The recent revolution in artificial intelligence (AI) promises to transform real-time payments fraud detection for central infrastructures (CIs) and payment networks. Network intelligence is one of the ways that the full benefit of AI can be realized.
Machine learning for payments fraud detection has limitations
As organizations look to drive adoption and increase transaction volume through their new real-time payment rails, fraud risks have been steadily climbing the agenda of data science and risk leaders within those organizations. Fraud isn’t more common with real-time payments, but increasing digital payment volumes and types create increasing fraud volumes and attack vectors. Thinner margins also mean the impact is felt more keenly on profitability and ROI, while in the case of important national utilities or economic infrastructures, it’s also vital to protect liquidity.
Machine learning — while it is one of the keys to mitigating these risks for CIs, payment networks and their members/customers — does have some practical limits. The knowledge reflected in machine learning models is largely restricted to the data found within the boundaries of a single organization. Privacy and compliance obligations make it too costly and too risky to horizontally expand their data sources any further, for example, with data from other financial institutions or from related third parties, such as telcos and eCommerce providers.
Recent developments in AI can help overcome these limits
Forrester describes federated machine learning as an AI 2.0 use case, which in the case of ACI’s own network intelligence capabilities facilitates the real-time sharing of information about payment fraud risks between organizations. It does so without compromising IP, data privacy or compliance obligations, effectively smashing through the barriers to horizontal data growth. By allowing all parties to supplement their proprietary data to improve the performance of their machine learning models, it transforms the way CIs and payment networks can protect and add value to their members and customers.
Network intelligence and maintaining data privacy compliance
Network intelligence connects members via a central body, creating a community that shares real-time fraud patterns in the shape of data features and signals.
A feature describes raw data without revealing it. It’s a data point made up of other pieces of information, which taken together — as a feature — communicates patterns in data sets and their corresponding predictability for fraud. And it does so in such a way that they can be used by other machine learning models on their own data sets. Features combine to create models, which organizations can then adopt – in full or as individually selected features – based on how strong a signal for fraud they are.
A great example of a signal or feature would be the number of transactions someone makes with a specific payment option within three hours. Most consumers are not in a rush to perform as many transactions as possible in a short time frame. When you compare the number of legitimate transactions to fraudulent ones, as the count of transactions within three hours increases, the probability that it is legitimate decreases rapidly. Velocity is thus a great indicator or “signal” for fraud in many cases.
How is network intelligence different to existing forums and consortiums for intelligence sharing?
In a traditional consortium set up, a central body uses member data to create a one-size-fits-all model that only really works best for the biggest players, because they get the heaviest weighting.
But the federated machine learning approach seen in network intelligence allows models and algorithms to be trained across multiple decentralized locations. Each location holds local data samples that don’t need to be exchanged with each other. This is privacy-by-design in action, ensuring that collaboration can occur without overstepping on data privacy, data security, data access rights and related compliance.
In practical terms, this allows the community’s central body to take features from members to create a consolidated community model that members can choose to adopt in full or — more often — only select the features that best fit their individual fraud strategy. The central body can also drive innovation by building features for members to test and provide their perspective of the features’ relative risk signals.
Enabling business outcomes for CIs and payment networks
Access to a vastly expanded community view of fraud and risk dramatically enhances the way CIs and payment networks can add value to their members and protect their users or customers. And the way they generate revenue.
Payment networks can differentiate on trust and reliability to grow confidence in and adoption of their real-time payment networks — while reducing the fraud losses that would undermine efforts to fill the gap left by declining card revenues. Offering network intelligence as a service also creates a monetizable, value-added enhancement to customers’ real-time payments fraud management strategies.
And from a CI perspective, it helps to better protect the financial ecosystem from liquidity risks arising from capital leaving the network, while offering better regulatory oversight to protect consumers. It also allows them to empower all their member financial institutions in the fight against fraud. Even smaller banks with limited resources can participate in building and leveraging shared community intelligence for managing risks in real time.
To learn more about network intelligence download our eBook AI 2.0 and the Revolution in Real-Time Payments Risk Management or visit aciworldwide.com/solutions/network-intelligence.