Beyond Functional Testing: How Financial Software Can Detect Fraud and Reduce Security Risks

The purpose of functional testing is to answer a simple question: is the software doing what it is supposed to do? For developers working in finance, however, there are many questions beyond this.

It is possible to build a transfer feature where the money moves correctly, balances are updated and the right information is displayed for both accounts. The feature may also pass every functional test. But if a criminal can use it to initiate fraudulent transfers, the software still presents a serious risk.

Functional testing does not determine whether an action is legitimate within a particular context. There are many situations in financial services where the action itself may not be incorrect. What matters is what is happening around it.

Functional Testing Does Not Detect Intent

Most software testing involves verifying that an application behaves as expected. A team might check whether a login form validates input correctly, whether a transfer reaches the intended account or whether the balance is updated accurately. In each case, the test begins with known input and an expected outcome.

Fraudulent behaviour does not follow the same pattern. For example, an account created using stolen personal information may still satisfy every registration requirement. Phishing can allow a criminal to obtain valid credentials and log in successfully.

From the software’s perspective, nothing may appear to have gone wrong. This is why fraud detection needs to complement conventional software testing rather than remain entirely separate from it.

Financial Software Must Monitor Behaviour

In many cases, it takes a combination of different behaviours to reveal the full picture. Consider a customer who initiates a transfer of £500. This may be completely normal for one person but highly unusual for another. There may also be legitimate explanations. The customer may have changed devices or travelled.

Individual behaviours can appear suspicious in isolation but become easier to interpret when viewed together.

Consider another scenario. A customer accesses an account from a new device, changes the password, adds a new beneficiary and then initiates an unusually large transfer. Each event could have a legitimate explanation. Taken together, however, they justify closer examination. Modern real-time monitoring systems can assess these surrounding conditions.

Signal What It May Help Detect
New device or browser Possible account takeover
Unusual transaction value Behaviour outside the normal pattern
Rapid transaction sequence Automated or coordinated activity
Failed login attempts Credential attacks
Linked accounts Organised or synthetic fraud
Unusual withdrawals Potential fraudulent cash-out

Fraud should not be assessed using a single signal. The combination of several indicators provides the context required to interpret an event more accurately.

Beyond Functional Testing: How Financial Software Can Detect Fraud and Reduce Security Risks

Interpretation Through Risk Scoring

Software may receive many different signals, but it still needs a way to interpret them. A weighted risk-scoring system can combine those signals into a broader assessment. A routine payment from a familiar customer may receive a low score. A series of transactions from a new device accompanied by unusual behaviour may receive a much higher one.

The software can then respond according to the level of risk. Normal activity can continue without interruption. Medium-risk activity may trigger additional verification. High-risk transactions may be placed on temporary hold and require further review.

This approach prevents every customer from being treated as equally suspicious. Guidance from the US Federal Reserve supports a risk-based approach. It recommends identifying higher-risk transactions and applying stronger controls where the potential for financial loss is greater.

Continuous Security After Authentication

A successful login does not mean that all subsequent activity is secure. Credentials and sessions can be compromised by attackers. A legitimate account holder may also choose to perform fraudulent actions. Continuous monitoring is therefore necessary after authentication. Recommended controls include:

  • audit and transaction logging;
  • behavioural analysis;
  • device monitoring;
  • transaction velocity monitoring;
  • anomaly detection;
  • alerts for changes to account details.

Federal Reserve guidance on fraud detection also highlights the importance of monitoring customer behaviour, transaction velocity and changes in login patterns. Maintaining transaction and audit logs supports both detection and investigation.

Monitoring should be part of normal operations rather than something introduced only after a security incident.

Hidden Relationships Between Devices

Accounts do not always exist independently of one another. Several apparently unrelated accounts may share devices, technical characteristics, payment methods or behavioural patterns. Device intelligence and relationship analysis help identify these connections.

Frogo’s approach to financial fraud detection combines real-time transaction monitoring with device fingerprinting, behavioural signals, risk scoring and graph-based analysis. Its financial product is designed to identify relationships between accounts, cards and sactions that may not be obvious when each event is reviewed separately.

Graph analysis is particularly useful when fraud involves a network rather than one suspicious customer. A single account may look ordinary. Twenty accounts connected through devices, cards or payment patterns tell a different story.

Addressing False Positives

An overly aggressive fraud system creates another problem: legitimate users may be blocked. False positives have real consequences. Payments are declined, withdrawals are delayed and customers may be required to complete repeated identity checks.

In some cases, a fraud detection system that blocks too many legitimate users can damage the overall customer experience and create unnecessary operational costs. Reducing false positives should therefore be an important objective for fraud teams. Relevant questions include:

  • How many legitimate transactions are being blocked?
  • Which rules generate the highest number of false positives?
  • Are legitimate users being challenged for ordinary behaviour?
  • Can final investigation outcomes be incorporated into future risk scoring?
  • Are customer behaviour patterns changing faster than fraud-detection thresholds?

The purpose of fraud software is not to block as much activity as possible. It is to distinguish suspicious activity more accurately.

Cybersecurity and Fraud Detection Need to Work Together

Fraud detection does not replace cybersecurity. The broader data-protection layer includes technologies such as encryption, multi-factor authentication, access controls and network security. Fraud detection addresses a different question: is an apparently legitimate action being carried out for a dishonest purpose?

NIST’s Cybersecurity Framework 2.0 recommends a connected approach to cybersecurity based on governance, identification, protection, detection, response and recovery. This model works particularly well for financial software because no single preventive control can address every type of risk.

At some point, activity will deviate from the norm. The important question is whether the system can recognise it.

Financial Software Should Be Tested Against Anomalous Behaviour

As security concerns increase, testing should go beyond verifying standard functional requirements. Unusual scenarios and behaviours should also be tested to determine how both the software and the team respond. Examples include repeated payments within a short period, sudden device changes or several account modifications made in rapid succession.

It is equally important to verify that alerts reach the correct people, that logs contain enough information to support an investigation and that additional authentication measures are triggered as intended.

Trying to predict every possible fraud scenario is unrealistic. The goal is instead to design a platform capable of recognising patterns of unusual behaviour that may not appear fraudulent individually but become suspicious when assessed together.

Financial Software Has to Understand Context

Reliable financial software must still perform every ordinary function correctly. Transfers need to reach the intended destination. Balances must remain accurate. Authentication has to work.

But functional correctness alone does not make a financial platform secure. Fraud often involves legitimate features being used in ways they were never intended to be used. Detecting this requires context: behaviour, devices, transaction history, relationships between accounts and changes over time.

Be the first to comment

Leave a Reply

Your email address will not be published.


*


This site uses Akismet to reduce spam. Learn how your comment data is processed.