AI in Financial Services: Challenges and How to Overcome Them

Artificial intelligence (AI) is disrupting critical business processes in virtually every industry, and finance consistently reports the highest levels of AI maturity out of them all. By leveraging new technology like AI, the financial services sector has made banking applications and products more user friendly and kept legacy institutions technologically relevant. However, this head start doesn’t mean there aren’t issues.

This article covers the challenges facing artificial intelligence in the financial services sector, including an overview of AI adoption, common obstacles and how to overcome them, as well as useful resources for taking advantage of artificial intelligence.

Adoption of AI in financial services

AI has been gaining prominence in the financial world since the 1980s, when expert systems were first used to predict market trends, provide customized plans, and reduce the risk of human mistakes. People working in the finance industry have been quick to recognize the potential of AI, and being early adopters has paid off. For example, hedge funds that leverage AI vastly outperform those that don’t. Research from consulting firm Cerulli found that “AI-led hedge funds produced cumulative returns of 34% [over the past three years]... compared with a 12% gain for the global hedge fund industry over the same period.”

According to a 2020 World Economic Forum survey, 85% of financial institution executives are already using artificial intelligence, and 77% expect AI to become essential to their business in 2022. Separate research from O’Reilly Media found that people in the financial services sector have the highest levels of AI maturity when compared with other industries. Although AI adoption in financial services is far along in a relative sense, it still has a long way to go. Next we’ll cover some of the obstacles facing AI adoption in financial services.

Challenges facing AI adoption

Early adoption means the financial services industry already has a well defined list of industry-specific AI roadblocks. Data privacy and security, data silos, access to high-quality training data, satisfying regulatory requirements, and skills gaps are some of the most essential challenges for organizations to anticipate and prepare for in order to ensure forward momentum isn’t stalled due to unforeseen blockers.

Data privacy and security

Financial services providers must collect, process, and store huge amounts of sensitive data that requires robust security and protection protocols. Additionally, increasingly strict data privacy laws make it prudent for financial services organizations to be aware of existing and forthcoming regulations. When building AI solutions, financial services providers need to take the following into account:

• Fast and secure infrastructure: AI applications rely on massive volumes of data that must be securely stored according to industry standards, and highly accessible to ensure processing speed doesn’t render a solution ineffective. Ensuring a strong digital backbone and appropriate infrastructure that includes a highly secure, low latency connection to move data from where it is processed and back again is essential to succeeding with AI. For more information on super.AI’s approach to data protection, check out our Data Security Whitepaper.
• Understand and follow relevant data privacy laws: The data that financial services providers rely on to build AI applications is typically personal. This is because it includes data generated from user activity such as shopping for new clothes, planning a vacation, or making investment decisions. New regulations such as GDPR and Payment Services Directive (PSD2) outline specific policies for handling this type of data, so it is important for financial services companies to be transparent with customer data and capable of explaining how they store and handle it. To learn more about the benefits of AI-powered personal identifiable information (PII) redaction, read our blog.

Data silos

Siloed pools of data are a massive blocker for artificial intelligence. Either due to regulations, company culture, or technology, businesses often find themselves with siloed units that can’t (or don’t want) to be brought together. Unfortunately, AI doesn’t like this.

As mentioned above, AI relies on massive datasets that must be readily available for processing and analysis. Additionally, because financial services organizations collect and generate huge amounts of data each day, customer data is often spread across a number of different systems with varying degrees of compatibility. This complexity, combined with an unclear or altogether missing data governance program, presents a major challenge to AI. Fortunately, there are solutions:

Tech-driven solutions

A surefire way to solve technology problems is to use more technology. Here are a few tech-driven solutions to overcoming data silos:

• Presto is an open-source, distributed SQL query engine built for big data. Its architecture enables users to query a variety of data sources such as Hadoop, AWS, MongoDB, and more. Data from multiple sources can be searched using a single query, making it possible to query data where it lives and unlock organization-wide analytics. Novel solutions like Presto offer a glimpse into the innovative approaches to overcoming data silos with technical solutions.
• By aggregating data into a cloud-based warehouse or data lake it is possible for companies to organize all their data in an accessible format. However, due to the type and size of the data stored by financial services companies, this may be a time and labor intensive undertaking. That said, this is not a novel concept–U.S. fund firm Alliance Bernstein embraced cloud computing as early as 2010–and most financial services organizations already heavily rely on the cloud for data storage and processing. Every business is unique and at different levels of technological sophistication, so the specific requirements for building a cloud-based data lake to fuel AI application development will vary accordingly.

Process-driven solutions

‍In order for financial services providers to build effective AI solutions, a clear strategy must be developed to ensure data is accessible to everyone that needs it. Process-driven solutions to data silos may involve technology, but are focused on democratizing access to data across organizations. This includes:

• Defining a data management strategy: Drafting plans and policies designed to safeguard the integrity and accessibility of data across an entire organization is no small undertaking. At least one person needs to take ownership of data management, ensuring that both silos don’t impede progress and sensitive information isn’t shared unnecessarily or unintentionally.
• Formalizing access controls: As we’ve established, financial services providers collect and store highly sensitive information. The pursuit of more accessible data should be weighed against the potential risks of improved accessibility. To make sure people cannot access data that falls outside their purview, companies should formalize their data access control policies, models, and mechanisms.
• Encouraging data literacy: Although only some (perhaps small) portion of a company’s employees will be directly contributing to AI application development, it is critical that everyone understands how to read, write, and communicate data in context. This is about understanding where data comes from, how it is structured, methods for analysis, use cases, and more.

