American banks launched websites in the 1990s to serve customers better. Now, today’s platforms—with their chatbots and interactive features—provide a powerful real-time window into local financial conditions, helping banks weather turbulence.
Banks that track and share customer data beyond their marketing departments are far more agile in managing impending recessions and recoveries, says an analysis of US banks by Harvard Business School Assistant Professor Jung Koo Kang.
Compared with the average bank, those with robust consumer-facing technology:
Anticipated deteriorating economic conditions better, adjusting loan loss provisions 2.2% faster.
Held 13.4% fewer non-performing loans, or those headed for default, after periods of economic distress.
“For the banks, macroeconomic conditions are a very important component for core operations,” Kang says. Consumer data helps banks “manage their risk, better navigate future uncertainties, and make decisions to address those issues.”
Kang’s findings arrive as major banks push aggressively into artificial intelligence, spending an estimated $53 billion on the technology this year alone, according to Statista. Banks that can use data and systems to create a “feedback loop between customer interactions and institutional decision-making” stand to benefit further from such investments, the research suggests.
Kang shares his findings in “Customer-Facing Technologies and Banks’ Macroeconomic Information Production.” He coauthored the January working paper with Wilbur Chen, assistant professor at the Hong Kong University of Science and Technology, Columbia Business School Assistant Professor Sehwa Kim, and HBS predoctoral fellow Ling Lin.
Measuring the scale of bank technology
To trace how US banks benefit from such data, the researchers used the BuiltWith platform to identify web technologies embedded in the source code of 4,700 banks’ websites from 2000 to 2023. Compared with apps, consumers spend more time on websites for longer-term business, such as loans and investments.
The researchers used AI to distinguish between web technologies that collect macroeconomically relevant customer information and those primarily used for marketing or website management. They combined bank-level data from regulatory filings with regional economic data from the Federal Reserve Bank of Philadelphia and other government sources.
Regulatory filings and conference call transcripts showed when and how often executives shared economic conditions. The authors excluded data from 2008 and 2009, when the global financial crisis roiled markets. They found that banks with stronger tech infrastructure:
Share more economic information. These banks included 4.7% more discussion of macroeconomic conditions in their regulatory filings than the average bank. This suggests they had a richer understanding of local economic trends and risks.
Extract insights from data more effectively. The researchers note that customer-facing technologies create the most value when banks have the large, digitally engaged customer bases and analytical capabilities needed to turn customer interactions into meaningful economic insights.
Forecast and managed risk more effectively. The paper notes that macroeconomic conditions are difficult to predict, but play a critical role in assessing credit risk.
“If you look across all loan portfolios and their [nonperforming loan] ratios, I think banks can save substantial credit costs with this early signal about macroeconomic conditions,” he says.
In California, where privacy laws limit some of the data banks can gather, insight into economic conditions diminished, the authors found, helping to validate the data’s usefulness in predicting potential losses and other insights.
The researchers also found that customer data was especially helpful to banks operating in “economically volatile regions,” where rapid shifts are more likely to cause pain for consumers. Traditional borrower information is less likely to yield meaningful insights in these areas, the paper says.
A new way to think about bank technology
So how can bank leaders harness their platforms to their fullest potential? Kang’s research points to several considerations:
Customer data must reach other core functions
Banks have long viewed tracking platforms, such as Google Analytics, and chatbots as marketing tools. However, Kang’s findings illustrate how such data can inform risk management, underwriting, and financial reporting.
Consider “designing an organization that could be more integrated where information can easily flow across departments, so that it can create synergies,” he says. “Because data can be multipurpose.”
Data collection, not design, delivers more benefits
A beautiful website that draws users in won’t help a bank forecast better if it doesn’t gather data. Banks gain the most when their chatbots, behavioral tracking, and other analytics tools go beyond customers’ personal preferences and use patterns.
“Technologies that track indicators of financial distress, borrowing intent, or geographic variation in service inquiries generate data that can be aggregated into meaningful proxies for local economic activity,” the researchers write.
Engagement and scale matter
Banks that have more assets and consumer lending naturally collect more information than small firms. But banks whose customers routinely use websites and digital tools to manage their money collect “richer behavioral data, thereby enhancing their macroeconomic forecasting ability,” the paper says.
Leveraging data takes the right talent
The researchers found that banks with a higher share of data-related employees were able to predict their loan-loss provisions more accurately. In addition to data scientists and analysts, banks need managers who can link data to strategy, Kang notes.
“As a manager, you should think about developing some ability to recognize the value the organizational data your firm collects as part of its everyday operations, and consider how that data can generate insights that support strategic decisions,” he says.
Illustration by Ariana Cohen-Halberstam.
