What Crime Statistics Can Teach Us About Market Surveillance
- Dr. Ulrich Harmes-Liedtke
- 4 hours ago
- 9 min read
The same structural blind spots that distort criminal justice data are quietly undermining product safety oversight - and the EU’s own 2024 border controls report confirms every one of them, in writing.
A number buried in the European Commission’s first annual report on product compliance at EU borders should stop anyone who reads it cold. Based on data of the DG TAXUD report in 2024, customs authorities across the EU refused entry to 4,137 al 64,322 non-compliant or dangerous products. Sounds like a lot. The most active Member State refused 175 products per million imports. The least active refused 0.1. That’s a ratio of 1,883 to one - across countries that are theoretically running the same regulatory system.
The Commission’s own reaction, diplomatic but unambiguous: “It is implausible that products imported in less performing Member States are inherently more compliant or less dangerous.” Exactly right. What the 1,883-to-one gap actually shows is that the statistics measure enforcement intensity, not product safety. The three worst performers carried out between one and five inspections per million items. At that rate, you are not surveilling a market, but documenting your own absence from it.
This is a well-known problem in criminology. German researchers call it the Dunkelfeld - the dark field, or dark figure - referring to the gap between crime that actually happens and crime that gets recorded. The Hellfeld is what police write down. The Dunkelfeld is everything else. The two are related, but the relationship is not what most people assume. When policing intensifies, recorded crime goes up. When trust in institutions rises, victims report more. A worsening statistic can mean the system is working better. A quiet statistic can mean nobody is looking.
Product market surveillance runs on exactly the same logic. And the EU’s last Directorate-General for Taxation and Customs Union (DG TAXUD) report - the first published under Article 25(6) of Regulation (EU) 2019/1020, covering customs controls at external EU borders in 2024 - is a near-perfect case study in how this plays out in practice.
What the headline numbers actually mean
Between 2023 and 2024, EU border refusals of non-compliant products rose by 73%. If you saw that figure in a press release, you would read it as alarming - a dramatic deterioration in what’s entering the market. The report itself suggests a more mundane explanation: several Member States simply started checking more. The General Product Safety Regulation (EU) 2023/988 also entered application in December 2024, expanding what counts as a violation. New rules, new violations, same underlying behaviour.
The report also admits a methodological wrinkle that makes year-on-year comparisons unreliable. Some Member States count every customs intervention as a suspension, because their customs authority and market surveillance authority (MSA) are the same body. Others do not. So when the suspension numbers change, you cannot tell whether the product risk has changed or whether a Member State reorganised its paperwork.
★ ★ ★ Intervention, suspention, refusal and release |
In EU customs terminology under Regulation (EU) 2019/1020, a suspension is a specific procedural step: customs has stopped a shipment from being released into free circulation and formally notified the market surveillance authority (MSA) to come and assess whether the product is safe and compliant. The three-step sequence is: 1. Intervention— customs actively examines a consignment (document check, physical inspection, lab test) 2. Suspension— customs formally holds the goods and hands the case to the MSA for a safety/compliance decision 3. Refusal or release— the MSA either clears the goods or orders them refused entry The counting distortion arises from institutional structure: where customs and the MSA are the same body, all interventions are recorded as suspensions; where they are separate, only escalated cases are. |
This is the reporting paradox that criminologists know well. Better institutions, more reports. More inspectors, more violations found. A quieter number is not a safer market, it may just be a less-watched one.
82 items per million
The headline control rate in the 2024 report is 0.0082%: 82 items actively inspected per million released. That means the recorded refusals are drawn from a population where 99.99% of items passed through without active scrutiny. Any statistical inference about the true non-compliance rate of EU imports, based on this data alone, is on shaky ground.
KEY FIGURES FROM THE 2024 REPORT
0.0082% - share of all imports actively controlled for product compliance
74% - of suspended items subsequently released, with no data on whether cleared by Market Surveilance Authorities (MSA) or auto-released by timeout
76× - detection gap between standard and e-commerce import channels
The 64,322 refusals are real. Each one is a dangerous or non-compliant product stopped at the door. But they are not the tip of an iceberg, but a fragment of a fragment. The report acknowledges this and calls for more controls. That is the right call. But intensifying controls will almost certainly produce a statistical surge in detected violations. Whether policymakers read that as progress or failure depends entirely on how well the context is communicated.
