Customers Bancorp expands AI loan processing after seven-day pilot closings

Customers Bancorp is expanding an AI-powered commercial-loan process after a pilot closed selected loans in seven days, compared with the bank’s stated usual 30–60 days. The seven-day milestone appeared in July’s second-quarter results. Its 22 October update reports increased use, additional functionality and a separate pilot opening commercial deposit accounts in minutes.[1][2]

What exactly became faster?

Commercial loan closing is a chain, not a single decision. Staff verify financial information, collect compliance materials, confirm collateral and negotiate terms. AI is potentially helpful for checking documents against a list, extracting values, flagging missing items, drafting communications and moving work between people. The bank's public announcement describes an AI-powered process rather than providing a step-by-step controlled evaluation assigning gains to each automated function.[1]

The most straightforward business question is how many loans were closed in seven days, and how representative they were. The company did not provide that denominator with its quoted pilot milestone. If cases were chosen for clean documentation, modest complexity or staff availability, the figure describes those circumstances. To estimate system-wide effect, an analyst would want a distribution of closing times across similar applicants before and after the intervention, ideally controlling for product type, size, documentation requirements and exceptions.[1]

A headline productivity figure with an instructive footnote

The third-quarter release reports twice as many AI agents deployed during the quarter and cumulative estimated time savings of 76,000 hours. Its definition of agents includes custom GPTs and similar products used in the previous six months. These counts describe adoption as well as automation.[2]

The hours-saved number came from periodic employee surveys of people using AI tools. The estimates were recorded at submission and were not subsequently remeasured or validated, the bank stated. They may include savings that later changed. It would be misleading to present 76,000 as a time-and-motion audit of completed tasks. It is a cumulative management estimate based on staff reports, which makes it a useful signal of adoption but a weaker measure of net economic gain.[2]

The annualized-capacity figure has its own definition. A footnote describes salary and benefits savings from positions not backfilled and open positions reduced year to date. It does not mean AI generated $15 million of additional cash sales, and it does not demonstrate that the estimated reduction in work hours is identical to those employment changes. The result depends on organizational decisions about hiring and redeploying labour as well as software performance.[2]

Extending the process to deposits

The October statement described another measurable workflow development: a completed trial in which new commercial deposit accounts were opened in minutes rather than days. It also disclosed a partnership with Palantir intended to support scaling, alongside growth in the company's payments and lending operations. These developments suggest the bank is targeting administrative bottlenecks across customer onboarding, not simply experimenting with a chatbot detached from operational systems.[2]

The financial results provide context but are not proof that AI caused them. The bank reported a third-quarter efficiency ratio of 47.2%, compared with 50.6% in the preceding quarter, and year-over-year growth across several balance-sheet measures. Interest income, portfolio mix, staffing, markets and other initiatives also influence such outcomes. Attribution would require isolating the relevant expenses, timing and counterfactual performance—not merely placing two numbers in the same investor presentation.[2]

The disclosures show a bank moving AI into several administrative workflows and, unusually, explain how some productivity estimates were constructed. The remaining test is repeatability across the normal mix of customers: a distribution of closing times alongside errors, exceptions and credit performance would make the seven-day example a service-wide result.[1][2]

Report an error or suggest a correction