Banks are getting faster at making decisions. The question is who, or what, is actually making them.
Financial institutions are already using AI and machine learning across lending, fraud detection and customer service. J.P. Morgan recently highlighted how machine learning is expanding in credit analysis. The bank was also specific: human expertise remains necessary to interpret model outputs, challenge errors and maintain trust.
So the central question for banks is not just whether AI can handle a job that used to belong to a person. It is whether a bank should let a machine make a decision that could significantly affect someone’s life without somebody at the bank who can understand it, challenge it or own the outcome.
How Americans actually feel about AI making financial decisions
New research from Tunnl measured where Americans actually stand on this. The firm surveyed 3,066 U.S. adults. 63% said they would not accept a faster loan or claims decision made entirely by AI if there was no way to appeal the decision to a human.
That 63% splits into two groups. 39% do not want AI involved in the decision at all. 25% are fine with an AI-made decision as long as they can appeal to a person afterward.
A separate finding in the same research showed that automated and AI-driven decisions tied for the top area where Americans want more oversight, at 52%. That put it alongside how customer data is used or sold.
More AI:
- Nvidia just made a move Wall Street wasn’t ready for
- Microsoft just took sides in AI policy fight
- OpenAI just disclosed something genuinely alarming
This is not a rejection of AI. Consumers are already using it. PwC’s 2026 Consumer Lending Radar found that nearly a third of borrowers now use AI tools to research loans. 67% expect AI to inform their next borrowing decision. At the same time, three out of four consumers still want a human involved in loan approvals and closings.
Consumers will trade some human interaction for speed. What they will not trade is recourse. When a loan gets denied or a claim gets rejected, they want to know someone is available if the system got it wrong.
How banks should decide which decisions AI can own
Automated systems have existed in finance for a long time. AI expands what those systems can handle. That is not the problem. The problem is figuring out when a faster, more complex AI decision crosses into territory where its consequences require a human to be in the loop.
“The dividing line shouldn’t simply be whether AI can make the decision. It should be the consequence of getting that decision wrong,” Sara Fagen, co-founder and CEO of Tunnl, told TheStreet in an interview.
A loan denial is not the same kind of error as a misrouted support ticket. One can affect someone’s financial stability for months or years. The other gets fixed in minutes. That is the distinction she is pointing to.
Automation works better when inputs are predictable and outputs are clear: validating a document, routing a standard application, flagging a duplicate. These are reasonable candidates for automation. As the judgment required increases, so does the need for a person to be making the final call rather than simply approving a machine’s recommendation after the fact.
What happens when AI makes the call and gets it wrong
The Government Accountability Office has looked at AI in financial services. Benefits are real: efficiency, lower costs, faster processing. So are the risks: potential lending bias, data-quality problems, privacy concerns and cybersecurity vulnerabilities.
The GAO also found something important. Most financial regulators it spoke with said AI outputs inform staff decisions. They are not the sole basis for action.
The institution owns the outcome, either way. A good algorithm call benefits the bank. A bad one? The bank still has to answer for it. And bad decisions can scale fast. An algorithm that makes a harmful call once can make the same call thousands of times before anyone catches it.
“The line gets crossed when they start outsourcing understanding, judgment, or accountability to it,” Nate Herk, founder and CEO of AI Automation Society, told TheStreet.
Banks face pressure from both sides here. Too little automation, and competitors pull ahead on speed and cost. Too much automation, and errors compound faster than any review process can respond. Getting the balance wrong in either direction creates real exposure.
Fagen added: “The most important part of that finding isn’t that consumers are rejecting AI. They’re rejecting a system in which AI has the final word. That’s a meaningful distinction.”
Where human judgment still creates competitive advantage
Every decision does not need a human. But the decisions most likely to seriously damage a consumer’s financial situation are the ones where human judgment matters most.
Herk added: “The fact that something can technically be automated does not mean it should be.”
He points to an example from his own business. A community manager role could be automated. He chose not to automate it. Members are not just there to get questions answered. They want to know a real person cares about what happens to them. Automating that role would look like efficiency from the inside and feel like abandonment from the outside.
Financial institutions face that same dynamic at their most consequential moments. When a customer gets denied, confused or faces a decision with real financial stakes, what the institution does next can determine whether that customer stays. Those moments may be exactly the ones that feel most automatable but are the most important to get right.
Research into AI and financial services has increasingly focused on calibrated trust. A recent study in Financial Innovation argues that financial institutions need consumers to rely on AI when it is likely to improve a decision, while withholding reliance when it is not. Financial errors can be costly and hard to reverse.
Machines can absorb the analytical workload. They cannot absorb the accountability. Banks that keep that distinction clear may find it is not just a compliance posture. It may be a competitive one.
Related: Michael Burry doubles down on his surprising AI bet
