A US regional bank spent four months on the technical implementation of an AI loan-underwriting system, then needed 14 more months before it could actually go live. Underwriters took six months to define the criteria for when they could trust the AI's judgment, five months to settle who was accountable when something went wrong, and three months to prepare documentation for regulators. Even once the technology was ready, the organization hadn't moved — and AI stayed frozen in place until it did.
In a recent McKinsey analysis of AI transformation in banking, three experts said they'd seen this pattern repeat itself over and over. Once AI shifts from "assistant" to "doer" — the point where a person is no longer checking and judging, but AI is directly processing and producing the outcome — technical readiness stops being the bottleneck. The holdup lives somewhere else. For Korean middle managers and solo founders alike, this pattern will sound familiar.
When AI becomes the doer, the nature of errors changes
When AI is used as a tool, a person still makes the final call. It drafts an email, summarizes a contract, cleans up a dataset — and a person reviews the output, judges it, and signs off. If AI makes a mistake at this stage, the final reviewer catches it. The chain of accountability is relatively clear.
Once AI becomes the doer, that flow changes. In banking, it's increasingly common for AI to handle everything independently, from receiving a loan application to sending the approval or denial notice. Sorting customer inquiries, detecting anomalous transactions, drafting compliance documents — all of these are shifting the same way. People step back into exception-handling and oversight roles. AI makes the bulk of the decisions, and a person only steps in when something looks off.
Banks report that this shift cuts time spent on repetitive document processing by 60 to 70 percent. At the same time, problems that never existed before start showing up.
When AI takes over execution, the nature of error changes. When a person makes a mistake, it's a person's mistake. When AI makes a mistake, whose mistake is it — the team that built the model, the executives who decided to deploy it, or the staff member responsible for oversight? An organization that can't answer that question clearly can't put AI into production. McKinsey's report names this question of accountability as the first design problem that has to be solved.
It's not a technology ceiling — it's a gap in design
What McKinsey's experts flagged all converge on the same three things: organizational design, governance, and talent placement. How much latitude AI gets to make its own calls, who decides that scope, and what path accountability follows when something goes wrong — an organization where these remain undefined can't move AI into the doer role, no matter how ready the technology is.
Regulatory demands sit in the same territory. In the financial sector, AI decisions have to be explainable. If a bank can't explain to auditors and regulators why the model rejected a particular loan, or why it flagged a particular transaction as suspicious, it doesn't get the green light to operate. This explainability isn't a problem for the technical team alone to solve — it's an organization-wide task that has to weave together legal, compliance, and risk management.
Data readiness is tangled up in the same problem. An AI system acting as doer needs clean, consistent data to base its decisions on. What banks actually found was data scattered across decades-old legacy systems, not even standardized in format. They had to clean up the data before AI could go live, and that work took longer than the technical implementation itself.
The skepticism toward moving AI into the doer role deserves honest treatment too. In a field like finance, where the cost of error runs high, the argument goes that shrinking human final judgment could carry more risk than efficiency gain. In 2023, an AI loan-underwriting system at a US financial firm was found to be systematically scoring applicants from certain zip codes lower. It was a pattern a human underwriter would likely have flagged as strange — but the AI just kept repeating it. The case shows that without an oversight structure, putting AI into the doer role can amplify bias. The 2023 bias case is concrete evidence for the argument that design has to come before speed. McKinsey's report tacitly acknowledges this point as well.
What to check before moving AI into the doer role
Banking's case studies don't translate directly to every other industry. But the same shift — moving AI from assistant to doer — is showing up well beyond finance. Marketing content generation, customer inquiry handling, tax invoice processing, scheduling — plenty of solo founders have already started handing this work to AI.
Before doing so, there are things worth checking first. What's the process for catching and correcting AI output when it's wrong? Who's accountable? How will you explain it to a customer? Hand execution over to AI without settling this flow in advance, and you'll hit the same bottleneck McKinsey observed inside banks. Being a small operation doesn't make the problem simpler.
There's an approach to financial decision-making that sketches out the loss scenario before calculating the return — thinking through what damage a wrong call could do before thinking through how much you'd gain if it goes right. The same order of operations applies to shifting AI into the doer role: picture what breaks when AI gets it wrong before picturing the efficiency gains from when it works. Only an organization that has drawn that picture out fully is ready to move AI from assistant to doer.
If your AI adoption checklist is missing an error-detection cadence, a clear accountability structure, and an explainable processing flow, then no matter how far ahead your technology is, your organization will spend 14 of those 18 months waiting. I'd call this "loss-first design."
The more work a tool takes on, the more you need a person and a process to decide when to stop it. What blocks you after the technology is ready lives right there.
