Payments glossary
Fraud scoring
Fraud scoring rates the risk of each transaction in real time, combining signals such as device fingerprint, location, purchase pattern and past behaviour into a single score. The score drives an automatic decision: approve, step up to extra authentication, or decline.
How a fraud score is built
Every transaction arrives wrapped in signals: the device and browser fingerprint, the IP location versus the delivery address, the age of the email address, how the order was assembled, velocity counts across cards and devices, and the history of this customer and this card. Rules catch the known patterns - a mismatched country pair, a flagged device - while machine-learning models weigh the subtler combinations. The output is one number: how much this transaction looks like fraud.
What happens at each score
Low scores pass straight to authorisation and the customer never notices. Middle scores are stepped up rather than rejected - typically a 3DS2 challenge, so the issuing bank confirms it is really the cardholder before the payment proceeds. High scores are declined or held for review. The craft is in the thresholds: set them too tight and genuine customers are blocked, which costs sales invisibly; set them too loose and the goods ship to fraudsters, which costs chargebacks visibly. Good scoring is tuned per merchant, because a pattern that is suspicious for a bookshop is normal for an airline.
Fraud scoring on LekkaPay
LekkaPay's risk engine scores every transaction in real time and can route borderline payments into a 3DS2 challenge instead of declining them outright - keeping the sale while keeping the protection.
Last updated:
Payments without the jargon.
One integration, every way Africa pays - we'll handle the acronyms.