Access to high-quality training data

It can be difficult for companies to source high-quality data, especially unstructured data. Of course, large technology companies like Apple, Amazon, Facebook, and Google collect huge amounts of data every second of every day. But smaller companies aren’t so fortunate. A former commissioner at the Federal Trade Commission (FTC), Rohit Chopra, went so far as to say, “Vast troves of consumer data collected by big technology companies allow them to gain a competitive edge and pose a threat to competition by creating entry barriers.”

Some pundits have argued for a “progressive data sharing mandate” that would require organizations of a certain size to share anonymized data with smaller rivals. Beyond this, even if companies can get good unstructured data, structuring it so that it can be used effectively for AI-powered automation is its own distinct challenge. Both of these issues have solutions:

• ‍Open-source datasets: There are thousands of publicly available datasets that can be used for AI/ML projects. For a jumping off point, check out this post on Medium that catalogues a number of data portals and aggregators. Additionally, larger financial services providers may have data they can use for AI application development, but remain unable to access it due to siloed data. See the process-driven solutions outlined above for more information about breaking down barriers to data access.‍
• Low-code and no-code AI: Thanks to rising computer power and declining computer costs, artificial intelligence has never been more accessible. With the emergence of low-code and no-code AI platforms, AI is reaching new levels of approachability. These tools make it possible for non-technical business users to structure unstructured data and use it in AI application development, lowering the barrier to entry considerably.

Overcoming regulatory obstacles

In computing a black box is a system that can be observed in terms of its inputs and outputs, without understanding how it works internally. The financial services industry is heavily regulated, and unexplainable AI software poses a unique hurdle for regulators as they want (and need) to understand how a given model works.

Explainable artificial intelligence (XAI) attempts to use methods and processes to ensure human users can trust and understand the results from machine learning algorithms. This emerging concept will only become more important as intelligent machines play an increasingly greater role in our lives, making it essential that we understand how they reach conclusions. For financial services providers considering building or implementing artificial intelligence must keep AI explainability top of mind.

Skills gaps

There is also another class of problems that center around a lack of resources and skills to implement AI at scale. It’s often easy to create a small demo project that looks really nice, but to implement at scale is something else entirely. This issue is compounded by the rapid pace at which AI technology is advancing, making it difficult for people to upskill quickly enough. Fortunately, no-code and low-code AI make it possible for non-technical business users to participate in building artificial intelligence solutions.

For example, super.AI’s Unstructured Data Processing (UDP) Platform makes it possible for anyone to train, test, and deploy custom AI solutions–all without learning to code. As platforms like ours grow and evolve to solve an increasingly broader set of problems, they will become essential tools that non-technical workers interact with on a daily basis.

Are the benefits of AI in financial services worth the hassle?

Despite all the challenges mentioned in this article, artificial intelligence is very much worth pursuing for financial services providers. From cost savings to productivity improvements, the benefits are huge. A few of the upsides to adopting of AI in financial services include:

• ‍Enhanced transaction data: Artificial intelligence makes it possible to identify hidden patterns buried in unstructured billing descriptor data, then automatically transform unrecognizable merchant names into clear transaction records. Both card issuers and cardholders benefit from having easily recognizable merchant names attached to each transaction. Benefits include fewer customer service inquiries, fewer payment disputes, lower operating costs, clearer statements, and better customer experience (CX). ‍
• Faster Know Your Customer (KYC) compliance: Automate KYC document processing to speed up new customer approval and onboarding, reduce human error, and lower operational costs while satisfying strict privacy compliance requirements.‍
• Intelligent document processing (IDP): Process, analyze, and extract relevant information from documents automatically to accelerate bank form processing, due diligence, credit application reviews, and more. Using artificial intelligence to process unstructured document data unlocks virtually endless possibilities for time and resource saving automations.‍
• Improved customer experience and engagement: AI can be used to generate personalized financial product recommendations customized based on user preferences and behavior, resulting in increased customer satisfaction and more revenue. Additionally, better customer experiences can be created by using AI to accelerate customer service response times, provide more relevant answers to inquiries, and more.‍
• Operational cost savings: There are huge operational savings at stake for financial services organizations that adopt artificial intelligence. According to research by Business Insider Intelligence, by 2023, banks are projected to save approximately $447 billion by developing and implementing AI applications. 

Additional AI for financial services resources

At super.AI, our mission is to automate boring work so that people can be more human. We strive to make artificial intelligence available to everyone with both the technology we build and the resources we create. If you’re interested in getting started with AI in financial services, check out the following resources:

• Schedule a personalized demo to see how AI can benefit your scenario.
• Read our blog: AI Technology Trends Transforming Financial Services