Fraud and member value

The question behind this research

Abound already manages credit risk as a tradeoff. In the July 22 follow-up, the credit union described roughly $700,000 in monthly loan charge-offs as a level it is comfortable carrying while continuing to lend.

The question now is how to bring that same discipline to fraud: how much fraud loss is acceptable while fraud controls still let legitimate members use their accounts, access their money, and resolve problems without unnecessary friction?

Abound is gathering data to build a “fraud-to-member-value ratio.” This report identifies measures that can inform that work, explains where each measure applies, and names the choices Abound still needs to make.

The short version: how to measure fraud and member value

Eight sources point to a practical starting point: a product-level scorecard that places fraud performance beside the effect of fraud controls on legitimate members.

• PULSE offers the closest direct model. Its debit-card analysis reads net fraud loss alongside transaction approval, suspected-fraud declines, false positives, and individual fraud-rule results.
• Member impact extends beyond approval. The sources also examine access to funds, recovery, effort, abandonment, complaints, and continued use.
• Cost-sensitive analysis can combine outcomes that have credible dollar values. Missed fraud and false positives carry different costs, and the calculation keeps those costs visible.
• The board and management have different decisions to make. The board sets Abound's risk appetite. Management sets and monitors product-level controls within it.

Bottom line: Begin with paired measures for one product or control. Show fraud loss, legitimate completion, access and recovery, member effort, and operating burden side by side. A single composite measure becomes useful where Abound has enough data to defend the values and weights inside it.

A map of the evidence

Eight sources address different parts of Abound's question.

PULSE offers the closest direct model

In brief: PULSE places fraud loss beside legitimate transaction approval, making it the closest direct precedent for the tradeoff Abound wants to understand.

PULSE's analysis of its 2024 Debit Issuer Study reported average net debit-fraud losses of 5.3 basis points, or 2.5 cents per transaction, during 2023. Issuers approved 92.3% of attempted debit transactions and declined 1.4% for suspected fraud.

A lower fraud-loss rate can accompany more declined transactions, including legitimate ones. PULSE therefore reads fraud loss alongside authorization rates, false-positive ratios, and the results of individual fraud rules. It reported that 35% of issuers performed strongly on both net fraud loss and approval.

These figures provide a comparison point for debit-card controls. Other products need their own baselines because a card authorization, an identity check, an account freeze, and a loan application create different decisions and member consequences.

A starting scorecard for fraud and legitimate-member outcomes

In brief: The sources support five groups of measures. Abound can select the measures that fit each product or control and review them together.

Why access and recovery belong beside fraud loss

CFPB examiners found delayed or absent notification after institutions froze accounts for suspected fraud, along with unclear instructions for restoring access. Some consumers lost access to their money for weeks or months. Examiners also found institutions that blocked calls from numbers associated with frozen accounts or diverted them to unhelpful recordings.

NIST tells identity-proofing programs to minimize legitimate-user rejection and application departures, review fraud controls periodically, and provide accessible redress that resolves identity-proofing failures, delays, and compromised accounts.

Together, the sources make notification time, restored access, and resolution part of the evidence about member impact.

How Filene measures member effort

Filene's 2016 study surveyed 5,247 members across 16 credit unions about consumer loans, mortgages, new accounts, and phone support. Its Member Effort Score measures how easy or difficult an interaction is to complete.

Using Member Effort Score after a fraud control or recovery process would be an Abound adaptation of Filene's general service measure. Abound would need to isolate the effort caused by the fraud control and decide how that effort should appear in the scorecard.

A credit-union example

A Visa DPS case study reports that one unnamed credit union prevented about $3.2 million in fraud during the first nine months of the engagement, a 43% increase in prevention effectiveness. The same case reports gross fraud rates of 0.08% for debit and 0.16% for credit, an authorization approval rate of 90.74%, and a client-reported positive member experience.

The case illustrates how one institution presented fraud performance and authorization results together. The credit union supplied the results, and Visa states that it makes no warranty about the completeness or accuracy of the participant's data.

A cost method for comparable outcomes

In brief: Cost-sensitive fraud analysis gives Abound a way to combine outcomes when it has credible dollar values for them.

The Fraud Detection Handbook treats fraud detection as a cost-sensitive problem. Missed fraud and blocked legitimate transactions carry different costs, so the method assigns a value to each outcome and calculates a weighted loss.

A cost matrix includes all four possible outcomes:

weighted loss per decision = [(true negatives × their cost) + (missed fraud × its cost) + (false positives × their cost) + (true positives × their cost)] ÷ total decisions

The handbook's discussion includes the fraudulent transaction, later fraud, customer-service work, and reputational effects in the cost of missed fraud. The cost of a false positive can include verification work and the inconvenience of blocking a legitimate payment.

This method covers the parts of member impact that Abound can value credibly. Access, effort, complaints, or continued use can remain visible as separate measures until Abound has reliable data and a defensible way to value them.

How the board and management use the scorecard

In brief: The board sets the overall risk appetite. Management translates that appetite into product-level controls, thresholds, and monitoring.

NCUA's Risk Management Components places approval of the credit union's risk appetite with the board. The CEO and executive team operate within that appetite, establish operating standards, and monitor exposure.

The scorecard gives Abound evidence for weighing fraud loss against easier member experiences. The board can agree on the strategic tradeoff the credit union is prepared to make. Management can then choose and adjust product-level controls within that appetite.

Decisions before Abound builds the scorecard or ratio

In brief: Before calculating a fraud-to-member-value ratio, Abound needs to make eight choices:

1. What decision will the scorecard inform? Name the product, control, and decision before choosing the measures.
2. What unit and time period will Abound use? A rate per transaction, per application, per member, or per dollar can produce different conclusions.
3. How will Abound measure the fraud side? Choose gross loss, recoveries, net loss, fraud prevented, or a combination, and use the same definition over time.
4. Which legitimate-member outcomes matter for this decision? Select the relevant completion, access, recovery, effort, complaint, or continued-use measures.
5. Which operating costs belong in the view? Include review work, decision time, service contacts, and other costs created by the control.
6. What baseline will show the control's effect? Compare the result with an earlier period, a control group, or another threshold so Abound can separate the control's effect from other causes.
7. Which measures should remain side by side, and which can share a dollar value? Keep the components visible even when Abound calculates a composite.
8. What belongs in the board's risk appetite, and what belongs in management's operating thresholds? Set the review cadence and identify who can adjust each threshold.

Source context

Each source contributes evidence from a particular setting:

• PULSE's article summarizes 2023 debit-card issuer data from its 2024 study.
• CFPB's anonymous findings come from select examinations generally completed from April through December 2023 and may reflect one or more supervised institutions.
• Filene's 2016 research measures effort across common credit-union interactions; applying the measure to fraud controls is an Abound use of that framework.
• NIST and NCUA provide standards and governance guidance.
• Visa's enterprise guidance is commercial and written for merchants. Its credit-union case study uses participant-supplied results, and Visa disclaims responsibility for their completeness or accuracy.
• The Fraud Detection Handbook provides a technical calculation method.

The product-level scorecard in this report is a synthesis of these sources for Abound. Each benchmark and measure retains the scope of its original source.