The e-commerce gap
Low-value e-commerce parcels, tracked under H7 customs declarations, grew from 1.4 billion items in 2022 to 4.6 billion in 2024. They now account for 88.9% of all items released for free circulation in the EU. Against that volume, here’s what the detection system produced:
★ ★ ★ EU DG TAXUD REPORT — KEY FINDING, 2024 |
Standard customs declarations (H1): 1 non-compliant item detected per 10,676 released items: a discovery rate of 94 per million. E-commerce / low-value declarations (H7): 1 non-compliant item detected per 815,429 released items: a discovery rate of 1.2 per million. A 76-fold gap between the two channels. H7 declarations represented 88.9% of all items released for free circulation in 2024. |
The obvious reading is that inspecting 4.6 billion small parcels is genuinely hard. That’s true. But volume is not the only problem. The platforms driving H7 growth are disproportionately associated with exactly the product categories that lead the refusal data: health products, electronics, and children’s toys. The dark figure in e-commerce is not spread randomly across a mostly-compliant flow. It is concentrated in the riskiest part of the market, in the channel with the lowest detection rate.
Put it differently: if you were designing a system to maximise the gap between what’s dangerous and what gets caught, the current arrangement is a reasonable approximation of it.
The auto-release problem nobody talks about
Of the 244,470 items suspended by customs in 2024, 180,148 (74%) were subsequently released. The report is honest about what it doesn’t know: those releases happened either because a MSA examined the product and cleared it, or because the MSA did not respond within the legally mandated 4 working days, at which point Article 27 of the Regulation requires automatic release.
The report cannot tell us which. The current data infrastructure does not distinguish a deliberate clearance from a bureaucratic default. Given the resource constraints on MSAs across 27 Member States, a meaningful share of those 180,148 releases are almost certainly administrative timeouts, i.e.products waved through not because they are safe, but because no one got to them in time. This is a dark figure in the bright field: the system counted these items as processed, but processing and safety clearance are not always the same.
The electronic interface connecting customs systems to the Information and Communication System for Market Surveillance (ICSMS) is still in the implementation phase. Once adopted, it should finally make this distinction visible. The catch: adoption is optional for Member States.
Where the risk-profiling loop bites
China accounts for the largest share of refusals across all three years covered by the report, with a 179% increase in 2024. The US and UK rank second and third. This pattern gets presented as evidence about those countries’ products. It’s also, at least partly, a story about how the risk-profiling system works.
The Customs Risk Management System (CRMS2) uses criteria developed at the national and EU levels to flag high-risk consignments. Those criteria are informed by past enforcement data. Past enforcement data reflects where enforcement was concentrated. Which means the system tends to find what it is already looking for, in the places it already looks. The report shows no control rates broken down by country of origin, so there is no way to separate genuine risk signals from a self-reinforcing detection loop. The more important question (which origins are systematically under-inspected?) goes unasked and unanswered.
Criminologists would recognise this immediately. Street-level crime is over-represented in arrest statistics, not because it causes the most harm, but because police are deployed where street-level crime happens. Corporate fraud causes far more aggregate damage and leaves a far lighter statistical trace.
How the two systems map onto each other
Criminal Justice Statistics | EU Product Market Surveillance |
Victim willingness to report | Consumer complaint and recall engagement rates |
Police density per district | Inspector intensity per Member State - 1,883:1 spread in 2024 |
Prosecution priorities | Risk-profiling criteria in CRMS2 |
Organised crime in the dark figure | E-commerce platforms: 1 detection per 815,429 items |
Legal reform inflates case counts | 73% refusal surge coincides with General Product Safety Regulation (GPSR) (EU) 2023/988 which entered in force, December 2024 |
Administrative non-response leaves cases unresolved | MSA 4-day timeout auto-releases suspended goods: 74% released, split unknown |
Slow-acting harm invisible to crime recording | Diffuse chemical exposure: no border detection mechanism exists for it |
What this system structurally cannot see
The report covers border controls on imports from third countries. That is the remit. But it is worth being clear about what that remit excludes.
