Ilia Zavialov

In numbers

Ilia Zavialov explains why the most common scam is rarely the costliest one

English version for the United States.

Two federal bodies published fraud numbers for 2025, and the totals do not match, because each counts a different set of reports in a different way. Ilia Zavialov reads figures like these for a living, and the total is rarely the useful part of either report. The useful part is the shape underneath it: which category takes the volume of complaints, and which one takes the actual money. This piece follows that shape through investment fraud, business email compromise, the new artificial intelligence marker and the arithmetic of older victims. The question underneath all four is the same one: does the category that happens most often also take the most money, or are the two lists led by entirely different things?

·14 min read·Ilia Zavialov

Ilia Zavialov on why the most common scam is rarely the costliest one, shown by category
Ilia Zavialov on why the most common scam is rarely the costliest one, shown by category

01Two tallies of the same year

For 2025 the FBI's Internet Crime Complaint Center, usually shortened to IC3, recorded 1,008,597 complaints and almost 21 billion dollars in losses, a rise of 26 percent on the year before. Over the same period the Federal Trade Commission ran a separate count and reached 3 million reports and 15.9 billion dollars in claimed losses, against 2.6 million reports and just over 12 billion a year earlier. The two figures sit close enough to look like the same measurement taken twice. They are built from different forms, filed with different offices, and a person who calls a bank first and a local police station second may never reach either federal channel at all.

Adding the two headline totals together produces a number that describes nothing real, since an unknown share of reports reached both agencies and an unknown share reached neither. Neither agency claims to be counting the other one's population, and neither total was built to be combined with anything outside itself. The honest use of either report stays internal to itself: how a given agency's own total splits across its own categories, compared against that same agency's own figures from a year earlier.

What survives the comparison is the shape shared by both counts: a small number of categories carrying most of the money, and a much larger number generating most of the individual complaints. Both agencies reached that shape independently, working from different forms and different populations of reporters. Two bodies that cannot agree on a shared total still arrive at the same underlying pattern, and that agreement is the part worth taking seriously.

02Where the volume sits and where the money sits

Investment fraud accounted for 8.648 billion dollars in the IC3 figures for 2025, more than forty percent of everything the agency recorded that year. No other single category came close. About half of all reported losses in the United States were connected to cryptocurrency, spread across several categories at once, and investment fraud is where the largest share of that connection tends to land.

The category holds so much money for structural reasons that have little to do with any single victim being careless. A scheme built around a fabricated balance has no natural ceiling on how much one person can be asked to send. The interface shows growth, the growth invites a larger deposit, and a request to withdraw is answered with one more fee standing between the victim and their own displayed total. A purchase scam ends the moment an item fails to arrive. An investment scheme can continue for as long as the victim keeps paying.

This also explains why the category stays quiet relative to its size. A person who loses forty dollars on a parcel that never turns up tells the story to anyone who will listen within the hour. A person who has spent eight months believing a balance was real reports far more slowly, because the loss is tangled up with an account of themselves as someone who was patiently building something over that time.

Reported fraud losses for 2025, by category
CategoryReported loss for 2025What the figure actually measures
Investment fraud8.648 billion dollarsMore than 40 percent of all IC3 losses that year
Business email compromise3.046 billion dollarsSecond largest IC3 category, about a third the size of investment fraud
Marked as AI-involved, all categoriesAbout 900 million dollarsFirst year IC3 used the marker, over 22,000 complaints
Marked as AI-involved, investment fraud only632 million dollarsOut of more than 8 billion dollars lost to investment fraud
Victims aged 60 and over, average per complaint38,501 dollarsAgainst 20,699 dollars averaged across all ages
All IC3 complaints, total for the yearAlmost 21 billion dollars1,008,597 complaints, up 26 percent on the year before
All FTC reports, total for the year15.9 billion dollars3 million reports, against 12 billion dollars a year earlier

03A category that barely needs to advertise itself

Business email compromise reached 3.046 billion dollars in the same IC3 figures, the second largest category after investment fraud and close to a third of its size. It produces comparatively few individual complaints, because each event tends to involve one organization, and a single successful message can redirect a payment meant for a supplier, a landlord or an entire payroll run. The people who ultimately notice are usually accountants reconciling a statement days or weeks after the transfer went through.

