Customers Bancorp (CUBI) AI Agents Save 76,000 Hours Across the Bank

On October 23, 2026, Customers Bancorp (CUBI) said on its third-quarter earnings call that its bank-wide AI agent program has saved an estimated 76,000 cumulative hours [1]. The company has not yet converted these hours saved into a dollar amount, and the savings do not show up in any specific expense line.

What Customers Bancorp is building with AI agents

Customers Bancorp is a bank holding company headquartered in Pennsylvania. Its subsidiary, Customers Bank, mainly serves business and institutional clients across commercial lending, specialty lending, deposits and payment services. Most of its revenue comes from the spread between loans and deposits, known as net interest income.

The AI application is a change in how work gets done across the whole workforce. The bank's internal AI transformation team pairs with employees who know the business, shadows real workflows, and builds AI agents on the bank's own data and systems. AI agents are software programs that carry out tasks automatically through defined steps. The team validates results first and then hands the agents to business units for daily use. The agents take over repetitive work in departments such as credit, account opening and compliance, saving staff hours. They do not charge customers directly. They act as a shared platform supporting every business line, and management positions the program as an operating tool to raise output per employee.

How the program changed from April to October 2026

The program's operating model was upgraded between April 24, 2026 and October 23, 2026:

• April 24, 2026: The first call to disclose an hours figure. At that stage, employees mainly built their own agents and custom GPTs, and cumulative savings exceeded 28,000 hours [2].
• July 24, 2026: The approach shifted to pods that pair engineers with subject-matter experts, starting with the highest-impact areas and measuring actual results before folding the agents into daily operations. Hours saved rose to at least 46,000, and every employee had an AI license [3].
• October 23, 2026: More than 60 people were focused on AI transformation, forward-deployed engineers from partners had joined the pods, and the focus moved to automating end-to-end workflows as a whole [1].

The 76,000 hours figure and its limits

The 76,000 hours is the program's most important result so far, and the three quarterly figures come from the same cumulative measure. On July 24, 2026, management said the 46,000 hours were up about 65% from the prior quarter and equivalent to about 24 FTEs [3]. The 28,000 hours reported on April 24, 2026 were described as equivalent to almost 15 FTEs [2].

These are cumulative management estimates. The company has not disclosed a start date or calculation method, and it has not broken the figures down by workflow. This quarter, management also said the number of deployed agents had doubled and estimated that the program had unlocked more than "15 million plus of annualized capacity" [1]. However, no base was given for the agent count, and no unit or method was given for the capacity figure. Neither can be compared directly with the earlier count of agents "built."

Where the savings could show up in the financials

If the hours saved translate into fewer hires, or into moving staff to revenue-generating roles, the most likely place to see them is noninterest expense and the efficiency ratio: revenue would grow faster than expenses, and the efficiency ratio would fall. Management ties this path to a target of bringing the operating efficiency ratio down to the low 40s by the end of 2027. On July 24, 2026, that ratio was about 50% [3].

This quarter, management said productivity gains already exceed its AI investment. Revenue per employee rose 13% year over year, and pre-tax profit per employee rose 34%. Over the past six months, revenue-generating staff increased 6% and made up about 60% of new hires, while total headcount declined slightly [1].

These are bank-wide figures, though. Over the same period, net interest income, loans and deposits grew and the staff mix changed, and the company actually raised its noninterest expense guidance. No single expense line has been shown to decline because of this program.

What is confirmed and what remains open

What can be confirmed is that CUBI has moved AI agents from employee-driven experimentation to workflow-by-workflow deployment by a dedicated team, and that hours saved have kept accumulating over three quarters to 76,000. The company has not separately quantified what those hours are worth or how much of its expense base they represent. The program's operating impact will only be validated if future quarters convert the hours saved into specific dollar amounts and show a matching change in noninterest expense or the efficiency ratio.

Application assessment

• Enterprise AI Workforce Enablement | Business position: shared platform | Stage: limited production | Scope: company-wide | Value type: cost reduction

Sources

[1] Drillr · Customers Bancorp, Inc. (CUBI) · 2026-10-23 · Earnings call

"Our team doubled the number of AI agents deployed. Cumulative estimated hours saved have now grown to 76,000. And we've unlocked an estimated 15 million plus of annualized capacity."

[2] Drillr · Customers Bancorp, Inc. (CUBI) · 2026-04-24 · Earnings call

"We have saved more than 28,000 hours through AI-enabled workflows, unlocking the equivalent of almost 15 FTEs."

[3] Drillr · Customers Bancorp, Inc. (CUBI) · 2026-07-24 · Earnings call

"our team has now saved at least 46,000 hours through AI-enabled workflow automation, up about 65% from last quarter and equivalent to 24 FTEs."