Robocall & fraud metrics — how the numbers are measured

The numbers that anchor robocall and phone-fraud coverage — billions of robocalls a month, billions of dollars lost to imposter scams, some share of calls flagged as unwanted — come from a small set of recurring sources. They are cited interchangeably often enough that it’s worth being precise about what each one actually measures. None of them is a census of the phone network. Each is a model, a sample, or a count of a particular kind of report, and the differences between those methods matter more than the headline figures suggest.

The central distinction is between volume and loss. Volume sources try to estimate how many unwanted or automated calls cross the network in a given period. Loss sources try to estimate how much money consumers lost to fraud, usually from reports filed after the fact. These are different quantities, measured by different means, and a large move in one does not imply a corresponding move in the other. A drop in robocall volume does not mean fraud losses fell; a rise in reported losses does not mean call volume rose. Conflating the two is the most common error in popular coverage, and it’s the reason a stable reference is useful.

This page treats each major source in turn — what it measures, how it derives its figures, and where its method introduces bias — and closes with guidance on reading the numbers together. Deliberately, it states no specific figures: the figures change constantly and are best taken from each source directly. What’s stable, and worth recording once, is the methodology.

YouMail Robocall Index

The YouMail Robocall Index is the most widely cited source for monthly US robocall volume. YouMail operates a call-blocking and visual-voicemail service, and the Index works by extrapolating from the calls observed across that user base to a national estimate. The company classifies the calls its users receive — telemarketing, scam, payment reminders, alerts, and so on — and scales the observed pattern up to a figure for the country as a whole.

The strength of the Index is its consistency over time. Because the methodology is applied the same way month after month, the trend line it produces is its most reliable output: whether robocall volume is rising or falling, and roughly how the mix of call types is shifting, are questions the Index is well suited to answer. It is the closest thing the industry has to a continuous public barometer of robocall activity.

The caveat is that the Index is an estimate produced by a model, not a direct count of every call on every network. It rests on the assumption that YouMail’s user base is representative enough of the broader calling population for extrapolation to hold, and on the company’s own classification of what counts as a robocall versus a legitimate automated message. Both are reasonable working assumptions and both are sources of uncertainty. The headline monthly number should be read as a modeled estimate with a meaningful margin around it, valuable chiefly for direction and magnitude rather than as a precise tally.

TransNexus

TransNexus approaches the problem from inside the network rather than from the consumer end. The company provides caller-authentication and call-analytics software to voice service providers, and it derives its STIR/SHAKEN attestation and robocall-trend statistics from the call traffic flowing through its service-provider customers. Where YouMail looks at calls as they arrive at consumers’ phones, TransNexus looks at calls as they transit the networks it serves.

This vantage point gives TransNexus genuine visibility into things the consumer-side sources cannot see directly — in particular, how calls are being signed and attested under STIR/SHAKEN, the share of traffic carrying full versus partial attestation, and how those proportions move over time. For questions about authentication and signing behavior, the provider-network view is the natural place to look.

The corresponding caveat is that the picture is shaped by which providers are TransNexus customers. The statistics describe the traffic on those particular networks, which is not necessarily a representative slice of all US voice traffic. A network view is inherently skewed toward the composition of its customer base — the mix of originating, intermediate, and terminating providers that happen to use the software — and trends in that traffic may or may not generalize to the network as a whole. The figures are best understood as an accurate read of a specific, sizable, but non-random portion of the network.

TNS (Transaction Network Services)

TNS publishes a periodic Robocall Investigation Report built on network-level analysis across the carriers it serves. TNS operates infrastructure and call-analytics services for a broad set of carriers, and that position lets it observe traffic moving between networks — not just within a single provider’s footprint. Its reports characterize the share of calls that are unwanted or high-risk, the behavior of bad-actor traffic, and patterns in how suspect calls move across carrier boundaries.

The cross-carrier vantage point is what distinguishes the TNS report. Because TNS sees traffic exchanged among many carriers, it can describe inter-carrier patterns — for example, how high-risk traffic concentrates on particular routes or originates disproportionately from particular kinds of provider — that a single-network view would miss. This makes it a useful complement to sources anchored to one user base or one provider’s traffic.

The caveat is again one of scope and definition. The analysis covers the carriers within the TNS footprint, which is large but not the entire network, and it depends on how TNS classifies calls as unwanted or high-risk. “High-risk” is a risk-scoring judgment, not a determination that a given call was illegal or fraudulent, and different analytics vendors draw those lines differently. The report is strongest as a network-behavior analysis and weakest if read as a precise, universal count of bad calls.

