The measured problem
Measurement limits
Reported losses are a floor, not a total. An incident must be recognized as fraud, sent through a reporting route and recorded with a stated financial loss before it can appear in a headline loss figure. The size of the gap is not quantified by a validated multiplier.
The FTC says Consumer Sentinel is based on unverified consumer reports rather than a consumer survey.1 Surveys confirm that many experiences never reach complaint systems, but their results depend on definitions and reporting routes. In the U.S. Bureau of Justice Statistics' 2017 survey population, 14 percent of an estimated 3.0 million adult personal-financial-fraud victims reported to police.2 In England and Wales, analysis for the year ending March 2025 estimated that approximately one in eight fraud offences was reported to police or Action Fraud.3
The funnel below shows why a reported-loss total narrows at several distinct stages.
<svg viewBox="0 0 590 410" xmlns="http://www.w3.org/2000/svg" font-family="system-ui, sans-serif">
<title>Illustrative funnel from fraud incidents to reported losses</title>
<g fill="none" stroke="currentColor" stroke-width="2">
<polygon points="35,35 555,35 520,95 70,95"/>
<polygon points="75,105 515,105 480,165 110,165"/>
<polygon points="115,175 475,175 440,235 150,235"/>
<polygon points="155,245 435,245 400,305 190,305"/>
<polygon points="195,315 395,315 360,375 230,375"/>
</g>
<g fill="currentColor" font-size="16" text-anchor="middle">
<text x="295" y="70">All fraud incidents</text>
<text x="295" y="140">Incident recognized as fraud</text>
<text x="295" y="210">Sent to a reporting channel</text>
<text x="295" y="280">Recorded as a complaint</text>
<text x="295" y="350">Reported loss stated</text>
<text x="500" y="398">Illustrative</text>
</g>
</svg>
Figure: Recognition, reporting and recording progressively narrow the incidents represented in reported-loss totals.
Each stage can exclude events for a different reason. Some victims may not recognize the conduct as fraud. Others may report to a bank but not a government agency, or use several routes for the same event. Classification rules also change totals by deciding which experiences count as fraud.
Complaint categories and AI-nexus flags do not establish what share of national losses is attributable to AI.5 Consequently, reported-loss data cannot reveal how much AI changed the total. The practical response is to focus analysis on channels, identity claims and payments.
Boundary. A purpose-built victimization survey can estimate incidents for its own definitions, period and surveyed population. That does not make an older reporting rate a valid multiplier for current complaint losses. Reporting probability varies with the fraud, loss, victim circumstances and available route.4
Quellen
Quizze
What can headline complaint-loss data establish about AI's contribution to fraud losses?
- Complaint data cannot support a numerical share
- The share follows from synthetic-media demonstrations
- The share equals the proportion of online reports
The headline systems classify complaints and losses without isolating AI involvement, so they cannot support that attribution.
Reported losses should be treated as a ___.
- floor within the reporting system
- validated estimate of every loss
- fixed fraction of national losses
Unreported and unrecognized incidents sit outside complaint totals, making recorded losses a lower bound within that route.
An older survey reporting rate can be applied as a fixed multiplier to current complaint losses.
- True
- False
Older survey reporting rates vary by definition and route, so they cannot validate a current loss multiplier.
Kommentare
Noch keine Kommentare. Fang das Gespräch an.