The Verification Files
TVF-003 · KYC · 30 min listen

Where Commercial KYC Data Comes From, and Why a Match Is Not Proof

Commercial KYC databases hold tens of millions of identity records. The legitimate sources cannot supply that volume. Old marketing lists and breached data can.

Key findings

  • A KYC match shows that two records agree. It does not show where the reference record came from, how it was collected, or whether the person knows it exists.
  • Roughly 15% of a population’s identity details change each year. Keeping a UK reference file current means re-collecting about 55 million adults a year. No legitimate consumer data source produces that volume.
  • In one sample of 100,000 records flowing into a KYC database, 100% appeared in known breaches. In 13%, the date of birth was later than the earliest breach of that identity.
  • Two databases that agree are not two sources if both bought the same record. Stolen data is sold many times.
  • Every party in a KYC check is paid when a match occurs. Only the person whose identity is being used gains from a failed one.

Summary assessment

Commercial KYC data is consumer identity data held by private data companies and queried by API when a business needs to confirm that an applicant exists. It is one of several inputs to electronic identity verification, alongside document scans, biometrics, mobile network checks and the electoral roll.

The assessment of this file is that most of this data is either very old or breached, with a small share that is fresh and properly collected. The checks built on it measure agreement between records. They do not measure where the records came from.

This file concerns the data, not the companies. Many of the firms holding and querying it operate legitimately. The question is what they are matching against.

Source: high. Simon Delaney was asked to run the breach test this file reports. Background in lead generation, identity, verification and software for data processing.

Evidence: the central point, that a match is not proof, follows from how the checks work. Stated for the breach findings. Contact source for evidence. In the one sample tested, every record of 100,000 appeared in a known breach. The replenishment arithmetic rests on the source’s estimates. Other commercial KYC databases have not been tested.

Challenge status: the source has put this assessment to an industry audience and invited rebuttal. None has been received.

What commercial KYC data is and how a 2+2 check uses it

A financial services firm takes an application for an account, a loan, an insurance policy or a crypto wallet. It holds a data spine for the applicant: name, first line of address, postcode and date of birth, sometimes with phone and email.

To confirm the identity, the firm or its KYC provider looks for the same details elsewhere. In a 2+2 check, also called a two by two check, it needs two corroborating sources. One may be the electoral roll. The second is often a commercial database: a privately held file of tens of millions of records, reached through an API.

If the applicant’s details appear in both, the identity is treated as confirmed and the application proceeds.

Line of inquiry 1: where the legacy records came from

Before GDPR, large consumer marketing databases were common in the UK. The figure quoted by seller after seller was the same: 47 million records. The files were sold, brokered and resold. The source, who worked in that market, assesses that most holders had the same data.

GDPR took effect in May 2018. A file with no traceable consent could no longer be used for email or telemarketing without risk. Its origin had been lost through years of resale. That market closed.

The data did not disappear. The assessment is that much of it moved into KYC reference files, for three reasons:

  • Fraud prevention offers a lawful purpose that marketing no longer did.
  • The data is out of sight. Nobody receives a message from it. The people in it cannot know it is held.
  • KYC and anti-money laundering are heavily regulated. That lends legitimacy to everything used in the decision, including data that is not itself regulated.

A regulated use does not make the data regulated.

Line of inquiry 2: the replenishment arithmetic

A KYC check is a check on a person now. A reference file has to keep up with the population. The source’s figures for the UK:

  1. About 15% of identity details change every year. People die, turn 18, move house, arrive and leave.
  2. That is about 8.5 million changed identities a year.
  3. The changes cannot be targeted. Nobody knows in advance which records have changed. The whole adult population, about 55 million, has to be collected again to find them.
  4. That is 6.67 records collected for every useful update. This assumes full national coverage, no duplicates, no fake data and complete identity fields in every record.
  5. The gap compounds. A file that is not refreshed is 30% out of date after two years, 45% after three and over half by year four, on simple addition.

The cost, at lead generation prices:

  • At £2 a record, 55 million records cost £110 million a year.
  • At £5 a record, a more realistic price on social media, £275 million a year.
  • By email, one million sign-ups a month requires 833 million emails a month, at a 4% open rate, 20% of opens clicking and 15% of clicks signing up. One million a month is still under a quarter of what is needed.

A collection operation on this scale would be among the largest consumer marketing programmes in the country. None is visible.

The bodies that do see the whole population are banks, mobile networks, the electoral roll and government. The source reports that their data does not flow into commercial databases.

That leaves lead generation and co-registration sites whose privacy policies include an identity verification opt-in. The source has reviewed those policies to identify which sites feed KYC, and has set up a KYC data company holding 300 million records. The finding: the declared sources do not generate anything close to the volume required.

Line of inquiry 3: breach analysis of live inflow

Status: first-hand source reporting. Single engagement.

A KYC data company asked for breach detection to be run on records flowing into its database. A one-in-N sample of 100,000 records was tested.

Findings:

  • 100% of the records appeared in known breaches.
  • In 13%, the date of birth was later than the earliest breach linked to that identifier. On paper, the person was born after their data was stolen.