Products already in the market are not counted. Goods that entered through a port with essentially no compliance controls are already on EU shelves. The stock of non-compliant products already in circulation is not estimated anywhere in the document.
Products made inside the EU are not covered at all. Non-compliant domestic manufacturing is tracked separately through Safety Gate — the EU's rapid alert system for dangerous products, which absorbed the old RAPEX function — and through ICSMS. Safety Gate published 4,137 alerts in 2024, with China as the stated origin for 40% of notified products. That looks like confirmation that risk comes from outside.
But Safety Gate records where a product was made, not how it entered the EU market — a cosmetic assembled in an EU member state from Chinese-sourced ingredients shows up very differently in each system. Notification rates also vary as wildly across Member States in Safety Gate as detection rates do in the DG TAXUD data: Italy's notifications jumped twentyfold between 2022 and 2024, driven by a shift in monitoring focus rather than a suddenly more dangerous Italian market.
The two datasets are measuring different things with different methods and different blind spots. Until someone builds an analytical bridge between them, the import-focused border statistics will keep producing the same implicit conclusion — that product risk is primarily something that arrives from outside — without anyone being able to say whether that conclusion is actually true.
The refusal categories reflect what border inspection can find. Health products, electronics, children’s goods dominate. These carry identifiable risks that show up in a document check or a lab test. Endocrine-disrupting plasticisers at concentrations just below threshold limits, microplastics leaching from repeated-use containers, cumulative low-level heavy metal exposure from daily-use products: none of these produces a refusal. They are not detectable through the methods the system uses. The dark figure for slow, chronic, diffuse harm is essentially total, not because the data is incomplete, but because the measurement apparatus was never designed to find this kind of risk.
Both systems measure institutional activity first and actual risk second. That’s not a flaw to fix. It’s a structural property to understand before drawing any conclusions from the numbers.
What to do with this
For regulators, the practical implication is that raw refusal counts are an unreliable management metric on their own. A 73% annual increase tells you something happened: enforcement, legislation, reporting methodology, or all three. It does not tell you whether EU consumers are safer. Publishing control rates, MSA response rates, and auto-release volumes alongside refusal counts would at least make the measurement environment visible. The electronic interface introduced in December 2025 enables this. Optional Member State adoption makes it uncertain.
For companies, the 1,883-to-one spread across Member States is strategically significant. Where your products enter the EU affects the probability of inspection substantially more than most compliance teams assume. Regulatory exposure is partly a function of geography, not just of product quality.
For anyone trying to understand the actual safety of products on the EU market: the refusal data is not a reliable guide to where the risk is. The 64,322 stopped products are real. But the goods entering through channels with 1-in-815,000 detection rates, the goods cleared through MSA timeout, the goods causing harm that accumulates over years rather than hours, are not in the report. They are in the dark figure. And the dark figure, almost by definition, is where the most serious problems tend to live.
TAKEAWAY
The DG TAXUD report is technically careful, and in several places more candid about its own limitations than official publications usually manage. But what the numbers describe, when read carefully, is primarily the behaviour of the institutions doing the measuring, not the state of the market they are measuring. The 73% surge in refusals is largely a measurement issue. The 1,883-to-one gap between Member States is a readout of enforcement density. The 0.0082% control rate is the outer boundary of what the system can see. And the e-commerce channel (88.9% of all imports, with a rate of 1.2 per million) is where the product safety challenge is actually building, mostly unobserved. The bright field matters. But the dark figure is the story.
Videos
References
European Commission, DG TAXUD. Report on controls on products entering the EU market with regard to product compliance in 2024. Publications Office of the European Union, Luxembourg, June 2025. ISBN 978-92-68-27749-2. doi: 10.2778/4250149. Drawn up pursuant to Article 25(6) of Regulation (EU) 2019/1020.
European Commission. The Safety Gate Rapid Alert System in 2025. European Commission, Brussels.
Comments