The financial services industry is under continued pressure to create more convenient ways of paying for goods and services. This trend goes beyond common hype: it’s a culture shift from long-standing payment methods traditionally defined by industry heavyweights to embracing market demands and delivering customer-centricity. With digitalisation, new market entrants and the creation of new channels such as PSD2, there is increased competition to acquire new market share, retain existing business and most importantly, to remain relevant. The volume of transactions is incidentally growing at an exponential rate worldwide, fuelled by digitalisation, new fintech offerings and on a larger scale, whole economies such as Sweden opting to operate cash-free.

Whilst change is certainly a constant in the financial services world, the pace at which it is happening is unprecedented. This is prompting many in the C-suite to re-evaluate their existing strategies and drive growth through innovation. This proactive stance towards market demand presents both an opportunity and a menace. New payment methods will deliver key improvements with regards to speed, mobility and ease of use but they will undeniably exert more pressure on fraud detection mechanisms. Unless updated and consolidated with new detection methodologies, these detection platforms are at risk of inviting fraud losses.

A macro view on fraud

Financial organisations have invested heavily in tools and resources to monitor their transactional flux and block fraud in real-time. Most financial institutions now claim to use analytics to drive the scoring and decisioning within their fraud solutions. The use of new data sources such as device information, geo-location or session data is now becoming standard. However, despite the deep level of scrutiny, the reality is that fraudsters are still able to circumvent many of these defences. Are these organisations therefore doing enough?

Fraud practitioners often claim that the ‘devil is in the detail’ and this is true in helping to uncover most fraudulent transactions. There are still many cases, typically low in volume but high in value, which remain undetected. These do not trigger alerts as the underlying evidence may not seem significant enough but when looked at through a different lens, they can uncover new modus operandi or large-scale organised fraud. Overlay techniques such as link analysis or social networks can help detect these new threats and uncover collusive behaviour by CPP/CPC (common point of purchase / common point of compromise): fraudulent merchants, fraud around points of sale or compromised ATMs infected by malware.

Operationalising AI for better fraud detection

The use of Artificial Intelligence & Robotic Process Automation (RPA) in the industry is still in the early stages but is maturing quickly. RPA can be applied for the automation of transactional rules-based tasks where structured data and predefined parameters are used. If correctly deployed and managed, it can be an effective means to manage operational costs and optimise human resources for more complex or uncommon cases. Many anti-fraud professionals are looking to leverage AI to extract complex patterns from their data to uncover a wide range of sophisticated fraud modus operandi. Techniques such as ensemble methods and deep learning are gradually making their way from offline data-labs into operations. As hardware infrastructure is becoming increasingly commoditized, AI is also being considered for dual negative and positive scoring: i.e. fraud detection as well as using the same data and the same transactional engine to upsell or promote new value-added services in real-time.

Contextual awareness

Many transactional fraud systems, although equipped with analytical engines, suffer from rising volumes of false positives. For example, a transaction made on 29th November, out of local time-zone hours for HKD 20,000  and for the purchase of high-value goods could be deemed high-risk and blocked. This scenario potentially breaches several rules around high amount and foreign spend as well as anomalies compared to common behaviour but is it effectively fraud? That transaction could be linked to a frequent traveller, used to spending in foreign currencies and where the spend value might only be a small fraction of their credit limit or their account balance. The purchase is also made on ‘Black Friday’, a popular date for many consumers to go on a shopping spree.

Most current systems are often tuned to look deep into the detail and extract relevant intelligence for fraud decisions but they often miss the bigger picture. An optimal fraud solution should be underpinned by multiple, potentially overlapping data segments and the score would consider several additional factors such as average disposable income, seasonal trends and average travel expenditure derived from account information or merchant category codes. Many market-leading solutions now feature customer signatures or entity profiles, with intrinsic machine learning and AI capabilities aimed at making better contextual decisions. The same approach can be used to defeat growing industry threats such as social engineering and account takeover. The quicker financial organisations adopt this bilateral strategy combining transaction focus with an overarching helicopter view, the more fraud attacks they will be able to stop in time.