Editorial Article
One misread plate is worth USD 2,500 and a camera earns USD 3,000 a year: the error budget is 31 records, and Vallejo's misread rate spends it in 89 reads
The Driver’s Privacy Protection Act does not regulate cameras. It regulates the single instant in which a plate string becomes an accountable person, and it attaches a price to that instant which has nothing to do with the cost of any technology involved.
That price is four paragraphs long and it is short enough to quote in full.
18 U.S.C. §2724(b): “(1) actual damages, but not less than liquidated damages in the amount of $2,500; (2) punitive damages upon proof of willful or reckless disregard of the law; (3) reasonable attorneys’ fees and other litigation costs reasonably incurred; and (4) such other preliminary and equitable relief as the court determines to be appropriate.“
Four remedies, and only one of them requires anything to be established beyond the purpose. No harm, no loss, no damaged party in any economic sense. A person whose record was pulled for a purpose the chapter does not permit is owed USD 2,500 before a lawyer’s letter is written, and the cost of writing it is recoverable on top of that.
What the statute is actually pricing
Section 2724(a) creates the cause of action against anyone who “knowingly obtains, discloses or uses personal information, from a motor vehicle record, for a purpose not permitted under this chapter”. The operative word is knowingly, and the operative failure is not accuracy. It is purpose.
This is a statute about the second step of the chain. A camera reads a plate; that is image processing. A system resolves the plate against a registry of registered keepers, and now there is a name, an address and a record of when it was asked for. The DPPA was drafted in 1994, three years before any commercial ALPR business existed at scale, to stop states from selling driver data. It survived a federalism challenge in Reno v. Condon in 2000, where the Supreme Court described a law that “regulates the disclosure and resale of personal information contained in the records of state DMVs”. It has since become the pricing mechanism for an industry it was never written for.
And it is priced per person, per improper act. One misconfigured query against one registration is not one event. It is one event for every individual whose record the query returned.
What a camera earns, and what that makes the error budget
The other side of the ledger is published in exactly one place: municipal contract awards. No vendor in this market lists a price, so the price a city actually pays is the only visible signal there is. Desert Hot Springs, California, put it on the record — a 26-camera deployment at USD 3,000 per camera per year, USD 78,000 annually.
Divide one published number by the other and the market acquires an accuracy specification nobody wrote into a tender.
| Measure | Value |
|---|---|
| Statutory floor per improper record | USD 2,500 |
| Annual charge per camera (published municipal price) | USD 3,000 |
| One record as a share of a single camera-year | 83.33 per cent |
| Records that equal a 26-camera deployment’s annual spend | 31.2 |
| Records per camera | 1.2 |
Thirty-one records. Across twenty-six cameras, one and two-tenths records per camera per year consumes the deployment’s entire annual fee in statutory exposure — before punitive damages, before attorneys’ fees, before a dollar of defence cost, and without anyone having to demonstrate that they were harmed.
A single improper record against a single camera-year is 83.33 per cent of what that camera is paid for a year of service.
The accuracy that implies, and why it is not a liability estimate
The nearest thing this market has to a published error rate is the BetaGov randomised controlled trial in Vallejo, California, which found that 35 per cent of hits were misreads. Apply that rate to the budget above and the camera is not permitted to generate ninety reads a year.
89.14 reads. One fixed reader on one access road passes that before lunch.
That comparison is deliberate arithmetic on two published numbers, and it is important to say what it is not. It is not a liability estimate, because the statute does not attach to a misread. It attaches to a purpose. A camera that misreads a plate and does nothing with it has disclosed nothing to anybody. The exposure sits on the workflow downstream of the read: the citation mailed, the boot ordered, the invoice posted, the record shared with another party.
Which is precisely why accuracy is sold as a product feature and liability is incurred as a process failure — and why the two are budgeted by different people in different departments. The accuracy number appears in a specification. The exposure appears in a risk register. Nothing in the procurement process requires the first to be divided by the second.
The margin the fine is measured against
Verra Mobility’s Q4 2025 segment margins map the asymmetry into a business. Parking Solutions ran at 8 per cent operating margin, Government Solutions at 24, and Commercial Services at 64. Against a USD 3,000 camera-year those are USD 240, USD 720 and USD 1,920 of operating profit respectively — and one improper record at the statutory floor is 10.42 times the first, 3.47 times the second and 1.30 times the third.
The same statute is a different business depending on where in the stack you sit. At 8 per cent, a single record destroys more than a decade of profit on that camera. At 64 per cent, it destroys a year and a bit. At any margin, it destroys more than the camera earns in a year, because the floor is 83.33 per cent of the annual charge and the margin is always a fraction of the charge.
