A customer receives a call from what sounds exactly like their bank manager. The voice is calm, familiar, and urgent. There has been suspicious activity on the account. Immediate verification is required.
The customer complies.
The voice was synthetic.
Not long ago, scams like these required time, coordination, and a degree of technical skill that limited how often they could happen. Generative AI has changed that equation. Fraud is no longer only becoming more sophisticated. It is becoming industrialised.
Banks are now facing a different type of fraud environment, one where synthetic identities, deepfake audio, AI-generated phishing campaigns, and automated scam networks can be created faster, scaled wider, and personalised more convincingly than traditional fraud operations were designed to handle.
According to Deloitte, generative AI could significantly accelerate fraud losses in the United States, potentially increasing them from approximately US$12 billion in 2023 to US$40 billion by 2027 as fraudsters gain access to more advanced and scalable tools. At the same time, McKinsey reports that generative AI adoption across enterprises nearly doubled in under a year, reflecting how quickly AI capabilities are entering mainstream use across industries.
Fraudsters are moving just as quickly.
Fraudsters are using AI to scale deception
Traditional fraud often depended on volume. Attackers sent millions of poorly written phishing emails hoping a small percentage of victims would respond. As mentioned in the BPC Guide Anatomy of New Fraudster - AI is shifting fraud towards precision.
Large language models can now generate convincing emails, messages, and social engineering scripts in multiple languages within seconds. Deepfake technology can replicate voices and video likenesses convincingly enough to bypass basic trust signals. Scam campaigns that once required organised teams can now be automated and personalised at scale.
The impact is already visible across financial services. Deloitte notes that generative AI is making fake content easier to create and harder to detect, particularly in banking environments where trust and urgency are central to customer interactions. Reuters also reported that AI-assisted crypto scams helped push global scam revenues to record levels in 2024, with “pig butchering” scams and synthetic social engineering tactics growing rapidly.
Even traditional scam indicators are becoming less reliable.
Grammar mistakes, robotic voices, and generic messaging once acted as warning signs for customers. AI-generated scams remove many of those signals. Fraud attempts can now mimic natural conversation patterns, local dialects, executive communication styles, and customer service workflows with alarming accuracy.
Europol has warned that AI is becoming an “accelerator” for organised cybercrime, allowing fraud networks to automate scams, personalise social engineering tactics, and deploy deepfake impersonation attacks at scale across Europe.
For banks, the challenge is no longer simply identifying suspicious transactions. It is identifying synthetic trust.
Why legacy fraud management systems struggle with AI Fraudsters
Many fraud management environments inside financial institutions were built for a very different fraud landscape.
Rules-based systems remain effective for known patterns. They can identify repeated behaviours, suspicious transaction thresholds, or previously flagged accounts. But AI-driven fraud does not always behave predictably enough for static logic to keep pace.
Attack patterns now evolve dynamically. Fraudsters can test variations of scams in real time, adapt messaging based on customer responses, and alter behavioural signals faster than manual rule updates can respond. What worked as a fraud indicator last month may no longer be reliable today.
This creates pressure on legacy fraud operations in several areas. First, fraud monitoring teams are dealing with growing alert volumes and increasing false positives. Traditional systems often rely on broad detection thresholds, which can create operational fatigue and negatively affect customer experience.
McKinsey notes that many payment providers still rely on “one-size-fits-all” fraud approaches that generate high levels of false positives instead of deploying more adaptive and specialised detection models.
Second, legacy environments are frequently fragmented across channels. Card fraud monitoring, digital banking fraud, account takeover detection, merchant fraud, and AML systems may operate separately, limiting the institution’s ability to build a complete behavioural picture across the customer journey.
That fragmentation becomes a larger problem when AI-powered scams move fluidly between channels. A fraud attempt may begin with a phishing message, continue through a spoofed phone call, and end in a real-time payment transaction within minutes.
Third, many traditional systems depend heavily on historical fraud patterns. AI-driven fraud introduces behaviours that may not closely resemble previous attack models at all. Static systems struggle when fraud itself becomes adaptive.
Why AI-powered fraud management platform is the best choice for the bank
AI-powered fraud management platforms are designed to operate more dynamically across large volumes of behavioural, transactional, and contextual data.
Instead of relying only on predefined rules, modern platforms combine machine learning, behavioural profiling, real-time analytics, link analysis, and adaptive risk scoring to identify anomalies as they emerge.
A transaction may appear legitimate in isolation. However, when combined with device behaviour, geolocation inconsistencies, customer interaction history, account activity patterns, and network relationships, the risk profile can change significantly. This is where AI-driven fraud management becomes particularly valuable.
Modern platforms can continuously analyse evolving behaviour across customers, merchants, devices, accounts, and channels simultaneously. They can also detect subtle deviations that would be difficult to identify manually or through static rules alone.
AI-powered fraud management is not about replacing human investigators. Fraud teams still play a critical role in validating threats, managing investigations, refining models, and understanding emerging attack tactics. The difference is that AI-powered systems can process and correlate signals at a scale that human analysts alone cannot realistically manage.
Where banks can gain momentum
For many institutions, the conversation around AI in fraud management is no longer theoretical. The challenge now is operational readiness.
Banks that are gaining traction tend to focus on several practical priorities. The first is building a more unified fraud view across channels and payment types. Fraud rarely stays confined to a single interaction point anymore. Connecting fraud intelligence across cards, accounts, digital banking, acquiring, and real-time payments helps institutions identify broader attack patterns earlier.
The second is reducing dependency on static rules alone. Rules still matter. However, they increasingly need to work alongside adaptive machine learning models, behavioural analytics, and real-time risk orchestration rather than operating independently.
The third is improving investigation efficiency. As alert volumes continue to rise, fraud teams need better prioritisation, clearer visualisation of relationships between entities, and faster access to contextual intelligence. AI can help reduce operational pressure by improving how alerts are scored, grouped, and escalated.
This is also where platforms such as SmartVista Fraud Management (SVFM) are evolving to support modern fraud operations. SmartVista Fraud Management combines real-time fraud monitoring, behavioural profiling, machine learning-based scoring, link analysis, and adaptive rule management across issuing, acquiring, digital banking, and payment environments.
The conversation around modern fraud management is no longer theoretical. Across different markets, banks are increasingly investing in technologies that provide greater visibility across channels, stronger behavioural analytics, and more adaptive detection capabilities.
In Saudi Arabia, QNB implemented SmartVista Enterprise Fraud Management to strengthen fraud prevention and detection across its digital channels. The bank sought to improve visibility across payment and non-financial activities while supporting real-time monitoring, automated controls, and customer behaviour profiling as digital transaction volumes continued to grow. Similarly, Malaysia's Co-opbank Pertama (CBP) deployed SmartVista Fraud Management to protect its retail and corporate digital banking services, introducing real-time monitoring, machine learning-driven analytics, and a consolidated view of fraud activity across multiple customer touchpoints.
The momentum extends beyond fraud monitoring alone. In the Dominican Republic, Banco Finandina adopted BPC's next-generation SmartVista platform and 3DS 2.0 technology to strengthen authentication for online payments and reduce exposure to identity fraud. While each institution addressed different priorities, the common objective was the same: building fraud and risk management capabilities that can adapt more quickly as digital transactions grow and attack methods become more sophisticated.
As fraud becomes more AI-enabled, fraud management itself can no longer remain static. The institutions that adapt fastest are unlikely to be the ones with the largest rule libraries. They will be the ones capable of learning, correlating, and responding faster than the fraud networks targeting them.
Because in the AI era, fraud is no longer only about stolen credentials or suspicious transactions. It is increasingly about manufactured trust.
