Join our Newsletter — 33% off our NHI Course
Home› Guides› Fraud Metrics and KPIs Guide: Measure Losses, Friction…
Guide Fraud Prevention

Fraud Metrics and KPIs Guide: Measure Losses, Friction and Reimbursement Properly

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 27 October 2026 13 min read
On this page

Most fraud programmes can say how much they lost last month, but far fewer can say how many good customers they blocked, how quickly they caught a new attack, or how their numbers compare with what regulators now publish. Fraud metrics fail when loss, friction and operations are measured separately, when labels arrive months late, or when definitions change between teams. This guide sets out the core fraud metrics with formulas, the regulator-defined measures that now matter (UK authorised push payment reimbursement and sending and receiving rates in particular), how to build a measurement pipeline that survives label delay, and which KPIs to show to operations, risk committees and boards. It is for fraud strategy, analytics and risk teams in banks, payment firms and digital businesses. The wider programme is covered in the Fraud Prevention Guide for Digital Businesses; this page goes deeper on measurement only.

Key takeaways

  • Measure loss and friction together. A falling loss rate means little if the false positive ratio doubled; every loss KPI needs a paired customer impact KPI.
  • Normalise by volume. Basis points of value and rates per million transactions let you compare months, products and peers; the UK Payment Systems Regulator publishes APP scam data in exactly these units.
  • Labels arrive late. Chargebacks, scam claims and first-party fraud can surface weeks or months after the event, so report by cohort (the month the transaction happened) and restate recent months.
  • Define fraud consistently. Separate authorised from unauthorised fraud before anything else, as the Federal Reserve's FraudClassifier model does, or trends will be distorted by reclassification.
  • Regulators now set some of the metrics. Since 7 October 2024 UK payment firms reimburse in-scope APP scam victims up to £85,000 per claim, which turns reimbursement cost and receiving-side mule rates into board-level numbers.

Classify before you count

Metrics are only comparable if every case is classified the same way. The FraudClassifier model, developed by the Fraud Definitions Industry Work Group and published through the Federal Reserve's payments improvement site, starts with one question: who initiated the payment? That splits fraud into authorised party and unauthorised party paths, independent of payment type or channel, and a companion ScamClassifier model classifies scams. Adopt a scheme like this, then add payment attributes (amount, date, type, channel) so trends can be cut consistently.

DimensionValues to recordWhy it matters for metrics
AuthorisationAuthorised (scam, APP), unauthorised (account takeover, stolen card, counterfeit), first partyDifferent owners, liability rules and reimbursement regimes
Fraud typeFrom a fixed taxonomy, for example the FraudClassifier and ScamClassifier categoriesPrevents "other" growing into the largest category
ChannelCard present, card not present, Faster Payments or ACH, cheque, mobile, branchControls differ by channel
Detection pointPrevented before payment, stopped in flight, recovered, lostSeparates prevention from recovery
Event and label datesTransaction date, report date, confirmation dateRequired for cohort reporting and label delay analysis
ValueAttempted, gross loss, recovered, net loss, reimbursedAvoids mixing attempted and actual loss

Core fraud metrics and how to calculate them

The formulas below are conventional definitions rather than regulatory ones. What matters most is that each is written down once, applied the same way in every report, and owned by a named person.

MetricFormulaNotes
Fraud loss rate (basis points)Net fraud loss value / total transaction value x 10,0001 basis point is 0.01%; report gross and net separately
Fraud rate by countFraudulent transactions / total transactions x 1,000,000Rate per million; useful when values are skewed
Attempted fraud rateAttempted fraud value (prevented plus lost) / total valueShows attack pressure, separate from control performance
Prevention rateFraud value prevented / attempted fraud valueThe headline effectiveness number; needs a clear rule for what counts as prevented
Detection rate (recall)Confirmed fraud caught by controls / all confirmed fraudFraud found only through customer claims counts as missed
Precision (hit rate)Alerts or declines confirmed as fraud / all alerts or declinesFor review queues: confirmed fraud cases / reviewed cases
False positive ratioLegitimate events declined or challenged / fraud events correctly stoppedA ratio of 10:1 means ten good customers affected per fraud stopped; also report as a rate of all legitimate traffic
Approval or completion rateLegitimate attempts completed / legitimate attemptsThe customer side of the trade-off
Chargeback ratioChargebacks in a month / transactions in the same or prior month, by count and valueCard schemes monitor merchants and acquirers on their own definitions and thresholds; use your acquirer's current rules
Recovery rateValue recovered / gross lossFunds frozen at receiving firms, chargebacks won, cheque returns
Time to detect a new attackFirst fraudulent event to first control responseMeasured per incident; the key measure of agility

Choosing denominators

  • Value-based rates: best for loss and financial reporting; dominated by a few large cases, so add count-based rates alongside.
  • Count-based rates: best for operations and customer impact; a thousand small card frauds may matter more to customers than one large scam.
  • Per customer or per account: best for account takeover and first-party fraud, where one account generates many transactions.
  • Per new account cohort: best for application and mule fraud, measured by the month accounts were opened. See the Fake Account Creation and Sign-up Fraud Guide.