The mechanism needs no malware and no cloned wallet. It needs a plausible message arriving at the moment an invoice is due, timed around a relationship the finance team already trusts, asking for a payment to move to an account that differs from the usual one by a single digit. The message can be technically simple as long as the timing is right and the amount looks ordinary.

This is the category where the working defense is a written procedure, applied every time, regardless of how convincing any single message looks. A changed bank detail gets confirmed by a phone call to a number that was already on file before the message arrived, never a number printed inside the message itself. A rule that does not depend on anyone noticing anything still holds on the day nobody happens to notice.

04A marker with a number attached to it

The 2025 IC3 report carries a category absent from earlier editions: complaints marked as connected to artificial intelligence. Earlier reports folded the same underlying schemes into whichever category they already belonged to, whether investment, romance or a fake job offer. The new marker covers more than 22,000 complaints and around 900 million dollars, recorded for the first time as its own line.

Set against almost 21 billion dollars in total losses for the year, 900 million looks like a footnote. Set against the investment category alone, the comparison sharpens: of the more than eight billion dollars lost to investment fraud that year, only 632 million carried the artificial intelligence marker. The marker measures visibility. It counts the fraud that involved generated material obviously enough for a victim or an investigator to notice and write it down, which is a narrower and smaller measure than the fraud that actually used generated material during the year.

That distinction limits what the number can do. It cannot describe what share of scripted calls, altered photographs or generated investment platforms already operate without being flagged, because a category this new gets judged mainly by its failures, the attempts clumsy enough to be caught. A floor is still worth publishing. It stops being useful only once it gets read as a ceiling.

05What happens to the arithmetic when the victim is older

Among IC3's figures for 2025, victims aged sixty and over lost an average of 38,501 dollars per complaint, against an overall average of 20,699 dollars across every age group, almost twice as high. The gap sits in the size of each complaint. Older victims file a smaller share of the total complaint count yet carry a larger share of the money, because the sum attached to each of their complaints, on average, runs higher.

Part of the explanation is arithmetic. A person further into a working life is more likely to hold a larger balance in a single account, a pension or a property, so a scheme that reaches that person reaches a bigger number by default, independent of how convincing the approach itself happened to be. Part of the explanation is also patience on the side of the scheme. An approach aimed at an older target can run for months, building a relationship or a sense of obligation long before any request for money appears, in a way that a scheme chasing a quick purchase never needs to.

The result is an average built from a genuine pattern across many complaints. A higher figure among older victims describes which schemes reach that age group and how patiently those schemes get built over time. It says little about how alert any individual victim was on the day a request for money finally arrived.

06What a percentage leaves out

Reading these categories side by side surfaces a habit worth naming directly: a striking percentage rarely means what it first appears to mean without the number it was taken from. Forty percent of all reported losses concentrated in one category is a genuine statement about where the money in fraud tends to end up. It leaves out how many people encountered that category and walked away before sending anything, a count neither report attempts to make.

The same caution runs in the other direction. A category producing a small share of total losses can still be the one a given reader is statistically most likely to encounter, because frequency and magnitude answer two different questions. One asks how often a category gets attempted. The other asks how much it costs on the occasions it succeeds. Business email compromise and investment fraud both score high on magnitude while reaching very different populations: one built around organizations and payment runs, the other around individuals with time and savings to commit.

None of the figures gathered in these reports are wrong for being partial. Each one answers exactly the question its collecting agency set out to ask. The error belongs to a reader who lifts a single figure away from the question it was answering. What remains describes one report, in one year, counted by one method, and nothing wider than that.

07The complaints that never get filed

Every figure in either report begins with somebody deciding to file something, and that decision is not evenly distributed across the population living through a loss. Embarrassment keeps a share of victims silent, particularly where the scheme ran for months and asked the victim to believe something flattering about themselves along the way. A person persuaded to move money for a stranger they believed they loved reports far less readily than a person whose parcel simply never arrived.

Business email compromise sits at the other extreme. An organization that loses a payroll run to a redirected invoice has accountants, insurers and sometimes a board that all expect a formal report to exist. A missing sum of that size is difficult to absorb quietly inside a set of accounts that somebody else is going to audit, so the category's habit of reporting probably tracks its real size more closely than most others do.