This is the arithmetic underneath the forecast’s most consequential line. Identity revenue — the resolution step, sold per reader — rises from 18.66 per cent of the market to 29.14 per cent at 20.81 per cent a year, contributing 34.64 per cent of the USD 7,970.0m the market adds between 2025 and 2033. Hardware falls from 36.00 per cent to 24.53 at 8.91. Software holds at 33.97 to 35.56. Services drift from 11.36 to 10.78.
Margin rises with distance from the sensor. So does exposure per dollar of margin. The market is migrating toward the step where a mistake costs most, and migrating there because that is where the contract is — the resolution step is the one the asset owner signs for, not the one the procurement officer specifies.
Tolerance narrows without anyone legislating
Measured per reader in service, the same movement has a second reading. The global market runs from USD 4,180.0m in 2025 to USD 12,150.0m in 2033 at 14.27 per cent, on readers in service growing from 1,860 thousand to 5,240 thousand at 13.82 — an increase of 181.72 per cent. Total revenue per reader is almost flat across the eight years, USD 2,247.31 to USD 2,318.70, a rise of 3.18 per cent. That flatness is not stability. It is two curves crossing: hardware per reader down 29.72 per cent, from USD 809.14 to USD 568.70, and identity per reader up 61.10 per cent, from USD 419.35 to USD 675.57.
Set the fixed statutory floor against the rising identity line and the market becomes less forgiving every year. One USD 2,500 record equals 5.96 reader-years of identity revenue in 2025 and 3.70 in 2033 — a narrowing of 37.92 per cent across the forecast window with no legislative change whatsoever. The penalty is a constant. The revenue it can destroy is not. Hold one still and grow the other and tolerance contracts by arithmetic alone.
How it scales, and what changes at scale
| Deployment | Cameras | Annual spend at USD 3,000 per camera | Error budget (records) | Statutory exposure at the floor | Multiple of annual spend |
|---|---|---|---|---|---|
| Published municipal award | 26 | USD 78,000 | 31.2 | USD 78,000 | 1x |
| Mid-size city | 1,000 | USD 3,000,000 | 31,200 | USD 78,000,000 | 26x |
| National operator | 10,000 | USD 30,000,000 | 312,000 | USD 780,000,000 | 26x |
The multiple is constant, because both sides of it scale with camera count. What changes with size is not the ratio but the discoverability. Thirty-one records in one town is an incident. Thirty-one thousand records across a thousand cameras is a class, and a class is the mechanism that converts a theoretical exposure into a claims estimate that an underwriter can price.
That is why the three changes capable of moving this forecast are all legal rather than technical: a federal amendment narrowing or widening the permitted purposes or altering the liquidated damages provision, which would reprice the entire United States identity arm inside one legislative cycle; a state retention floor at 24 hours or less in a large market, which would strand archive-based products; and a certified class in the Florida litigation. None of the three is in the model. All three are inside the forecast window.
Where liability actually attaches
Section 2721(b) lists the permitted uses, and the mapping to this market is unusually exact. Paragraph (10) covers the operation of private toll transportation facilities — tolling sits on an explicit permission. Paragraph (1) covers any government agency performing its functions and any private entity acting on its behalf — municipal enforcement sits there when the operator is genuinely acting for the authority. Paragraph (4) covers use in connection with a civil, criminal, administrative or arbitral proceeding.
Private parking on private land fits none of them cleanly. It must rely on paragraph (3), which permits a legitimate business to verify information the individual supplied and to correct it, but only to prevent fraud, pursue legal remedies or recover a debt or security interest. That permission requires a genuine debt. If the underlying parking event did not happen, the purpose fails — and it fails per person, at USD 2,500, with fees.
The Levine litigation in Florida is the shape of it. A plaintiff who had driven through a private lot twice without parking was invoiced for “nearly four hours”, the operator’s system having converted presence into duration, and the claim alleges that private operators obtained home addresses from motor vehicle records in order to mail boot-and-tow invoices. The statutory question in that case is not whether the plate was read correctly. It is whether the purpose existed at the moment the record was pulled.
Section 2721(c) then supplies the discovery engine. A recipient may redisclose only for a permitted use, and must keep records identifying each recipient and each permitted purpose for five years. The purpose log the statute compels for compliance is the same log that establishes liability in litigation. One artefact, two functions — which is a large part of why the identity step carries the margin: the compliance record and the commercial record are the same file. The mobile parking payment market runs on the same identity: there the session record, not the payment, is the asset.
