Proprietary fraud tools will be rendered obsolete by 2023 due to continued growth in transaction volumes and out-dated software architecture, according to managed service provider Instaclustr. With debit card transactions alone predicted to reach 20bn a year by 2023, proprietary tools will soon no longer be viable for banks and payment companies, either due to the prohibitive cost or lack of scalability. Instaclustr forecasts that all major financial institutions will need to transition to open source solutions or risk facing unacceptable fraud losses each year as a result.
Fraudsters and cyber-criminals attempt nearly £3bn worth of financial scams annually, making off with more than £1.2bn in illicit gains. With 98 per cent of cases resulting in customers being fully refunded, this cost is overwhelmingly falling on banks and payment providers. Ensuring that the right digital tools are in place to monitor and prevent anomalous – and potentially fraudulent – transactions in real-time is therefore a business necessity.
“The problem facing banks is that effective anomaly detection systems, used to uncover fraudsters, involve a highly advanced combination of technologies,” commented Paul Brebner, Chief Technology Evangelist at Instaclustr. “Everything from machine learning and statistical analysis, to algorithmic optimisation techniques and data layer technologies are needed to ingest, process, analyse, disseminate and store streaming data in real-time. Running such applications at a massive scale means analysing millions of events every day, which is a huge computational challenge – one that is beyond the capabilities of proprietary systems, especially those operating a per transactions fee model. The only way to have successful, cost-effective fraud monitoring is through open source tools.”
The potential of open source to handle such volume was demonstrated in September 2019, when researchers at Instaclustr demonstrated that, using Apache Cassandra and Apache Kafka, it was possible to build a system capable of monitoring and evaluating 19bn events in a single day. Thanks to the scalable structure of Cassandra and Kafka, nodes could be added and subtracted as needed to meet any transaction volume, with no theoretical upper limit.
“The truth is the proprietary model is dying,” continued Brebner. “Partly this is because companies are far savvier about the need to avoid vendor lock-in these days, but mainly it’s because most of the systems on the market simply weren’t designed for the volume of data modern enterprises need to process. Banks and payment providers need to shift towards open source as quickly as possible to manage their fraud detection programmes, otherwise they’re going to continue to suffer losses when they’re unable to identify all the problematic payments and scams that are happening. And while it can seem daunting to transition to a whole new system, it’s very manageable with the right expertise on board – and besides, the alternatives are far less palatable.”