Scalable AI is one of the most important features of a robust fintech platform. Sure, there’s been a lot of hype on the topic, but for growing organizations, scalability and speed affect the bottom line every day.
Before we talk about how scalable AI helps your business, let’s look at exactly what the term means. Scalability is the solution’s ability to grow and handle an increasing dataset, workload, or additional processes. The more scalable the AI solution, the more reliable it is in returning the results that determine the outcome of good or bad transactions.
Beyond providing the security your organization needs, scalability directly affects your customer service. If the system lags, transactions don’t go through. If the system can’t process in real time, fraudulent transactions may be approved manually. And if credit risk isn’t reliably identified, a delinquent customer may become a default. Scalability enables rapid growth and increased transaction speed.
Why scalable AI matters
When assessing technology before making an investment, scalable AI is an important consideration for all organizations. Whether preventing transaction fraud, assessing credit risk or detecting healthcare fraud, waste and abuse, your AI solution must be fast, efficient and reliable while capably handling growth. Scalable AI comprises these elements.
The problem with databases
Data analysis solutions based on databases or legacy rules-based technologies are not highly scalable. As datasets grow, these platforms slow down significantly and impact the time required for results to be determined.
Additionally, risk systems based only on rules detect anomalous behavior associated with just the existing rules; they cannot identify new anomalies that can occur daily. As a result, systems based on rules are outdated almost as soon as they are implemented.
A platform’s scalability boils down to how a solution handles increased throughput in terms of additional users, more computations, more input/output while still delivering the same performance or end user experience. And when any or all of these things increase, how does your solution expand to handle that additional volume?
Scalable AI architecture must be built for significant change from the ground up. A good AI platform allows expansion for the long term, not just a few years.
Scalable AI supports rapid growth
More data is being created today than ever before. The World Economic Forum estimates that by 2025, 463 exabytes of data will be created each day globally. That’s 463,0006 bytes, or the equivalent of 212,765,957 DVDs per day. With more data than ever before, what we need is intelligence to make sense of it all.
Scalable AI prepares organizations for sudden shifts. It needs to flex as datasets grow, regardless of whether you’ve added new customers, launched a line of new products, or experienced a sudden increase in transactions, such as the pandemic-influenced surge in online sales.
“Scalable AI is important because most companies want to grow, process more data, get more customers, or bring on new portfolios,” says Paul Cordero, Director, Product Sales at Brighterion. “Having a system that is truly scalable and won’t have a catastrophic failure when you add more data is key for any mission-critical system.”
Scalable AI also allows additional use cases to be added once an AI model is in production. For example, global acquirer Worldpay initially deployed Brighterion AI for anti-money laundering (AML) during the 2012 Olympics. One of the selling points of Brighterion was its scalability, and soon the credit risk and payments fraud portfolios were added to the solution.
Scalable AI delivers reliable, fast decision speed
We’ve established that scalability affects how AI handles the datasets, the volume of information processed, and the ability for the datasets to grow and have the solution grow with it.
As the volume of data grows, many platforms bog down and results take longer. A scalable AI solution performs under pressure and ensures reliable throughput.
Why does speed matter? In fraud or risk detection, decision speed is an important element in flagging anomalies and providing real-time scoring. If your system can’t process in real time, fraudulent transactions may be approved manually. And if credit risk isn’t reliably identified, a delinquent customer may become a default. Expansive scalability ensures no lag time.
Distributed architecture is engineered for scalability
Achieving highly scalable AI is at the heart of Brighterion’s engineering. Using distributed architecture that is carefully engineered to run on entry-level servers, Brighterion’s AI solution installs virtual servers within the solution, using less RAM. The workload is balanced among the virtual servers, meaning that there is no lag time or outages. Brighterion’s customers benefit from 99.9999% uptime.
As Cordero explains, “it’s not dependent on databases or large amounts of RAM. The scoring file system is very efficient and fast and doesn’t analyze all of the data; it uses the relevant data elements for the model to identify the potentially risky behavior.”
With distributed architecture, the solution takes the card number, cardholder name or other identifier for each transaction or decision. It instantly goes through the file system, finds and updates the account’s history with the current decisioning event, and scores it. Once the system determines if the event is valid or looks suspicious, the results are sent back out.
This decision is made in less than 10 milliseconds while completing more than 100,000 decisions per second. The lightning fast speed enables real-time decisioning every time.
“The process is very intricate and intense,” says Cordero. “With some solutions, the performance is hurt because they don’t have the real-time history to score the data against. That’s why we’re able to have such a high-performance rate with a low false positive rate — we are actually able to do all this in real-time.”
Highly scalable AI enables future-proofing
According to Wikipedia, “the term ‘future-proof’ refers to the ability of something to continue to be of value into the distant future – that the item does not become obsolete.”
Highly scalable AI is one way of future-proofing your business. As we’ve seen in the Worldpay example, adding use cases was done with ease. And as Worldpay’s business grows, they are able to add their own servers, if necessary.
When Mastercard started using Brighterion more than 20 years ago, our AI was deployed only for card present transactions. Over time, more portfolios were added and Mastercard now uses Brighterion AI for its full dataset, including card-not-present transactions and issuer fraud.
Recognized for scalable AI
Aite Group, the leading analysts of financial services technologies, named Brighterion the most scalable AI platform used for anti-money laundering (AML). Aite found our platform enables more than twice that of our nearest competitor, citing “62,000 transactions per second (TPS) speed. Its streaming infrastructure with no underlying databases is a key driver of this impressive performance,” Aite reported. We have since upgraded our speed to 100,000 TPS.
Download the AITE Group report AIM Evaluation: Fraud and AML Machine Learning Platform Vendors to learn more.