Baseline for comparison:

  • Data freshly collected within the last three years typically shows a breach rate of about 50%.
  • Old data runs higher but does not reach 100%. Some identifiers have never been exposed.
  • Sources rated as highly trustworthy, tested the same way, showed about 50%. They also supplied 90% less volume.

Reading: a batch in which every record is breached is consistent with a file taken from a breach. The impossible dates of birth point the same way. Stolen files often lack a field. Where date of birth is missing, someone has to invent it.

One further point. This data is probably accurate. A breach of a bank, a retailer or another transactional system yields better records than the co-registration site the data claims to come from. Accuracy is what makes it pass.

Line of inquiry 4: one record, many databases

In 2022 a journalist investigating how his own data had spread traced it from a single point of entry. He believed it had been stolen from that one place. He mapped where it went next. It had reached a large number of companies, and many of them were involved in KYC.

Stolen and brokered data is not sold once. It is sold to every buyer available.

Reading: identity checks treat agreement between two databases as strong evidence. If both databases bought the same record, the 2+2 check has found one source twice. The same applies to recycled and synthetic data.

Who is paid when a record matches

  • The firm taking the application wants sign-ups.
  • The KYC provider is paid when a match is returned.
  • The database holder is paid when its record is the one that matches.

Each party earns more as the match rate rises. Only one party gains from a failed match: the person whose identity is being used. The source’s reading is that this is a volume market, as list data was before GDPR, and that it rewards quantity of data over its quality.

Context: identity fraud in the UK

Status: corroborated by published industry data.

  • Cifas, January to June 2026: more than 220,000 fraud cases were filed to the National Fraud Database, the highest first-half total on record. Identity fraud accounted for 59%, close to 130,000 cases, up 9%. Unauthorised SIM swap cases rose 402%.
  • UK Finance: £1.28 billion was stolen through payment fraud in 2025, up 4%.

Billions of identity records are stolen every year. The source assesses that they are monetised twice. They are used to commit fraud. They are also sold into the reference data that fraud checks rely on.

If so, the check favours the fraudster. An identity taken from a breach is more likely to match a reference file built from the same breach. This is assessment, not established fact. It is not known whether fraudsters select identities on this basis.

What counts as provenance

Not provenance:

  • A collection URL on its own. Without the metadata of the submission, it can be invented.
  • A signed contract. Few are read before signing, least of all where the data is of doubtful origin.
  • An opt-in timestamp. It can be invented.
  • Ownership of the data.
  • Several holders of the same record.

Provenance:

  • The collection URL together with the metadata of the original submission: where, when and how the person signed up.
  • Confirmation that the person knows it happened, which rules out a bot.
  • The same checks run across batches. A single record with plausible metadata can look sound. A pile of them from one place does not. See why valid data is not a real person.

Four tests apply to any reference record: origin, collection, freshness and independence.

Assessment

A match rate shows that two records agree. It does not show where the reference record came from, how it was collected, whose data it is or whether that person knows.

The provenance of the commercial KYC data layer is unverified. The assessment stands until evidence is produced against it. For how fabricated leads can enter the same reference files, see how fake leads become trusted KYC data.

The countermeasures follow below.

Listen to episode 003.

Countermeasures · TVF-003

Countermeasures: KYC data of unknown origin

A KYC match shows that two records agree. If the reference record is old, breached or copied across databases, the match says nothing about the applicant.

  1. Ask how the data was collected

    Put five questions to any provider of commercial KYC data. How was it collected? Was the collection incentivised? What is the original sign-up date for each person? Which URL did they sign up on, and what metadata sits behind it? How was traffic driven to that site? A provider that cannot answer does not know what it holds.

  2. Get the original sign-up date

    The date passed on is often the date someone opened an email or clicked a link. That is not the date they gave their details. Without the original date and time, the age of the record is unknown. Breached data is often nine or more years old.

  3. Treat very large volumes as a warning

    Driving traffic at scale is difficult and expensive. A source offering very large volumes has to show where the traffic came from.

  4. Weight the match by its provenance

    A match is not one grade of evidence. A record from a high-intent mortgage enquiry five days ago is strong. A record from an incentivised competition entered five years ago is weak. Know the type of site, the method and the date, and weight the match to fit.

  5. Test whether sources are independent

    Compare the type of site and the original registration dates across sources. The same person on two co-registration sites on the same or very similar dates, with the same breach patterns across both, most likely came from one place. That is one source, not two.

  6. Test the data by originating site

    Run a one-in-N sample on each originating website, not on the aggregator as a whole. Check the breach rate. Check dates of birth against the earliest breach of each identity. Check the timing of original submissions for mechanical patterns. Look for clustering in dates of birth or surnames, which is uncommon but points to invented fields. Fresh data runs at about 50% breached. Ask harder questions as a source approaches 70%.

  7. Require provenance with every record

    A supplier should deliver more than the identity fields. Each record needs the URL it was collected on, the date and time of the original sign-up, where the traffic came from, and the supporting metadata: IP address, device, operating system and browser. The URL counts only when the metadata comes with it.

