Insurance brokers often rely on a clean Driver and Vehicle Licensing Agency (DVLA) or Police National Computer (PNC) check to reassure clients on a used vehicle purchase, or to help inform motor underwriting decisions.
New research from QuestGates, the UK's largest owner-managed loss adjusting and claims solutions group, shared this week, suggests that data may understate the true scale of vehicle theft, driven by a police administrative process known as weeding.
A vehicle can be genuinely stolen and still return a clean result on both the PNC and DVLA within weeks of the theft being reported. That has direct implications for brokers who facilitate or recommend provenance checks as part of a used vehicle sale or finance arrangement. A client who later discovers they have bought a stolen vehicle that showed as clear at the point of purchase may look to the broker who pointed them to that check, even though the underlying failure sits with police data handling rather than the broker's own process.
There is also a slower-moving underwriting risk. Motor theft pricing models draw partly on recorded theft data by postcode and vehicle type. If actual thefts are running meaningfully higher than official figures suggest, certain areas or vehicle categories could be systematically underpriced, an issue that would typically only surface later through deteriorating loss ratios.
When a vehicle theft is reported, a Lost or Stolen (LoS) marker is placed on the PNC.
If that marker is not confirmed by the constabulary within six weeks, the record is automatically removed, known as weeding, and the LoS status is never passed to the DVLA.
Affected vehicle registrations then show as clear on PNC and DVLA checks, sharply reducing the chances of recovery and leaving the data brokers and insurers rely on incomplete.
QuestGates submitted Freedom of Information requests to 44 individual constabularies across England and Wales.
The analysis found that the DVLA recorded 87,452 stolen vehicles in 2025, compared with a police-reported total of between 107,734 and 115,710, depending on methodology. Using the most conservative approach, that leaves a discrepancy of 20,282 vehicles, an 18.83% gap between the two datasets, rising to as much as 24.42% under alternative methodologies, or roughly one in four reported thefts.
The research has prompted the National Audit Office (NAO) to raise the issue with the Home Office.
"We thank the National Audit Office for engaging with this important matter and for raising it with the Home Office," said Philip Swift, technical director for motor at QuestGates. "As far as we are aware, this is the first time the government has recognised that weeding is a problem."
Swift noted that legitimate factors, such as prompt recoveries or cancelled reports, can also explain some difference between the two datasets, but said such trends would be expected to appear evenly across constabularies. Instead, the research found wide variation. In Merseyside and Gwent, for example, the DVLA is alerted to fewer than half of all reported vehicle thefts.
QuestGates is calling for the issue to be prioritised as part of the new National Law Enforcement Data Service, for an interim measure removing the requirement to double-confirm reported thefts, and for an official audit of historically weeded records with a view to reinstating them.
For brokers, the practical takeaway is that a clean check should be treated as necessary but not sufficient when advising clients on used vehicle purchases, and factored in as a caveat when discussing provenance or history checks as part of any add-on service.
For underwriters and portfolio managers, the wider question is whether current theft data is a reliable enough base for pricing and risk selection in affected postcodes and vehicle categories.
Whether this leads to reform will depend on what comes of the NAO's engagement with the Home Office. If the interim fixes QuestGates is seeking are adopted, both fraud detection and underwriting accuracy in motor could improve.
If not, the gap between reported and actual theft risk will likely keep widening beneath the surface of data the sector currently treats as authoritative.