Retention is the architecture cost
The federal statute sets the price of the exception. The states set the cost of the architecture. At least 28 have statutes expressly addressing ALPR use or retention, and at least 19 impose retention requirements with three more requiring a local policy: three minutes in New Hampshire, 21 days in Maine, 90 in Kentucky, 18 months in Montana, three years in Colorado, five in Alabama. The spread between three minutes and five years is 876,000 times, and no single archive design satisfies both ends of it. The same pressure is visible one layer down in the parking occupancy sensing market, where processing moved to the sensor for identical reasons of retention and minimisation.
Two state models are worth isolating because they fail differently. California imposes a full operational code — reasonable security, a public usage and privacy policy, an access record carrying date, time, plate or query, username and purpose, a designated custodian, a public comment opportunity before a public agency implements a system, a prohibition on sale or transfer, and a civil cause of action for breach. A vendor complies with California by improving process. Connecticut’s 2026 statute drafts from the other end — a short list of permitted purposes that expressly includes parking enforcement and access control, a 21-day deletion default, a 30-day carve-out for highway usage fee collection. A vendor complies with Connecticut only by changing architecture.
That is the difference this market has to price. It also explains why the European market has converged on ephemeral matching for reasons of principle — GDPR purpose limitation, minimisation and storage limitation produce the same practical instruction — while the United States converges on it for reasons of damages. Same architecture, entirely different motive, and a vendor that builds for one can sell into the other.
What the fieldwork found
Two findings from our own programme sit underneath all of this, and they are ours rather than anyone’s published statistics: 31 semi-structured interviews across seven markets, nine conference-floor interviews at a European traffic-technology exhibition, one closed roundtable with five operators and one toll authority, and 47 written survey responses.
The first is that most operators we interviewed could not state the recognition error rate of the deployment they had bought, and none could state it in the unit the statute prices — records per camera per year. The second is that where indemnity appears in a contract at all, it is negotiated as a substitute for accuracy rather than a consequence of failing to achieve it. The clause transfers the cost of error without ever naming a tolerance. The 31.2 records in the table above exist whether or not anybody writes the number down.
Who is selling into it
The consolidators are buying resolution, not optics. Motorola Solutions acquired VaaS for USD 445m on roughly USD 100m of revenue — 4.45 times — in a deal whose stated asset was a vehicle location database. Metropolis bought SP Plus, converting around 4,000 locations, and Flock Safety reports more than 100,000 readers across roughly 12,000 customers. In Europe the hardware incumbents are Kapsch TrafficCom, Jenoptik and Vitronic, with Genetec and Conduent holding the integration and back-office layers.
Verra Mobility is the most instructive listed comparable in this market and, notably, carries no Wikidata entity — no identifier survives verification — so it is named here without a link. Its margin split is the cleanest available read on where value accrues: 8 per cent at the parking edge, 24 in government services, 64 in commercial services. The company that operates the curb earns least. The company that resolves and bills earns most.
The number nobody writes down
What this market sells is access. What the law prices is the resolution. Between the sensor and the invoice there is an act — a plate string turned into an accountable person — and that act is where the margin migrates and where liability has always sat. A camera whose hardware revenue per reader deflates 29.72 per cent over eight years is not a business under pressure; it is a business becoming a sensor while the identity step above it compounds at 20.81.
Whether that is a good trade depends on a figure that appears in no tender we reviewed and in none of the contracts our interviewees described. For a municipal deployment of twenty-six cameras it is 31 records. Across a thousand it is thirty-one thousand. The statute sets it at USD 2,500 each, and the market that has to live inside it is growing 14.27 per cent a year.
Frequently Asked Questions
What is the statutory minimum penalty for one improper motor vehicle record lookup in the United States?
Under 18 U.S.C. §2724(b)(1) a court may award actual damages "but not less than liquidated damages in the amount of $2,500". No harm has to be proved — the floor applies on the fact of an impermissible purpose — and punitive damages, attorneys' fees and litigation costs are available on top.
How many misread records can an ALPR deployment absorb before the statutory exposure exceeds what it earns?
At Desert Hot Springs' published price of USD 3,000 per camera per year, a 26-camera deployment bills USD 78,000 annually. Dividing that by the USD 2,500 floor gives 31.2 records — 1.2 per camera — as the point where statutory exposure equals the whole year's spend.
Does a misread plate automatically create DPPA liability?
No. The statute attaches to obtaining, disclosing or using personal information for a purpose the chapter does not permit. A camera that misreads a plate and takes no action discloses nothing. Liability sits on the workflow downstream of the read — the citation issued, the boot ordered, the invoice posted or the record shared.
Why does the error budget get tighter every year without any change in the law?
Because identity revenue per reader in service rises while the statutory floor is fixed. In 2025 it took 5.96 reader-years of identity revenue to equal one USD 2,500 record; by 2033 it takes 3.70. Tolerance narrows 37.92 per cent across the forecast window with no legislative change at all.