GASA (Global Anti-Scam Alliance)

The Global Anti-Scam Alliance measures a different quantity entirely. GASA aggregates scam-loss and victimization data across many countries, drawing on consumer surveys, member organizations, and partner reporting to build a global picture of how widespread scams are and how much financial harm they cause. Its concern is the harm and prevalence of scams generally, not the volume of phone calls.

This makes GASA the natural reference for the human and financial toll of fraud at a global scale, and for cross-country comparison of how scam exposure and losses differ between markets. Because it synthesizes many national sources, it can speak to questions — how prevalent scams are worldwide, what share of people encounter them, how losses compare across regions — that no single-country or single-network source can address.

The caveats follow from the method. Survey-based and member-reported data depend on who was surveyed, how questions were framed, and which organizations contributed in a given year, all of which can vary between editions and between countries. Self-reported victimization is subject to recall and to willingness to disclose. And because GASA measures financial harm and prevalence rather than call counts, its figures are not comparable to robocall-volume estimates at all — they answer the “how much harm” question, not the “how many calls” question.

Truecaller Global Insights

Truecaller operates a large caller-ID and spam-blocking application, and its Global Insights and US spam-and-scam reporting estimate spam and scam call activity from that app’s user base and its crowd-sourced labeling. When Truecaller users mark calls as spam or scam, those labels feed a shared database; the company’s published estimates extrapolate from that labeling activity to figures about how much spam and scam calling users are exposed to.

Truecaller’s particular strength is geographic. Its user base is very large outside the United States, especially in markets where it is a default calling companion for many users, which makes its insights especially valuable for regions that the US-centric sources do not cover well. For a global or non-US view of spam calling, Truecaller is one of the few continuous sources.

The caveat is the familiar one for any app-derived statistic: the figures reflect Truecaller’s users, who are not a random sample of all phone users, and they reflect crowd-sourced labels, which depend on users choosing to flag calls and on the accuracy of those flags. Labeling behavior varies by market and by how engaged a user base is, so cross-market comparisons carry extra uncertainty. As with the other user-base sources, the value is in the trend and the relative picture rather than in a precise count.

FTC Consumer Sentinel and FTC fraud reporting

The Federal Trade Commission’s Consumer Sentinel Network is a different kind of source again: it is a repository of self-reported consumer complaints, together with the losses consumers report when they file. The FTC’s periodic fraud reports — including its breakdowns of imposter scams and other categories — are built from these reports and from data contributed by partner agencies and organizations. The headline figures describe reported complaints and reported dollar losses.

The authority of Consumer Sentinel comes from its standing and breadth: it is an official, long-running, government-maintained collection, and for the question of what consumers report — which scam types are most complained about, how reported losses are distributed across categories — it is the definitive US source. Its category breakdowns are the closest thing to an official ledger of consumer fraud reporting.

The decisive caveat is that complaint data systematically undercounts actual fraud, because most fraud is never reported. Many victims do not realize they were defrauded, do not know where to report, are embarrassed to, or conclude that reporting will not help. The FTC itself is explicit that its totals represent a fraction of true fraud. As a result, Consumer Sentinel figures should be read as a floor on reported harm and as an authoritative guide to the composition and direction of reported fraud — not as an estimate of total losses, and certainly not as a measure of call volume.

Methodology caveats in brief

A few cross-cutting points are worth holding onto when any of these numbers comes up. Volume and loss are different quantities. Robocall-volume estimates and reported-loss data measure separate things and should never be substituted for one another; a change in one carries no automatic implication for the other. User-base extrapolation introduces sampling bias. Every source that scales up from its own users or customers — YouMail, Truecaller, and to a degree the provider-network sources — assumes its sample is representative, and that assumption is itself a source of error. The terms are not synonyms. A “robocall” is not necessarily a scam, a scam is not necessarily a robocall, and neither is the same as a quantified “fraud loss”; sources that count one should not be read as measuring another. Reported losses undercount. Any figure built from complaints or self-reports is a lower bound on real harm, because most fraud goes unreported.

Reading the numbers

The sound way to use these sources is to match each to the question it actually answers and to resist combining them naïvely. Use YouMail, TransNexus, and TNS for questions about call volume, attestation, and network behavior; use Truecaller for global spam-calling exposure; use GASA for the scale of scam harm worldwide; use the FTC’s Consumer Sentinel for the composition and direction of reported US fraud. Treat the volume figures as modeled estimates valued for their trends, treat the loss figures as reported floors rather than totals, and keep “robocall,” “scam call,” and “fraud loss” as distinct ideas even when a single headline blurs them. Read this way, the sources are complementary; read interchangeably, they mislead.