By Amritha Reddy, Senior Director of Fraud Product Management, TransUnion Africa
For years, organisations treated fraud as a transaction problem: identify a suspicious payment, compromised account or fraudulent application and stop it before money leaves the system.
That model is no longer enough. Fraud increasingly begins long before the final transaction, through compromised credentials, synthetic identities, social engineering and AI-enabled scams. By the time suspicious activity appears, the attack may already be well advanced.
Fraud has evolved from a transaction problem into an identity and trust problem.
The Battleground Has Moved Upstream
Fraud rarely begins with the transaction itself. More often, it develops over time, with each step creating the conditions for the next.
A fraudster might change contact details, steal login credentials, open a mule account or build a convincing synthetic profile. The financial loss comes later, often after the groundwork has already been laid.
TransUnion’s H1 2026 Fraud Trends research reflects this change. In 2025, 3% of digital transactions involving South African consumers were suspected of fraud. Of the South African consumers who said they had lost money to digital fraud, 33% pointed to scams by third-party sellers on legitimate e-commerce sites. Globally, the suspected digital fraud rate for account takeover rose by 37% between 2024 and 2025. The pattern is clear: criminals are increasingly exploiting identities, accounts and trusted platforms, not just individual transactions.
That is why controls focused only on the transaction can miss what happened before it. By then, the organisation may already be responding to the final act rather than the wider scheme.
Organisations need to follow risk across the customer journey, from onboarding and login to profile changes, account activity and payments. Timing matters at every stage: criminals adapt quickly, warning signs need to be recognised early, and teams must be able to act while there is still time to prevent a loss. A delay at any point can make even strong detection less effective.
AI Is Changing the Rules, but Not the Objective
AI is changing the fraud landscape, but it should not distract us from the real objective. Fraudsters are already using it to create more convincing scams, build synthetic identities and launch attacks at scale. The answer, however, is not simply to keep building more sophisticated models.
A model may spot the risk, but that does not mean the organisation can act in time. Even an accurate system can create unnecessary friction in a legitimate customer’s journey if it triggers the incorrect response. What matters is the quality of the decision: spotting risk early, responding in the right way, acting before the opportunity is lost and learning from what happens next.
No Single Solution Sees the Whole Picture
There is no single answer to fraud. Rules still play an important role in managing known risks and policy requirements, while machine learning can pick up less obvious patterns and adapt as those patterns change.
Device and behavioural intelligence add context by showing whether an interaction looks right for that customer. Network analysis adds another view, revealing links between identities, devices, accounts and transactions that may not be visible on their own.
Each layer adds useful context, but none is enough on its own.
The strongest fraud programmes bring these signals together to make a clearer decision in the moment. This helps organisations respond to genuine risks without adding unnecessary checks or delays for trusted customers. The real test is not whether one tool works well in isolation. It is whether the combined intelligence helps the organisation spot risk earlier, act with confidence and keep the experience smooth for genuine customers.
Look Beyond the Credentials
Authentication remains essential, but credentials only tell part of the story. A password may confirm what a customer knows; it does not show whether the interaction fits how they normally behave. A new device, a change in contact details, an unfamiliar beneficiary or an unusual payment may seem harmless on its own.
Seen together, these signals can point to a customer at risk. Recognising that pattern early gives organisations more opportunities to protect the customer before money moves, without putting unnecessary hurdles in the way of genuine activity.
Technology can help, but it cannot work in isolation. Fraud, data, technology, operations, customer experience and compliance teams need to work together so that useful intelligence leads to a timely, consistent response for the customer.
This matters even more as organisations introduce AI. Its most immediate value may be in focused uses, such as bringing case evidence together, finding relevant information quickly and helping investigators make informed decisions. Human judgement, clear governance and accountability must remain at the centre.
Measure What Matters to Customers and the Business
A fraud programme should be judged by the difference it makes, not only by how well the model performs.
Better analytical accuracy does not automatically mean lower losses or a better experience for customers. Organisations need to know whether they are spotting risk earlier, preventing harm, keeping genuine customer journeys simple and helping investigators focus on the cases that matter most. Regular feedback is also essential to identify when models are becoming less effective as fraud tactics and customer behaviour change.
Across Africa, this is made harder by fragmented identity data, mobile-first channels, cross-border activity and different levels of digital maturity. Organisations are also working within real limits in data, skills and operations. The answer is not simply more technology. It is intelligence designed for local conditions and connected to decisions that teams can act on while keeping the customer experience in view.
The Real Advantage Is Acting Before Trust Is Lost
Fraudsters are moving faster, working across connected channels and using AI to scale their attacks. Organisations will not stay ahead by chasing the most sophisticated model. They will do it by recognising risk earlier, bringing the right signals together and acting before a customer is harmed.
The strongest fraud programmes will do more than detect fraud. They will protect customers sooner, reduce friction in genuine journeys and learn from every outcome. The real win is not stopping fraud after it happens, it is protecting trust before it is broken.
