A calculated future
What artificial intelligence means for banking
By Gabe Krajicek and Pradeep Ittycheria
Artificial intelligence (AI) is often used as a way to describe computer systems that can learn and perform tasks that previously required human intelligence.
Movies portray computer-powered machines that mirror our senses and surpass our cognitive functions. These aspirational capabilities are referred to as “general AI” and remain mostly in the realm of science fiction. What has progressed, however, is “narrow AI,” where advanced software is built around performing specific tasks. Image classification on social networks such as Pinterest and Facebook is a real-world example.
Let the machines do the math
As a topic, AI comprises multiple approaches, including machine learning (ML) and deep learning (DL). Machine learning, at its core, is the discipline of parsing data using algorithms, learning from it and then making a determination or prediction about the environment where those data were sourced.
Instead of writing code with finite instructions to carry out a task, the software looks for patterns in a large body of data and uses the findings to determine how a task should be performed. This process can be repeated an infinite number of times to refine the analysis and move toward the objective. For example, increased efficiency in a municipal power-grid is one possible application.
Deep learning relies on methods that mimic how human brains process information. Humans use neural networks to process large quantities of seemingly unconnected information and make decisions. Deep learning allows computers to simulate neural networks and create connections between disparate data points. This allows a computer to solve problems such as finding a stop sign in group of pictures that may or may not depict stop signs.
More than a glorified calculator, less than an overlord
AI earns a significant amount of hype, which is due in part to its complexity and the “science fiction factor,” where it is easier to assume the movies have it right than to evaluate the true potential.
In reality, its uses are significant, and the benefits will be felt across industries. According to the annual Gartner CIO agenda survey for 2019, the deployment of AI grew from 4% in 2018 to 14% in 2019.
Google CEO Sundar Pichai went as far as to claim that while the last 10 years have been about building a world that is mobile-first, the next 10 years will be AI-first. That is significant for a company that has the largest market share in mobile operating systems, and in a world where nearly 60% of all web searches happen on mobile devices.
How well is AI learning the ropes of banking?
Many of the mature uses of AI in banking are on the front end of the consumer experience. This includes customer identification and authentication and the simulation of live employees through chatbots and voice assistants. Other examples include the personalization of product pages on bank websites and marketing messaging.
The first chatbots capable of replying to basic questions appeared in 2015, with Ally Bank deploying its Ally Assist virtual assistant. Today, some of the most popular chatbots are Erica from Bank of America and Eno from Capital One. Banking consumers engage through many channels, including web, mobile, social and in-branch.
Going beyond customer support, banks can use predictive analytics to engage with consumers by anticipating their needs before the consumers even recognize it themselves. For example, offering a low-interest personal loan product when they start using their credit cards to buy large ticket items. Such engagement improves consumer loyalty but also generates new sales and revenue opportunities for the bank.
By automating repetitive tasks, banks can reduce costs and free staff to focus on more pressing needs.
Conversational banking can happen in an omnichannel environment through the combined use of AI-powered text conversations and in-branch human support. By automating repetitive tasks, banks can reduce costs and free staff to focus on more pressing needs.
For middle-office functions, AI is powering anti-fraud measures, risk management and credit underwriting. These capabilities are improving processes for anti-money laundering (AML) and know-your-customer (KYC) regulatory checks. This application of AI promises to lower compliance costs.
Already, AML/KYC due diligence processes have shrunk from hours into seconds. Data classification algorithms allow banks to identify what kind of document is being uploaded by recognizing template elements. Following classification, extraction of information found in the document to populate relevant fields improves the user experience. Finally, using enhanced optical character recognition (OCR), banks can check internal consistency and detect fraudulent alterations.
AI has unlocked alternative credit underwriting models that go beyond regression analysis. By analyzing hundreds of variables and thousands of consumer interactions to develop new risk models, banks can expand credit to the unbanked and underbanked without feeling constrained by credit scores alone.
Further out on the edge of the technology curve, AI will allow banks to authenticate payments using biometrics, as well as facilitating smart contracts, and automate compliance workflows.
Before there was AI, there was (and is) advanced analytics
AI deployments are not cheap and require large databases to create meaningful machine learning models or do any deep learning. The largest banks are spending billions on technology and have access to large data sets by virtue of their size.
JPMorgan, the largest spender, put roughly half of its $11.5 billion technology budget for 2019 toward innovation. Smaller banks need to evaluate the landscape before they decide on a course of action — opportunities may be hiding in plain sight.
The AI-enabled banking technology vendor space is swelling with hundreds of millions of dollars invested across scores of companies. This breed of specialized knowledge is extremely difficult for banks to cultivate internally, thus partnerships can prove fruitful.
Looking at banks that have successfully undergone an AI-enabled transformation, it is clear that a holistic strategy is required. Ideally, an implementation of AI will extend across business lines, usable data and partnerships with third parties.
However, before banks jump into the AI pool, they should examine their current data analysis capabilities. They may be underestimating their own ability to harvest meaningful insights from their data. Using a three-tiered maturity model of data analytics, we see that AI occupies the cutting edge of the highest tier, but the preceding tiers are foundational and offer significant value.
The first tier is historical reporting where we can answer questions about “What happened in the past?” Every bank has systems that support historical reporting; these are essential for regulatory compliance and performance tracking.
The subsequent tier is ad-hoc analysis, where we attempt to reveal “Why did it happen?” or “What is happening now?” Again, the third and highest tier is predictive analytics, where we can discern possible answers for “What is going to happen next?” or “What is the best that could happen?”
Even predictive analytics can offer value without requiring the application of AI. Banks do not need machine learning models to determine if a customer is going to reach a not sufficient funds (NSF) state. Collating average monthly account balances, average weekly spending and details of upcoming autopay bills can permit trigger-based messages for applicable products or services. This type of conclusion can be reached with conventional, non-AI models.
Predictive behavior modeling doesn’t require AI-intervention and is a proven way to facilitate personalized marketing offers at scale. Banks should seek partners that have proven behavior models that can effectively cross-sell additional products and encourage beneficial behaviors. It’s true that AI can improve the accuracy of such models, thanks to continuous learning capacity, but there is plenty of opportunity within reach of most banks.
Test, but verify.
Banks that do not have billions or even millions to spend on an AI strategy should first define a data strategy that goes beyond historical reporting. They should build partnerships with technology firms that can leverage models proven against hundreds of financial institutions and tailored for the individual bank.
This allows banks to augment the size of their database (which may be too limited for effective modeling). Banks should also view technology partners as a way to close skill gaps. While most banks have resources that can handle historical reporting, predictive analytics requires specialized skills.
Data science (the discipline that encompasses the three tiers of analytical maturity) is a blend of various tools, algorithms and machine learning principles for discovering hidden patterns in raw data.
The right technology partnership gives banks the best chance to harness data science and grow into a successful future.
Data science isn’t a silver bullet — it doesn’t work flawlessly. The quality of bank data can be a serious constraint. Data quality issues range from completeness, validity, accuracy, integrity, consistency, all the way to timeliness.
Most banks don’t have significant quality issues with first-party (in-house) data that is generated in core banking systems and collected directly by the bank. However, all data science around customer engagement and marketing is built on third-party sources. This is when data quality becomes a limiting factor.
Finally, data science depends on constant testing and re-evaluation. Many banks invest in data science testing but do not invest in their ability to verify the results of their experiments. The highly regulated and risk-averse banking industry does not easily accommodate experimentation and testing. Here again, the right technology partnership gives banks the best chance to harness data science and grow into a successful future.