Regulator-defined measures: UK APP scams

The Payment Systems Regulator publishes APP scam performance data for the largest firms in three measures, with the latest comparison period from 1 January 2024 to 6 October 2024, the day before mandatory reimbursement began.

PSR measureDefinitionUnitsInternal equivalent
Metric A: reimbursementShare of reported APP scam losses refunded by sending firms, and share of cases fully or partially reimbursed (14 largest banking groups)% by value and % by volumeReimbursement rate and cost, by scam type
Metric B: money sent by victimsAPP scams sent by a firm's customers (14 largest banking groups)£ per £1 million of transactions; cases per million transactionsOutbound scam rate, the sending-side prevention KPI
Metric C: money received by fraudstersAPP scam funds received into a firm's accounts (top 20 receiving firms)£ per £1 million of transactions; cases per million transactionsInbound mule rate, the receiving-side KPI. See the Money Mule Detection Guide

Two published examples show the spread: the PSR data shows Metro Bank's customers sent £377 per £1 million of transactions to APP scams, and Prepay Technologies received £3,132 per £1 million, in the 2024 pre-reimbursement period. The PSR confirmed a maximum reimbursement level of £85,000 per claim from 7 October 2024 and stated that 99.8% of Faster Payments APP scams by volume, and 90% by value, are fully reimbursed within that limit where they are in scope. Its APP scams page says most victims will be reimbursed within five business days, which makes time to reimburse a KPI in its own right. For the scheme itself see the Authorised Push Payment (APP) Fraud Guide.

Other external benchmarks

  • US consumer reports: the FTC reported on 10 March 2025 that consumers reported losing more than $12.5 billion to fraud in 2024, a 25% increase on 2023, with 38% of reporters losing money (27% in 2023) and investment scams the largest loss category at $5.7 billion.
  • UK industry losses: UK Finance's annual fraud report page states over £1.1 billion of losses in 2024.
  • US suspicious activity: FinCEN's February 2023 mail theft alert counted over 680,000 check fraud SARs in 2022. See the Cheque Fraud Guide.
  • EU card and credit transfer fraud: Commission Delegated Regulation (EU) 2018/389, the PSD2 strong customer authentication RTS, ties the transaction risk analysis exemption to a provider's monitored fraud rates, so fraud rate calculation is a compliance metric for EU payment providers.

External figures are context, not targets: consumer report data reflects what victims report, not total fraud, and industry totals mix very different firms.

Measuring rules, models and operations

Programme-level numbers hide which controls are doing the work. Every rule and model should carry its own scorecard, reviewed on a fixed cycle, so that weak rules are retired rather than left to add friction.

Control level metricWhat it tells youAction when it degrades
Rule hit volume and precisionWhether a rule still targets live fraudRetune or retire rules with falling precision; check whether attackers moved
Unique fraud caughtFraud stopped by this control and no otherRules with no unique catches are candidates for removal
Model score distribution and population stabilityDrift in the incoming populationInvestigate the source, then retrain or recalibrate
Recall at operating thresholdShare of later-confirmed fraud the model would have flaggedReview features for the missed cases
Review queue hit rateWhether analysts are spending time on real fraudRaise thresholds or improve routing
Average handling time and backlog ageOperations capacity against demandAdd automation or staff before customers wait
Decision overturn rateShare of declines reversed on appeal or complaintA rising rate signals false positives the holdout missed

Detection and measurement in practice

  • Log what you need to replay: request, signals, rule and model versions, score, decision, reviewer, outcome and final label.
  • Monitor in near real time: sudden spikes in declines, challenge failures or new-payee volume are often the first sign of a new attack or a broken rule.
  • Close the loop: feed confirmed fraud and confirmed good decisions back as labels within days, not quarters.
  • Audit the numbers: have finance reconcile reported fraud loss to the general ledger write-off each quarter.

