“Data is the new oil” is a phrase attributed to Clive Humby, a U.K. mathematician and the architect of Tesco Clubcard, one of the most successful brand-loyalty schemes ever created. The phrase has been more recently employed in reference to the huge datasets and associated valuations of big tech companies such as Facebook, Google, Amazon, Tesla, and Microsoft, including an Economist front page proclaiming data as “The World’s Most Valuable Resource.”


The critical importance of data is not missed in financial services. Former Citibank CEO and fintech pioneer Walter Wriston made a similar comment when he suggested that information about money is as important as the money itself.


These phrases on the importance of data caught the public imagination when big data and associated technologies such as Hadoop were at the peak of their hype curve. However, the same emphasis on data applies in the emerging world of deep learning and artificial intelligence (AI); good quality, real-world data is essential to successfully training deep learning systems in financial services and helps to reduce false inferences.


Financial services companies see interest in deep learning and AI in several areas. Many use cases are focused on sales, marketing, and cost reduction, such as lowering customer sign-up friction through chatbots for initial engagement and subsequently with solutions for automated ID verification and credit risk assessment. Emphasis is also on the use of deep learning to detect fraud, particularly in insurance, and almost all organizations could use it to meet and reduce the cost of know-your-customer (KYC) and anti–money-laundering (AML) requirements. There is wider use in systems to accelerate the review of legal and contract documentation, particularly in response to legal disclosure requests.


Tools have been developed to aid market forecasting and social media analysis to shape trading strategy. In most of these use cases, AI and deep learning are still being used in supportive roles, with audit and compliance requirements still maintaining that key financial decisions be taken by humans and not completely outsourced to machines.


There have been some exceptions to this approach, with people putting greater trust in machine-based systems. A good example in the United States is Betterment, which provides customers with an online robo-advisor service to help support investment decisions. In reality, this is not too far removed from the use of deep learning by hedge funds to support decision making, but it is allowing an automated service delivery to a customer.


More dramatically, there was the now-infamous 2016 crowd-funded attempt using Ethereum smart contracts to build an organization called the DAO (Decentralized Autonomous Organization). The DAO offered a fully decentralized business model without a conventional management or board, with the aim of managing investment for commercial and nonprofit customers. The DAO was hacked shortly after its launch because of a weakness in its open-source code, and a third of the funds were stolen. Although deep learning and AI were not implemented in the blockchain itself, they were to have connected into it through trustless smart contracts. The DAO failed because of security, but it is easy to envisage financial disasters that might also occur in a system with more experimental robotic traders and a lack of the traditional skill switch of stock exchanges. (Note: The DAO hack resulted in a blockchain fork to allow restoration of the missing funds.)
The application of deep learning and AI techniques to financial services certainly has value; its abilities to reduce cost, enhance the customer experience, reduce fraud, and enhance investment decisions are all key. However, it is recognized that financial ecosystems are highly complex, bringing together many aspects such as financial advice and decisions, customer data, contracts, legal, security, and the handling of financial assets. When looking to develop and ultimately deploy systems using deep learning and AI, financial institutions need to be in a position to accurately assess whether a solution will achieve its objectives and meet all the necessary security and regulatory requirements.


Such decision making can be guided by data scientists and machine-learning experts, but to reach accurate conclusions, organizations will require good quality, real-world data. This data is also a prerequisite for achieving high-quality outputs from machine learning and avoiding bias and false conclusions. So, for financial services organizations that want to embark on machine-learning projects, it is essential that they put data quality, collection, and management at the heart of their operations.

Vasilis Kapsalis