AI Agents & Business Automation

A Bank Doubled Its AI Agents and Its Efficiency Ratio Dropped 340 Basis Points

Customers Bancorp's Q3 2026 results show AI agents moving from pilots to measured operations, with 76,000 hours saved and a 47.2% efficiency ratio. Here's what that means for small businesses.

Back to NewsA Bank Doubled Its AI Agents and Its Efficiency Ratio Dropped 340 Basis Points

When a mid-size bank tells its investors that it doubled the number of AI agents it has running, and then points to a 340 basis point improvement in its efficiency ratio, you should pay attention. Customers Bancorp (NYSE: CUBI) held its Q3 2026 earnings call on October 23, and the results read like a case study in what happens when AI agents stop being a pilot project and start being an operating discipline. For a $20 billion-plus bank, the numbers are impressive. For a five-person plumbing company or a two-lawyer firm, the signal is bigger: the argument that agents are still experimental just got a lot weaker.

Key points

  • Customers Bancorp reported Q3 2026 EPS of $2.25, up roughly 10% sequentially, with record deposits of $22.6 billion and record loans of $18.5 billion.
  • The bank's AI program has identified $15 million in annualized capacity, saved 76,000 cumulative hours, and cut loan underwriting cycle times by a reported 95%.
  • Revenue per employee rose 13% annualized, which management ranked first among banks with $10B to $100B in assets; pretax profit per employee rose 34%.
  • The bank doubled its deployed AI agents in the quarter and targets a low-40s efficiency ratio by the end of 2027.
  • Management flagged real risks too: a Q4 expense step-up and potential borrower stress from 2027 rate resets, a useful reminder that agent investments sit inside a business with normal pressures.

Start with the financials, because they frame everything else. Customers Bancorp reported net income available to common shareholders of $78.4 million, or $2.25 per diluted share, up roughly 10% from the prior quarter. Return on equity came in at 14%, up 70 basis points sequentially, and return on assets at 1.16%. The efficiency ratio improved to 47.2%. Deposits hit a record $22.6 billion, up 4% quarter over quarter and 11% year over year. Loans hit a record $18.5 billion, up 3% and 14% on the same basis. Tangible book value per share reached $65.66, the 17th consecutive quarterly record. All of these figures are as reported by management on the call and summarized by the Motley Fool transcript and other outlets covering the results. The bank also bought back about 492,000 shares, roughly 1.5% of common stock, for $38 million, and its CET1 capital ratio stood at 12.6%. A consent order from the Pennsylvania Department of Banking and Securities was lifted during the period, which management framed as evidence that risk management has become a strength rather than a liability. Non-performing assets fell to 0.29% of total assets, below the 0.36% peer median they cited.

Here's the part that matters if you build or buy agents for a living. Management said the AI and automation program has identified $15 million in annualized capacity, saved 76,000 cumulative hours, and cut loan underwriting cycle times by a reported 95%. Commercial onboarding times dropped by more than 90%. More than 60 employees are dedicated to the AI transformation effort, working alongside forward-deployed partner engineers, and the number of deployed AI agents doubled during the quarter. Notice what the bank measures. It doesn't brag about chatbot conversation counts or demo videos. It measures hours, capacity, and cycle time, and then it ties those to per-employee economics: revenue per employee up 13% annualized, which management ranked first among banks in the $10B to $100B asset class, and pretax profit per employee up 34%. That's the honest way to evaluate agents. If you run a small business, the equivalent questions are concrete: how many missed calls came back as booked jobs, how many hours of follow-up did the team get back, and how much faster does a lead go from first contact to first appointment.

Strip away the banking specifics and the underlying pattern is one you see in every service business. The two biggest wins CUBI reported, faster underwriting and faster commercial onboarding, are both versions of the same problem: a process where the gap between a customer's initial interest and their actual engagement is long enough that people drop off. A 95% reduction in underwriting cycle time means a borrower who would have waited days gets an answer in minutes. A 90% reduction in onboarding time means a commercial client who might have drifted to a competitor gets set up while they're still motivated. That's exactly what an intake agent, an appointment setter, or a follow-up agent does for a trades business, a law firm, a medical or dental practice, an insurance agency, or a real estate team. Someone calls after hours, or fills out a form, or texts a number they found online. The window to respond is short. An agent that answers, qualifies, books, and routes that person compresses the cycle the same way CUBI's underwriting agents compress theirs. The mechanics differ. The economics don't.

One smaller data point from the call deserves more attention than it will probably get. After the September rate increase, the bank's deposit beta ran about 33% over the following 30 days, versus roughly 60% in the prior rising-rate cycle. In plain terms, the bank kept more of the benefit of higher rates this time because customers were stickier, partly because of products like the cubiX platform, which is on pace to exceed $2 trillion in annual transaction volume and has diversified well beyond its original digital-asset clientele into real estate and fund finance. The lesson generalizes: businesses that invest in faster, better customer experiences end up with pricing power, because switching costs go up when the experience is genuinely easier. That's true for a bank's deposit platform and it's true for the contractor whose customers can book a job at 9 p.m. without leaving a voicemail.

To be fair to the bank's own framing, management didn't present AI as a cure-all, and neither should you read it that way. The CFO flagged a Q4 expense step-up driven by incentive compensation and continued technology investment. He also pointed to potential borrower stress when 2027 interest-rate resets arrive. CRE sentiment was described as somewhat more cautious. These are normal business pressures, and they're a useful reminder that agents sit inside a business with real risks, not above them. It's also worth stating clearly: these are company-reported figures from one bank's operations. They tell you what was possible for Customers Bancorp. They don't tell you what any other business will get, and anyone who promises you a specific result from deploying agents is guessing.

The most useful takeaway from CUBI's quarter isn't the EPS beat or the record deposits. It's the proof that a large, regulated, risk-conscious organization can run AI agents in production, measure them in hours and cycle time, and tie the results to per-employee profitability, with a stated target of a low-40s efficiency ratio by the end of 2027. When banks start treating agents as core operating infrastructure, the businesses that sell to banks, borrow from them, and compete alongside them will feel the pace change. If you've been waiting for evidence that agents are past the experiment stage, this is a bank handing you the receipts. The question for a small business owner isn't whether the big players are doing it. It's whether the missed calls and slow follow-ups in your own shop are worth fixing before everyone else fixes theirs.

Stay in the Loop