KPIs by audience

AudienceKPIsCadence
Fraud operationsQueue volume, review hit rate, average handling time, SLA breaches, time to detect new attacks, rule hit countsDaily and weekly
Fraud strategy and analyticsBasis points by product and channel, precision and recall per rule and model, false positive ratio, approval rate, drift indicatorsWeekly and monthly
Customer and product ownersChallenge rate, completion rate, complaint rate linked to fraud controls, time to reimburseMonthly
Risk committee and boardNet loss against appetite, reimbursement cost, sent and received APP rates against peers, top emerging threats, regulatory findingsQuarterly

Setting appetite and thresholds

  • Loss appetite: express it in basis points per product with an amber and red level, agreed with finance.
  • Friction appetite: a maximum false positive ratio or challenge rate per journey, agreed with product owners.
  • Small firms: start with five numbers (gross loss, net loss in basis points, prevention rate, false positive ratio, chargeback ratio) and a monthly review.
  • Large firms: add cohort reporting, model monitoring, peer benchmarking against regulator data and independent validation of definitions.

How to implement a fraud measurement programme

  1. Agree definitions: a written metric dictionary with formulas, numerators, denominators, inclusion rules and owners, signed off by fraud, finance and risk.
  2. Fix the data: every decision logged with inputs, score, rules fired, outcome and later label, joined to transaction and customer records. See the Transaction Monitoring and Behavioural Biometrics Guide.
  3. Measure the unseen: keep a small random holdout or below-threshold sample to estimate fraud you did not stop and good customers you did block.
  4. Handle label delay: report by event cohort, publish an expected maturation curve (share of fraud known after 30, 60, 90 days) and restate prior months.
  5. Pilot dashboards: run the new metrics alongside old reports for two or three months and reconcile differences before switching.
  6. Govern changes: version the dictionary; when a definition changes, restate history or mark a series break. Roll back to the previous definition if reconciliation fails.

Where identity fits

Many fraud losses are identity failures: account takeover through stolen credentials, synthetic identities at onboarding and mule accounts opened with real but recruited identities. Identity metrics, such as account takeover rate per million logins, step-up success rate and onboarding identity verification pass rate, belong in the fraud dashboard. See the Identity Security Metrics Guide and the Account Takeover Prevention Guide.

Framework mapping

Measurement areaReferenceLink to metrics
Fraud classificationFederal Reserve FraudClassifier and ScamClassifier modelsConsistent numerators across channels
APP scam reporting and reimbursementPSR APP scam performance data and reimbursement requirementMetrics A, B and C; reimbursement cost and timeliness
EU SCA exemptionsCommission Delegated Regulation (EU) 2018/389Fraud rate monitoring for transaction risk analysis
Security monitoringNIST CSF 2.0 DE.CM Continuous Monitoring, DE.AE Adverse Event AnalysisDetection timeliness and coverage
GovernanceNIST CSF 2.0 GV.RM Risk Management StrategyLoss and friction appetite statements
Wider security measurementSecurity metrics programmeSee the Security Metrics and KPIs Programme Guide

Common failure patterns

  • Counting attempts as losses: mixing attempted and actual loss inflates the problem and hides control performance.
  • Reporting only the last full month: without cohort restatement the latest month always looks best, because its fraud has not been reported yet.
  • No measure of blocked good customers: without a holdout or appeals data the false positive ratio is a guess, and rules only ever get tighter.
  • Definitions drift: a team reclassifies scams as first-party fraud and the scam trend "improves" overnight.
  • Ignoring the receiving side: firms measure what their customers send but not what their accounts receive, missing mule exposure that the PSR publishes by firm.
  • Targets that reward the wrong thing: paying analysts on cases closed per hour, or setting a loss target with no friction limit, pushes teams to decline good customers or skim reviews.
  • Vanity accuracy: quoting model accuracy on data where fraud is a tiny fraction of events, instead of precision, recall and alert volume.

Practitioner checklist

  • Publish a metric dictionary with formulas and owners, and version it.
  • Classify every case as authorised, unauthorised or first party before counting.
  • Report loss in basis points and per million transactions, gross and net.
  • Pair every loss KPI with a friction KPI such as false positive ratio or approval rate.
  • Report by event cohort and restate recent months as labels mature.
  • Keep a holdout or sample to measure missed fraud and blocked good customers.
  • For UK payment firms, track sending and receiving APP rates per £1 million and time to reimburse against the PSR data.
  • Show the board net loss against appetite, reimbursement cost and peer position each quarter.

Standards and references

Related NHI Mgmt Group resources: Fraud Prevention Guide for Digital Businesses · Fraud Prevention Maturity Model · Authorised Push Payment (APP) Fraud Guide · Cheque Fraud Guide

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 27 October 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.