This unevenness means the shape described across these sections should be read as a lower bound on some categories and something closer to a full account on others. Investment fraud, business email compromise and the broad older victim numbers all likely undercount the true scale by different amounts, for different reasons. That unevenness is one more reason to treat the shape of the categories as the durable finding and any single total as provisional.

08What the shape implies about a single decision

None of this changes what a person actually does at the moment a message, a call or an investment opportunity arrives. It changes which categories deserve the most attention before that moment happens. A frequent, cheap event is answered well by the payment method used to settle it, since a route that somebody other than the seller can reverse limits the damage a purchase scam can do regardless of how convincing the listing looked.

A rare, expensive event answers to a different kind of preparation. Investment offers and altered payment details are both defeated by a procedure fixed in advance, applied before persuasion has a chance to work on the day. No money moves on the strength of a single conversation, and no changed detail gets accepted without a call back placed to a number obtained earlier, independent of whatever channel the request arrived through.

The artificial intelligence marker and the older victim numbers point at the same idea from two directions. A generated voice or video raises the quality of the persuasion without changing what a sound procedure requires, and a target with more time and a larger balance simply raises the stakes on the same test. The categories differ in who they reach and how much they take. The test that defeats each of them keeps turning out to be the same one.

09A shape that outlives the year

Next year's reports will carry different totals, because the totals never repeat exactly and the categories occasionally get renamed or split apart by the agencies that track them. The 26 percent rise in IC3's own total between 2024 and 2025 is itself evidence that the number moves substantially from one year to the next. A reader who anchors on last year's figure will always be slightly out of date by the time the next report lands.

What tends not to move nearly as much is the ranking underneath the total: a small number of categories, built around persuasion sustained over time, taking the largest share of the money, and a much larger number of categories, built around a single fast decision, taking the largest share of the complaints. That pattern held across two independent counting systems for 2025. It has held across enough previous years that betting against it would take more confidence than the evidence actually supports.

This is the reading Ilia Zavialov keeps coming back to whenever a single headline number gets treated as the whole story. The category that happens to you most often and the category that would cost you the most rarely turn out to be the same one. Knowing which is which, category by category, does more for a decision than any total ever will.

10Questions and answers

Which type of fraud causes the biggest financial losses?

Investment fraud. In the IC3 figures for 2025 it accounted for 8.648 billion dollars, more than forty percent of everything the agency recorded that year, and about half of all reported losses in the United States were connected to cryptocurrency.

Why do the FBI and the FTC publish different fraud totals for the same year?

The two bodies collect separate reports through separate forms, so an unknown share of victims reaches only one of them or neither. For 2025 the IC3 recorded almost 21 billion dollars in losses while the FTC recorded 15.9 billion, and the totals are not meant to be added together.

Does artificial intelligence account for most fraud losses now?

No single report supports that conclusion. The 2025 IC3 figures mark about 900 million dollars as AI-involved out of almost 21 billion in total losses. That figure reflects what victims and investigators managed to notice, and noticing captures less than the true total.

Why do older victims lose more money on average than younger ones?

Partly because older victims tend to hold larger balances in a single account, a pension or a property, so a scheme that reaches them reaches a bigger number by default. Partly because schemes aimed at older victims can afford to run for months, building trust before any request for money appears. In the 2025 IC3 figures the average loss for victims aged sixty and over reached 38,501 dollars against an overall average of 20,699 dollars.

Is business email compromise a risk for individuals or only for companies?

It is almost entirely a company risk. Each event tends to redirect a payment meant for a supplier, a landlord or a payroll run, so the loss lands inside an organization's own accounts. It reached 3.046 billion dollars in the 2025 IC3 figures, the second largest category after investment fraud.

Does the most common type of fraud also cause the most financial damage?

Usually not. Frequency and total loss answer different questions, and the 2025 figures show investment fraud taking the largest share of the money while other categories generate far more individual complaints. Reading only one of the two numbers gives an incomplete picture of where the real risk actually sits.

Monogram of Ilia Zavialov

Written by Ilia Zavialov, digital security consultant. Profile of Ilia Zavialov

11Read next

Back to the blog