Methodology
How the numbers on this site are collected, cleaned, mapped and shown — and, more importantly, the things they cannot tell you.
The short version
- Five official datasets from four countries, refreshed daily where the publisher allows it.
- Every figure keeps the publisher's own category name. We never rename a category to match a scheme.
- Losses are converted to US dollars at the annual average rate for the year of the loss.
- Everything is reported losses, which are a floor. Nothing here is an estimate of true harm.
- The cleaned data is downloadable on the data page so you can check us.
What counts as a source
Statistics come from national reporting bodies only: the FTC and FBI in the United States, the Canadian Anti-Fraud Centre, the National Anti-Scam Centre in Australia, and UK Finance. Cases come from court records and prosecutors' offices, regulator actions, and established news organisations. Where a case is only reported in the press, the page says so.
Nothing on this site is sourced from a security vendor's marketing report, an industry survey with an
undisclosed method, a press release about a product, or another explainer site. If a number cannot be
traced to a document we fetched and cited, it does not go on the page — the content linter refuses to
publish a scheme page carrying an unresolved TODO(source).
The datasets
| Publisher | Covers | How we get it |
|---|---|---|
| Federal Trade Commission (US) | 2022 – 2024 | Fetched and parsed automatically |
| Canadian Anti-Fraud Centre / RCMP | 2021-01-02 – 2025-09-29 | Fetched and parsed automatically |
| FBI Internet Crime Complaint Center (IC3) | 2023 – 2025 | Fetched and parsed automatically |
| National Anti-Scam Centre (ACCC), Australia | 2021 – 2025 | Hand-transcribed from the published report |
| UK Finance | 2025 – 2025 | Hand-transcribed from the published report |
Why some sources are hand-transcribed
Two of the five publish no machine-readable export. Australia's scam data moved off the national open data portal into a Power BI dashboard, and both of the government domains that host it refuse automated requests. UK Finance publishes its category breakdown only inside a PDF. For those two, the figures are transcribed by hand from the named report, each row carrying the page it came from, and both the data page and every chart footnote say the data is transcribed. Where we could not verify a category in the primary document, the row is absent rather than estimated — the UK seed carries an explicit note listing the categories still missing for that reason.
What happens each morning
A scheduled job re-fetches every source, re-parses it, and re-runs validation. Cleaned data is only replaced if the new copy passes: a broken fetch leaves yesterday's working charts in place rather than blanking them. If a fetcher's parse breaks — because a publisher restructured an export, which they all do eventually — that source falls back to its seeded CSV and is flagged as being on fallback everywhere it appears. Raw downloads are archived with a SHA-256 hash so any figure can be traced back to the exact file it came from.
Mapping agency categories to scheme pages
This is the part most likely to mislead, so it is worth being precise about. Agencies do not classify fraud the way this site does. The FBI has one category called Investment; we have five separate pages that all fall inside it. The Canadian Anti-Fraud Centre files business email compromise under Spear Phishing. Australia's Other category is a residual, not a scam type.
So we keep a mapping file with 156 agency labels in it, and each entry records not just which scheme pages the label relates to but how well it fits:
- exact
- The agency category and the scheme describe the same thing — IC3's SIM Swap, for instance.
- superset
- The agency category is broader than the scheme. Most mappings are this.
- subset
- The agency category is narrower and covers only part of the scheme.
Charts on a scheme page therefore label their series with the agency's category name, not the scheme name, and say in the footnote that the category is broader than the page. A chart on the pig butchering page reads "Investment (US)", never "pig butchering losses", because nobody publishes a pig-butchering line item. The mapping decides which chart appears on which page; it never relabels a number.
Categories we deliberately exclude — billing disputes, credit reporting complaints, product complaints — are listed explicitly in the same file, so "out of scope" is distinguishable from "nobody has looked at this yet".
Currency conversion
Cross-country charts are shown in US dollars, converted at the annual average rate for the year the loss was reported, using the IRS's published yearly average table. Converting at today's spot rate would rewrite every historical figure whenever a currency moved, which is why we do not. Charts that mix currencies say "USD at annual average rate" on their face, and the rate table is published on the data page.
What the numbers cannot tell you
They are reported losses, and reporting is rare
Every agency here collects reports that people chose to make. Most people do not. Shame, confusion, not knowing where to report, believing nothing can be done, and the sheer fact that many victims never realise they were scammed all suppress the count. The FTC has published an estimate that the true cost of fraud to US consumers in a year could run to many times the reported figure. Australia's National Anti-Scam Centre notes that around 86% of the reports it receives involve no financial loss at all, which pulls its median down. Treat every total on this site as a floor.
A falling line does not always mean less fraud
Report counts move for reasons that have nothing to do with crime: a new reporting channel opens, a campaign raises awareness, an agency changes a form, a bank starts handling reports itself. Australia saw text-message scam reports fall by 62% in a year while losses through the same channel rose. Read reports and losses as two different signals, and be suspicious of any story that only uses one.
Categories move between years
IC3 added a Charity category and stopped separately reporting Crimes Against Children between editions. The FTC restructures its subcategories. Where a category simply did not exist in an earlier year, the source prints an asterisk and our data carries a null — never a zero, which would read as "there was none of this".
Countries are not comparable
The four national systems differ in what they cover, who reports to them, what threshold applies and whether business losses are included. UK Finance counts only losses recorded by member banks. IC3 covers internet-enabled crime generally, not just consumer scams. Stacking them, as the homepage chart does, gives an order-of-magnitude picture and nothing finer; the chart says so.
Averages hide the shape
Scam losses are extremely skewed: a handful of victims lose life-altering sums and most lose a few hundred. Where a source publishes a median we show the median, because the mean of a distribution like this describes almost nobody. Where a chart shows an average loss per report, it says so and explains that the denominator includes reports with no loss.
How cases and named people are handled
Cases are drawn from indictments, plea agreements, sentencing announcements, regulator actions and court filings, plus reporting from established news organisations where the primary document is not public. Each case page states the outcome as it stood on the last review date — charged, convicted, sentenced, ongoing, unresolved — because a charge is not a conviction and the distinction matters.
A person is named only when they have been charged, convicted or sanctioned in a public proceeding, or when they are a public figure speaking on the record. Only the legal facts of the proceeding appear. Victims are not named unless they made themselves public in the source we cite. The full policy is on the about page.
Where geography claims come from
Several schemes genuinely do run from identifiable places, and saying otherwise would make the pages useless. Where a page says an operation runs from somewhere, it cites the specific document that documents it — a UN human-rights report, a US Institute of Peace study, a DOJ indictment, a Europol threat assessment. Those claims describe criminal enterprises and the places they operate, never nationalities or ethnic groups, and they are dated: enforcement moves operations, and a claim true in 2023 may not be true now.
Checks that run before anything publishes
- Data validation: schema, ranges, currency coverage, missing datasets. A malformed dataset fails the build.
- Staleness: a warning when a dataset's latest year is more than 400 days behind, so old data is noticed rather than quietly served.
- Content lint: every scheme page must carry all ten required sections in order, at least three sourced cases, and at least five sources.
- Source liveness: cited URLs are checked, and dead links are surfaced for repair.
- A fact audit that lists every numeric claim in a scheme body next to its nearest citation, for review in one pass.
Corrections
If a number here is wrong, or a mapping misrepresents a category, please say so — send the page, the figure and the source that contradicts it via the contact page. Corrections are made promptly and the page's last-reviewed date is bumped. The raw data is downloadable precisely so that this is checkable by someone other than us.