2019 was not a good year for fraud detection. The Cifas Fraudscape report showed that fraud in the UK actually rose by 6% in 2019. Accompanied by the fact that 9 of 10 flagged transactions that are suspected fraud are later cleared, this rocky landscape can ruin relationships with consumers and waste company time and money. In an industry that infamously has poor user experience, incumbent financial institutions must adapt before losing their user base to newer, more technologically competent challenger banks.
The digitization of the financial system has made life much easier for the consumer but unfortunately these same developments have made life easier for fraudsters. With an ever-increasing use of the internet for shopping, banking, filing insurance claims etc and the growing number of channels used by consumers to access the banking system, fraud has evolved in whole new, unpredictable ways. For example, contactless cards have made our lives more efficient by reducing the time it takes to make physical transactions but in the wrong hands, allow fraudsters to take money without any authorisation (up to your contactless cap or until the card is cancelled).
Fraud has become a major problem in finance and a lot of resources are being invested to recognize and prevent it. According to UK Finance, £1.2 billion was successfully stolen by fraudsters in the UK alone in 2018. But it’s not only fraudsters upping their game – the regulators are also imposing more and larger fines. For example, the Bank of Scotland was fined £45.5m by the FCA for not disclosing information about suspicions that internal fraud may have occurred in 2019.
It’s no wonder then that financial institutions are turning to RegTechs to help solve their fraud and compliance woes, further fuelling the RegTech revolution and leading to the paradigm shift alluded to in the title – from rules-based systems into the new generation of tech – machine learning.
RegTech to the Rescue
Numerous RegTech companies offer software-as-a-service (SaaS) point or platform solutions, helping to fight financial crime more generally and detect fraud more specifically. Products target all types of fraud including APP fraud, credit card fraud, payment fraud, internal fraud and corporate fraud. One thing all the top vendors employ however is machine learning (ML).
Evolution of fraud detection technology
Traditional approaches to identifying fraud have been rule-based. This means that hard and fast rules for flagging a transaction as fraudulent have to be established manually and in advance. But this system isn’t flexible and inevitably results in an arms race between the seller’s fraud detection system and criminals finding ways to circumnavigate these rules.
So with money and reputations on the line, the solutions had to develop to keep up with the rapidly developing tech and fraudsters in an ever-changing economic environment.
Enter machine learning.
Fraud detection has to be done in real time in order to stop the illicit trail of money. This is on the contrary to other forms of financial crime like money-laundering where suspicious activity is usually flagged after the fact. By leveraging the processing power of machines, large scale data such as transactions can be processed with sub 10 milliseconds of latency, seen as ‘real time’ and stopping crimes before they can occur.
Across the fraud prevention space, all leading vendors use ML in some way or another. New kids on the block such a Bleckwen and DataVisor, challengers such as Feedzai and Featurespace and even incumbents such as SAS, FICO and fiserv are starting to adopt the technology. A combination of the old-school, incumbent rules systems with supervised (and sometimes – but less frequently – unsupervised) machine learning is deployed. This has seen a huge reduction in false positives compared to the solely rules-based systems; Featurespace claims a 75% reduction whilst NetGuardians claim an 83% reduction.
As with all innovative technologies, there are risks as well as benefits. For ML learning products, one of the key risks to manage is the opacity of ML models. To satisfy regulatory expectations, the algorithmic reasoning behind each case of fraud needs to be explainable in order to create an audit trail. This is perhaps the biggest stepping stone AI/ML vendors face today and, understandably, given the recent harsher regulations, is an important point for the businesses using RegTech. In addition, being able to explain how the decisions were made to block specific payments is also key to the customer experience, a key competitive advantage in digital banking.
The fraud detection space is evolving rapidly so keeping up with the key trends is important for the tech vendors. Four key areas to look our for in 2020 are:
- Platforms addressing multiple financial crime use cases. Leading fraud detection vendors such as Feedzai and Featurespace are moving toward fully comprehensive financial crime platforms that cover all bases and as each use-case can employ similar underlying technology.
- Point solutions addressing very specific types of fraud or fraud in particular sectors. For example, Bleckwen’s ML solution focuses on payments fraud in financial services, and particularly APP fraud. Chainalysis targets fraud in the cryptocurrency market whereas Kount focuses on eCommerce across a range of different industries.
- Ability to explain results of ML Models. With the introduction of ML, model explainability has become a priority to vendors in order to abide by existing and new legislation and regulation such as the Future of AI Act and SR 11-7. Aside from this, understanding how your own systems work will help improve model accuracy as well as prevent bias.
- Synthetic data for collaboration. Vendors commonly claim KPIs such as reductions in false positives need a baseline data set to make them meaningful. An 80% reduction in false positives when the incumbent system was 30 years old is very different to an 80% reduction when the incumbent system is 3 years old. With help from the new wave of synthetic data companies such as hazy, Statice, Synthesized and Mostly.ai, a common transaction dataset could be used between the vendors to increase transparency in the sector and motivate competition between vendors’ models.
In order to move forward and reduce crime as much as possible, collaboration between the regulators, regulated institutions and vendors is crucial to be able to understand the regulations required and how to reach said requirements. With this collaboration, the industry can stay ahead of the fraudsters, keep the regulators at ease and reduce the rate of fraud so that as an industry, in future years, we can reflect on the statistics with pride.
We have conducted a market assessment of all the Fraud Detection products in the RegTech Directory which maps the vendor landscape and highlights the capabilities of the vendors in this market segment. Click here to download the report