How real-time data technology drives AI in financial services
Thu 18 Apr 2019 | Roshan Kumar
Roshan Kumar, senior product manager at Redis Labs, discusses the most pressing challenges for AI in financial services
About 50 years ago, Barclays opened its first automatic teller machine (ATM) and changed the way we did banking. To this day, the ATM has been one of the most disruptive innovations of the financial sector, which forever seeks to stay on the leading edge of innovation. These days, you can deposit a cheque just by taking a picture of it using your mobile phone. Behind the scenes, AI deciphers the numbers on your cheque and deposits the appropriate amount into your account.
Of course, AI does a lot more than that. Financial services firms use AI-driven solutions for fraud detection, risk assessment, asset management, investment portfolio optimisation, stock predictions based on social trends, customer engagement, and much more.
Three emerging challenges for AI in financial services
As financial services organisations adopt the latest and greatest AI-based solutions to support their evolving business needs, they encounter new problems they didn’t face before.
In today’s on-demand, always-on mobile era, people expect instant responses from every app or website they use. Thus organisations encounter the problem of instant response at scale.
A delayed response may result in losing a customer, missing a stock trade or failing to flag fraud. To meet this expectation, AI solutions must run thousands, if not millions, of decisions in a split second, and do so cost-effectively at scale.
In-memory databases must be adopted as a key component powering the AI engine.
Organisations are also pressed to improve the quality of learning over time. What distinguishes AI from other computational solutions is its ability to learn and adapt to new situations. This process of learning must result in more accurate and applicable responses as time passes. However, quality improves only if the AI solution is able to capture and process all of the data points available and recalibrate its decision models effectively.
Companies also require autonomous AI solutions. Network connectivity is usually taken for granted, leading to centralised solution designs. However, the ATMs and AI solutions of the future must work even when they are disconnected from each other. Financial services firms are working hard to push their solutions to the network edge, where autonomous AI solutions can learn and make decisions on their own.
How financial services can drive their real-time AI needs using high-speed databases
Today’s most effective database platforms are constantly evolving to deliver instant responses for the applications they support. In-memory databases, in particular, are enabling hackers, innovators, developers and architects to deliver solutions for the instant economy. Some of them can run over a million operations per second with sub-millisecond latency using limited computational resources. Financial services firms use these database platforms to deliver real-time personalised user experiences, job and queue management, high-speed transactions, user activity tracking, product recommendations and more. For instance, companies have embraced open source projects, like Redis, to power specific AI needs.
These solutions allow them to meet the challenge of instant response for decision models. AI solutions react to situations they encounter by collecting data points, processing them and running them through decision models.
For example, an option trading company might use AI to compute the optimal price for buying and selling options in real-time. A bank or a credit card company could run every financial transaction through an AI model to detect and mitigate fraud. In-memory databases and computation engines can act as a ‘serving layer’ to run pre-trained neural network or tree ensemble models at unprecedented speed.
As financial services organisations adopt the latest and greatest AI-based solutions to support their evolving business needs, they encounter new problems they didn’t face before.
High-speed in-memory databases also accelerate faster model training. AI-based solutions in financial services, such as fraud mitigation, compliance and governance validations and price optimisation, build their decision models using supervised learning. During this process, data scientists put a lot of effort into calibrating and re-calibrating their AI models. However, as AI models only understand numbers, all text data must be mapped to numerical data during this process.
The faster applications convert it, the faster it trains the AI solution. By using a high-speed, low-latency in-memory database as a dictionary server, data scientists can perform over a million conversions in less than a second. This enables them to retrain their models and deliver better results many times faster.
Finally, the challenge of capturing high volumes of data in real-time data becomes can be met. The financial sector collects more real-time data than most other industries, including information about stock trades, commodity prices, interest rates and currency transactions.
A stream database can help manage these volumes by ensuring all the data points are fully captured and processed. For example, a sophisticated stream data structure can connect a multitude of producers and consumers. If the database is binary safe, it can also capture streaming audio, video and pictures. This is critical for a portfolio management company that uses its database platform to track user activity around research interests in order to make personalized investment recommendations.
Given the speed at which innovation is occurring in the financial sector, it won’t be too long before we see an autonomous vehicle (that also happens to be an ATM) stop at your door, recognise your face, lend you the money you want, and also recommends how to rebalance your investment portfolio. However, in order for this to happen, in-memory databases must be adopted as a key component powering the AI engine.