  8. Audit the file you already hold

    Data holders need far richer provenance for existing records as well as new ones. Test the whole file, or sample it one in N by element. Where breached data is found in the inflow, remove the data or remove the source.

  9. Pay on provenance, not on a match

    Payment today is per match or per API call. Tie it to provenance instead, with different rates for different grades. A recent, high-intent, well-evidenced record should earn more than an old co-sponsored sign-up.

  10. Check what any persistent key is built on

    A persistent key links the same person or address across datasets under one identifier. The key is only as good as the data that supplies it. Tools and services built on weak reference data inherit the weakness. Ask what data sits underneath before relying on the key.

What does not work

  • A match on its ownIt makes money for everyone involved. It says nothing about where the reference record came from.
  • A click or open dateOften passed on as the sign-up date. It is not one.
  • A collection URL on its ownA URL with no metadata behind it proves nothing. The URL together with the submission metadata is the evidence.
  • A signed contractIt can exist without the sign-up having happened. Few are read.
  • More sourcesTwo databases holding the same record are one source.
  • A regulated settingKYC is regulated. The data used in it is not.
  • An accuracy testBreached data describes real people and is usually accurate. Accuracy does not make it lawful or current.

Limits

Breached data is real, so no accuracy check will reject it. The failures lie elsewhere. The data is probably stolen, which is a compliance problem and funds the crime the check exists to prevent. Its true age is lost with its provenance. And no countermeasure removes the appeal of easy money from a very large database.

Example checks

Example checkIllustrative
Reference record A: provenance
site type
mortgage enquiry
sign-up
original date, 5 days ago
traffic
source shown
metadata
IP, device, browser
EvidencedMatch with provenance. Carries weight.
Example checkIllustrative
Reference record B: provenance
site type
incentivised competition
sign-up
click date only
breach
exposed 9 years ago
second source
same record, same date
FlaggedMatch without provenance. Carries none.

Example checks are illustrative. They are not real records.

Operational resources

Relevant services and organisations that can help implement these countermeasures.

Provero

Provero was founded by Simon Delaney, who produces The Verification Files.

  • Provero: Source audit

    Assesses a supplier’s whole file or a sample before it is bought or used, and reads breach history across the file. It does not prove the provenance or legitimacy of commercial KYC data.

    Supports: 06 Test the data by originating site · 08 Audit the file you already hold

A resource is one way to carry out a countermeasure. Listing is not an endorsement. Some countermeasures may only be able to be applied via a resource. How resources are chosen

Frequently asked questions

Where does commercial KYC data come from?

Commercial KYC data is consumer identity data held by private data companies and queried by API during identity checks. Its declared sources are lead generation and co-registration sites whose privacy policies permit use for identity verification. The assessment in this file is that those sources cannot supply the volume required, and that much of the remainder is pre-GDPR marketing data and breached records.

What does a KYC match prove?

A KYC match proves that the details an applicant gave agree with a record in a reference database. It does not prove that the reference record is genuine, that it was lawfully collected, or that the applicant is the person it describes. If the reference data was built from a breach, a fraudster using the same breached identity will match.

What is a 2+2 check in KYC?

A 2+2 check, also called a two by two check, confirms an identity by finding it in two separate data sources. One is often the electoral roll. The second is often a commercial database. The check assumes the two sources are independent. If both trace back to the same original record, it is one source counted twice.

Why is a 100% breach rate in KYC data a warning sign?

Because legitimately collected data does not reach it. In the source’s testing, data collected within the last three years shows a breach rate of about 50%. Even old data falls short of 100%, because some identifiers have never been exposed. A batch in which every record appears in a breach is consistent with data taken from a breach.

How much data does it take to keep a KYC database current?

About 15% of identity details change each year through deaths, people turning 18, house moves and migration. The changed records cannot be targeted in advance, so the whole adult population has to be collected again to find them. For the UK that is about 55 million records a year, or 6.67 records collected for each useful update.

Do several matching KYC data sources confirm an identity?

Only if the sources are independent. Breached and brokered data is sold to many buyers at once. When the same record sits in several databases, a match in each looks like corroboration and is in fact a single origin repeated.

What should you ask a KYC data provider about its data?

Ask how the data was collected, whether collection was incentivised, the original sign-up date for each person, the URL they signed up on with the metadata behind it, and how traffic was driven to that site. Ask for the original sign-up date specifically, because the date supplied is often the date of an email open or a click. Very large volumes are a warning sign, because traffic at that scale is difficult and expensive to generate.

What counts as provenance for KYC data?

Evidence of origin, collection, freshness and independence. That means the metadata of the original submission, confirmation that the person knows the submission happened, and patterns checked across batches of records. A collection URL on its own, a signed contract, an opt-in timestamp and ownership of the data are not provenance. Each can be produced without the event having taken place. A URL becomes evidence only when the submission metadata comes with it.

Exhibits: sources and further reading

Related files