Governments do not measure if a fraud case ends. They measure reports received, losses reported, arrests made, websites taken down, funds frozen and campaigns viewed. Not one of those describes what happened to the person who was defrauded.
When the Home Office last examined where fraud cases are lost, it scored them against a nine-outcome police framework, since expanded to eighteen. Not one outcome in either version describes anything that happens to the victim.
This is not administrative laziness, and treating it as such is why the problem has outlived every reform aimed at it. Measurement follows liability. A state measures what it has made itself answerable for, and on resolution almost no state has made itself answerable for anything.
What is currently believed
The orthodoxy deserves its strongest statement, because it is not foolish. An agency cannot be held accountable for an outcome it does not own. A police force controls whether it makes an arrest. It does not control whether a bank reimburses, whether a platform removes a listing, whether a receiving institution freezes an account in time, or whether a foreign jurisdiction answers a request within the year.
Resolution depends on a chain of actors, most of them private and some of them beyond reach. Measure an agency against a number it cannot move and you get gaming of the system, demoralisation of employees, or both. On that reasoning, activity metrics are not an evasion at all. They are the honest limit of what a department can promise, and publishing them is a form of restraint.
The reasoning is sound. The conclusion drawn from it is not, because it treats the ownership of resolution as an external condition rather than as something a government decides.
The evidence
One regulator has crossed the line, and only one
The UK Payment Systems Regulator publishes, by named firm, the share of authorised push payment scam losses reimbursed by value and by volume, the speed at which claims close, and the reasons claims are refused. Over the eighteen months from October 2024 it reports 88% of reimbursable claim value returned. Of the national fraud authorities examined here, no other publishes anything comparable.
Two figures look like exceptions and are not
Both are often raised against any argument of this kind, and so should be disposed of first.
Singapore reports S$140.5m of scam losses recovered in 2025, and the police footnote states that this refers to funds frozen before onward transfer and does not refer to monies returned to scam victims.
The FBI's Internet Crime Complaint Center reports a 66% success rate for its Recovery Asset Team, which is also funds frozen at recipient banks, calculated on 3,020 complaints out of 859,532 received.
- United States: no outcome, resolution or reimbursement measure anywhere in the Federal Trade Commission's annual data book
- Australia: a national report whose victim support section consists of two activity counts, referrals made and support responses given
- United Kingdom: an outcomes framework containing no victim outcome, and a charge or summons rate for fraud of 4%, against 11% for theft and 17% for crime overall
The Australian case is the sharpest. A government measured its own effort across sixty pages, and the only number describing what happened to victims had to be provided by a third-party legal centre reporting that a third of its callers who lost money were being pursued by banks for scam-related debts.
Why it happens
What gets measured gets resourced, and what gets measured is what someone has been made to answer for. The Payment Systems Regulator can publish a reimbursement rate because a rule made in October 2024 created a duty. A duty has a discharge, and a discharge can be counted.
Everywhere else the state has no locus in the victim's recovery. It has not assigned itself a duty and it has not assigned one to anybody else, so there is no discharge to count. The department is left measuring the only thing it controls, which is its own activity, and the activity count then quietly becomes the target. That is how ten separate public awareness campaigns can run at once for five years after consolidation was formally recommended: reach is measurable, and resolution is not.
The sequence runs in one direction only. Liability first, measurement second, accountability third. Every attempt to reverse it, to demand better statistics before deciding who owes what, has produced better statistics about activity. This is why the measurement gap has survived reform. It is not a reporting failure sitting downstream of the real problem. It is the visible shape of the real problem, which is that resolution belongs to nobody.
It also explains why the exception is a payments regulator rather than a police force or a consumer agency. The Payment Systems Regulator was the one institution positioned to create a duty and then measure it, because it regulates the firms through whom the money moves. It did not acquire better data. It acquired a defendant.
The strongest objection
A resolution rate measures the wrong thing. A victim can be reimbursed while the money remains untraced, the mule account stays open, nobody is charged, and the same victim is reachable again next week. Reimbursement is an insurance outcome. Resolution is a case outcome, and conflating them would let a government declare success by writing cheques while the underlying operation continues untouched.
This objection is correct, and it should be conceded rather than argued away. The answer is not that the distinction is unimportant but that a partial measure held by someone beats a complete measure held by nobody. A reimbursement rate is narrow and real. What exists in its place elsewhere is not a better measure. It is no measure.
What follows
For governments and regulators
Publish a resolution rate, and calculate it on everyone who reported rather than on everyone who qualified. Start with the denominator, because that is where the number is made. Over eighteen months, 438,300 claims were submitted under the UK scheme. Of those, 301,500 were judged eligible for reimbursement. The published 88% is a share of the value of those eligible claims alone. The other 136,800 claims, close to a third, sit outside the calculation entirely. They were not counted as successes, and they were not counted as failures either. They are simply omitted from the analysis.
A regulator can raise its published figure by narrowing who qualifies, and nothing in the published figure would show it.
A rate built on everyone who reported would be a worse number and a truer one. It should also be broken out by payment type, and by whether the money went abroad. Those are the hardest cases to resolve, and an agency that quietly stopped taking them on would otherwise look like it was getting better at its job.
For industry
Expect the second metric, and prepare for it. The Payment Systems Regulator already publishes scam value received per million pounds received, by receiving firm. It prices a firm's contribution to a harm it did not cause and cannot be sued for, and puts its name against the number. Singapore has taken one step in the same direction by naming the operator with the most mobile lines disrupted. That is the entire global inventory, and it will not stay that small.
What would show this wrong
Two things would show this wrong.
The first would be a government that has made nobody responsible for resolving fraud and yet publishes an honest count of how many victims got a result. If that exists, the argument here fails, because measurement would not be following liability after all.
The second is the UK figure itself. If it holds steady year after year while fraud keeps rising at the same pace as other markets, then it is not telling anyone how victims fare. It is telling them whether banks are following a rule, which is a different thing and should be given a different name.
Sources and method
Payment Systems Regulator, APP fraud performance data and reimbursement dashboard, Q1 2026, and PS25/5 Consolidated policy statement, May 2025. Singapore Police Force, Annual Scam and Cybercrime Brief 2025, February 2026; the recovery definition is footnote 1 to the recovery figure. FBI Internet Crime Complaint Center, Internet Crime Report 2024. Federal Trade Commission, Consumer Sentinel Network Data Book 2024, March 2025. National Anti-Scam Centre, Targeting Scams 2025, March 2026. Home Office Research Report 97, 2018, for the nine-outcome framework and the attrition analysis. National Audit Office, Progress combatting fraud, HC 654, November 2022, for campaign proliferation and charging volumes.
No figure in this report has been rescaled, modelled or extrapolated. Two are arithmetic drawn from figures in the cited sources, and both are shown in the text: 3,020 of 859,532 complaints, and 438,300 claims less 301,500. Each is quoted from the publication named above on that publication's own base, and each base is stated in the text where it affects the reading. Where two figures are not comparable, that is said rather than reconciled.
