Customers Bank says its GenAI loan underwriting process cut cycle time by 95% and freed an estimated 76,000 hours.

Customers Bank Scales GenAI Loan Underwriting

Customers Bank is a US commercial bank and the operating arm of Customers Bancorp. On 22 October it reported third-quarter results. The following day, chief executive Sam Sidhu held the earnings call. He used it to set out what the bank’s GenAI loan underwriting work has delivered.

The bank piloted the process earlier in the year on selected commercial loans. In the third quarter it scaled that pilot, adding functionality and increasing use across its commercial lending teams.

What GenAI Loan Underwriting Replaced

Commercial credit approval is document-heavy work. Reporting by Bobsguide earlier in the year said the bank’s work with OpenAI targets two steps. First, collecting complex data on a borrower. Second, producing a first draft of the credit memorandum.

Customers Bank built the new process with OpenAI. Sidhu said on the call that the bank is leaning in heavily with OpenAI on the lending work. PYMNTS reported that the bank’s count of deployed agents includes custom GPTs and similar tools. More than 60 people now work on AI transformation at the bank. Alongside them sit 12 or more engineers embedded on site by its technology partners.

PYMNTS also reported a result. Selected commercial and commercial real estate loans now close in about seven days, against a typical 30 to 60.

The Results Customers Bank Reported

On the call, Sidhu said cycle time for the GenAI loan underwriting process is down 95%. The same team is handling significantly more loan volume. Cumulative estimated hours saved have grown to 76,000. The bank says this unlocked an estimated $15 million or more of annualised capacity. PYMNTS reported that the figure counts salary and benefits savings from posts left unfilled or reduced.

The number of AI agents the bank has deployed doubled during the quarter. Sidhu also said the productivity gains are showing up in financial performance. He cited revenue per employee up 13%, and pretax profit per employee up 34% year over year.

These are the bank’s own estimates, disclosed in its quarterly reporting. The hours figure comes from an internal employee survey. Other regulated lenders have taken comparable steps. Hiscox, for instance, has built a GenAI underwriting platform on the insurance side.

Why It Matters

  • Hours saved is a soft number until someone converts it. The discipline worth copying here is the restatement of saved hours as annualised capacity tied to specific roles left unfilled. Programmes that stop at the hours figure rarely locate the money.
  • In document-heavy work, calendar cycle time is the honest measure. If a file still takes the same number of days to reach a customer, nothing has changed for them yet.
  • Engineers embedded on site by the model provider are becoming part of the commercial arrangement. Any bank weighing something similar should budget for that, rather than assume a licence and an internal team will do.
  • Self-reported survey estimates deserve to be read as direction, not precision. A board should ask which figures will still show in the cost base next year.