Why PEP data quality matters
A politically exposed person is someone entrusted with a prominent public function, such as a minister, a judge or an ambassador. International standards ask regulated firms to identify PEPs and apply enhanced due diligence. The work is only as good as the list: every wrong row costs an analyst's time, and every missing row is a gap an examiner may find.
Much open PEP data traces back, at least partly, to Wikidata. It is large, open and current. It is also crowd-edited, and it records history: every minister who ever held office, alive or not. How a provider turns that raw material into a screening list decides how many of its alerts are real.
A definition you can test
We started with a definition precise enough to count against. A PEP, for this dataset, is a person who holds an office in one of ten categories and has a name that can be screened.
- 10
- office categories: head of state, head of government, minister, MP, ambassador, governor, mayor, judge, central bank governor, legislator
- ≤5%
- pass mark: no country may lose more than 5% of its living PEPs
- 156
- Azerbaijani minister records (128 people) used as a known-answer check
Method: shadow runs
The previous dataset was built by querying Wikidata live, country by country. Queries that time out silently drop people. The new method reads the complete Wikidata dump once a month and applies the definition to every person in it.
Before changing anything, we ran the new method in shadow mode: it computed what it would add and remove and wrote a report, while screening kept using the old data. We repeated shadow runs, inspected the would-be removals, fixed the rules, and ran again. The live switch is bound to the exact list of removals that was reviewed, rows are switched off and never deleted, and the change history is kept.
What the new definition removed
Figure 1
Rows removed by the new PEP definition, by reason
Show the data
| Category | Value |
|---|---|
| Deceased | 165,000 |
| Occupation only, no office | 28,000 |
| Office outside the 10 categories | 8,000 |
| No usable name | 4,000 |
| Implausible birth year, no death | 2,700 |
The largest group by far is people who have died. A deceased former minister cannot open an account, and their record turns every customer with a similar name into a false alarm. The second group, about 28,000 rows, carried the occupation “politician” without any office held: candidates, activists and party members whose exposure, if any, is not what PEP rules are about.
Figure 2
Share of removed rows that were people who had died
- Deceased165,364 (79.5%)
- All other reasons42,636 (20.5%)
Show the data
| Part | Count | Share |
|---|---|---|
| Deceased | 165,364 | 79.5% |
| All other reasons | 42,636 | 20.5% |
Why a blanket safety cap fails
A sensible safeguard for any data update is a cap: if an update would remove more than a set share of a country's rows, stop and ask. Our monthly update uses a 30% cap. For the one-time switch we first proposed 60%. The shadow runs showed why no single number works.
Figure 3
Share of a country's existing PEP rows removed by the clean-up
Show the data
| Category | Value (%) |
|---|---|
| Vatican | 89 |
| Japan | 77 |
Countries with long, well-documented political histories carry many deceased office holders. A correct clean-up in those countries looks, to a blanket cap, exactly like a software bug. The fix was to replace the cap with review: the switch may only remove rows that were individually approved.
Recall against precision
One shadow run found that about a quarter of the living people it would remove held a generic office, such as “mayor” or “minister”, with no country attached. We tried assigning such offices the holder's citizenship. The wide version of that rule backfired.
Figure 4
A wider country rule: people added against real PEPs recovered
Show the data
| Category | Value |
|---|---|
| People added | 19,000 |
| Real PEPs recovered | 2,500 |
Among those added were UNICEF and UNESCO goodwill ambassadors, whose title happens to contain “ambassador”, and a chess grandmaster. More rows is not better coverage. Each wrong PEP is an alert an analyst must clear and explain.
Who we kept
The pass mark was about living people. Under the final definition, 216 living office holders were lost in total, and every country kept at least 95% of its living PEPs except three small ones, which were over the line by 4 to 8 people each. The pass mark also required Azerbaijan to keep its living ministers, our known-answer check.
- 216
- living office holders lost across all countries
- 3
- small countries over the 5% line, by 4 to 8 people each
- 0
- rows deleted: removals are switched off, with history kept
What compliance teams should ask
- What is your definition of a PEP, and can you show it in writing?
- How do you handle people who have died, and how quickly do you remove them?
- Do you list people by occupation, or only by office held?
- When your data changes, how do you know the change is correct and not a bug?
- Is PEP data ever scored like sanctions data? (It should not be.)
Read how Verifex keeps PEPs apart from sanctions on the PEP screening page, and our companion report on measuring screening accuracy.
This report describes measurements and general regulatory context. It is not legal advice; confirm your obligations with a qualified adviser.
Limitations
- Wikidata is crowd-edited. A missing date of death means the death is not recorded, not that the person is alive.
- Counts other than the deceased total are rounded in our records and are shown as “about”.
- The study covers the Wikidata-derived rows only. About 398,000 rows from national PEP registers (for example Brazil) were not in scope.
- Local council members (about 8,000 people) are outside the 10 categories by decision; some regulators may expect them.
- Relatives and close associates of PEPs are a separate problem and are not measured here.
- Shadow runs measure what an update would do. Results after the live switch will be reported separately.
Data and code
Figures come from internal, read-only shadow runs of the Verifex PEP update (September to October 2026) recorded in the Verifex engineering decision log. No customer data was used. Per-country tables are available to customers on request.