Advances in machine learning and artificial intelligence are transforming how banks and other financial services organizations operate.
By understanding and applying the newest predictive and prescriptive capabilities of machine learning technology, banks can find new ways to improve efficiency, increase profitability, reduce risk, enhance customer engagement, and gain competitive advantage.
While most banks are working to improve their analytics and machine learning capabilities, most also are not satisfied with how they’re progressing. In a recent Crowe bank survey, 56 percent of the responding banks said they had a road map for the types of analytics projects they would like to do. Even more (77 percent) said they had data quality standards in place along with a plan to improve data governance practices related to advanced analytics. Despite this attention, however, only 33 percent of the respondents said they were getting the results they expected from advanced analytics.1
To understand where the shortfalls could be happening – and to recognize the opportunities that might be overlooked – it is necessary to begin with a clear understanding of the language of advanced analytics. Some of the key terms can be misunderstood or misinterpreted.
From a practical standpoint, artificial intelligence typically refers to artificial “specific” intelligence, which refers to machines programmed to learn a specific task, such as identifying customer churn. This is not to be confused with artificial “general” intelligence, which refers to the more far-fetched and often sensationalized idea of a machine capable of human-like thought and reasoning.
In the past, teaching a computer to perform complex tasks involved writing a large array of if-else rule statements. The rule sets often were limited in scope due to financial considerations and did not cover all cases in the data. Now these complex tasks can be completed using artificial intelligence based on machine learning algorithms. These self-adaptive algorithms learn to identify important patterns in data and update those patterns as new data is collected.
In banking, machine learning models can be used to improve customer segmentation, to estimate the probability of early loan repayment, and for other specific functions banks use to more accurately predict customer behaviors. One advanced type of machine learning, known as “deep learning,” employs algorithms that are particularly well-suited to tasks that involve large amounts of data. Deep learning also makes it possible for software to employ image recognition and natural language processing, both of which require processing enormous amounts of data.
By applying artificial intelligence, machine learning, and deep learning capabilities, banks can begin to apply advanced analytic capabilities that allow them to lower costs, modernize their data systems, and ultimately derive greater insights into expected customer behaviors. The turning point in achieving these goals comes when banks can move beyond traditional analytics (which involve looking back to evaluate prior trends and events) and begin instead to employ predictive analytics that use forward-looking models to analyze expected outcomes.
The exhibit helps demonstrate this concept by offering examples of various levels of analytics in five general areas of concern: customer intelligence, portfolio management, marketing and sales, risk and compliance, and operations.
Source: Crowe analysis
The first row of the grid offers examples of ways that traditional analytics can be applied in each of the areas of concern. These tools include a variety of useful functions such as the ability to see a 360-degree view of customers or branch performance, as well as live monitoring of customer activities, regulatory compliance, and other concerns.
The second row begins to explore some of the new possibilities that can be opened up through the use of predictive analytics. The third and fourth rows carry those possibilities even further, demonstrating some of the opportunities that can be pursued through the use of artificial intelligence and machine learning.
The first column of the exhibit, which focuses on customer intelligence, offers a good example of where these capabilities can lead.
Most of today’s widely used analytics systems can provide a bank with a 360-degree view of its customers, offering a consolidated picture of each customer’s full range of relationships with the bank, visible through a single portal. While this capability might have been considered state-of-the-art a few years ago, today it represents a baseline level of analytic ability, delivering a single view that draws on various data sources, including customer profiles and account data, transaction records, and records of other customer interactions with the bank.
Predictive analytics build on this foundation and begin to open broad new possibilities in customer intelligence. By applying various analytic engines and machine learning models to the same data sources, banks can begin to target high-value customers whose recent activities and behaviors indicate an opportunity to expand the relationship with the bank or, conversely, a propensity to end it prematurely.
Deeper advanced predictive capabilities can be achieved by supplementing the bank’s internal data sources with various external and online sources. Enriching it in this way expands the data volume exponentially, with the models taking into account many more factors and data points than any individual could hope to analyze manually.
Ultimately, artificial intelligence can be taken a step further, going beyond predictive analytics to prescriptive analytics. Predictive models employ artificial intelligence capabilities to develop a specific, systematic, and proactive retention strategy for each customer. Looking beyond the individual customer, such analytics also provide a foundation for improved product development strategies, as well as more cost-effective marketing and more responsive customer service efforts.
Above all, it should be noted that this power also can be applied in a similar fashion to a wide variety of other data-intensive processes in bank operations. These include anti-money laundering customer and transaction monitoring programs and various forecasting efforts related to stress-testing, loan portfolio management, and general capital management strategies.
In recent years, a growing number of banks have begun to discover that conventional project approaches that have served them well in the past are not necessarily the most effective way to move ahead with machine learning and artificial intelligence initiatives. The typical approach, with extensive planning and design phases, driven by a sensitivity to risk and an aversion to failure, actually can cause such projects to stall.
Instead, experience suggests that the effort to advance banks’ artificial intelligence and machine learning capabilities often requires a more accelerated process. This approach begins with some exploratory analyses but then quickly moves on to linking previously disconnected systems, and the rapid development of experimental models for testing and validation.
It is important to remember that, in such an iterative process, failures are not fatal. In fact, they should be expected. Initial assumptions often prove to be invalid and the results often are not what was expected. Those managing such projects should not expect the models to be perfect the first time. Instead, they should expect surprises and be prepared to move quickly to build on the lessons learned.
In today’s competitive banking industry, the ability to analyze historical data no longer is an adequate basis for success. Machine learning and other artificial intelligence capabilities can generate powerful insights and have a significant business effect in almost all areas of the organization. However, the key to capturing these benefits is to begin quickly and move forward assertively.
1 Crowe Data Governance and Analytics Study, January 2016