Lloyds Bank's AI cost-cutting plan has a hidden tab: who pays when the machine gets it wrong?
A researcher argues that banks counting only the minutes AI saves are ignoring the hours other workers spend fixing the mistakes it creates.
Key points
- Lloyds Bank announced a £2 billion cost-cutting plan on 30 July built around AI-driven automation.
- Dr Gleb Tsipursky, writing in The Guardian AI, argues the bank's productivity figures only count time saved for the worker using the AI tool.
- Hidden costs include colleagues checking invented facts, fixing customer messages, explaining rejected loan applications and escalating errors.
- A single team can appear more productive on paper while the actual workload shifts, quietly, to other parts of the bank.
- No public breakdown of these downstream costs has been published.
Lloyds Bank made headlines at the end of July when it unveiled a plan to cut £2 billion in costs, with artificial intelligence, the technology that powers tools like chatbots and automated document processing, doing much of the heavy lifting. The numbers looked clean. The plan did not.
Dr Gleb Tsipursky, an expert in decision-making and AI risk, is asking a question the headline figures do not answer: when the AI gets something wrong, whose time pays for it?
What does that hidden cost actually look like?
It looks like a colleague reading back through a customer letter the AI drafted and catching a made-up interest rate. It looks like a branch worker calling a customer to explain why their mortgage application was rejected based on a system error. It looks like an escalation queue that was not there before.
AI tools that generate text, summarise documents or process applications can produce what researchers call hallucinations, meaning confident-sounding answers that are factually wrong. Catching those errors takes human time. That time belongs to somebody.
The problem, Tsipursky argues, is simple accounting. Banks measure the minutes saved by the employee directly using the AI. They do not measure the minutes spent downstream by everyone cleaning up after it. One team posts a productivity gain. The risk and the effort move sideways.
Should bank customers be worried?
Not in a dramatic way, but there are things worth watching. If a bank's AI drafts a letter about your account, read it carefully before acting on any figures or deadlines it contains. If an application is rejected and the reason feels wrong or vague, ask for a human explanation in writing.
The practical concern is not that AI will cause a financial crisis. It is that errors get diffused across many small interactions, each one annoying or costly for the customer involved, while the bank's internal metrics show only the savings.
Lloyds has not published a breakdown of how it will track and report the downstream human cost of its AI programme. Until it does, the £2 billion figure tells only half the story.
What should readers watch for?
If you are a Lloyds customer or an employee at any bank rolling out AI tools, four things are worth your attention: letters or messages that quote specific numbers (rates, balances, deadlines) without a named human author; rejection notices that offer no clear route to appeal; tasks that used to take one person and now seem to require two; and any process where the AI's output goes to you without a human review step between the machine and your money.
