviewer’s name is shown.