Banking Processes Managed by AI

Consumer40

I recently made a large faster payment transfer, above my bank's online limit, and so had to submit the request over the phone. The agent carrying out verification checks was concerned that I was not phoning from one of my registered numbers. I thought I had notified the bank of revised contact details earlier in the year but accepted that I was mistaken.

I subsequently logged in and updated my phone number, observing that the update process generated a secure message. Later I was notified of a new secure message and logged in to read that my details had been updated - the revised details were set out in the message and were correct.

To be absolutely sure that the process had worked, I revisited the 'Contact Details' page and was surprised to see that it showed my old details. This encouraged me to take another look at the secure messages page where I discovered that I had in fact notified the bank of the change in May (as well as last week) and that also resulted in a response confirming that the changes had been applied.

I logged a complaint with the bank. The complaint was upheld. The agent believed that the update process was managed by AI.

I imagine most of us have experienced AI giving incorrect answers but I am a bit alarmed that a mechanism that is inherently non-deterministic, and so liable to behave differently every time, might be allowed to process bank account updates.

Does anyone with knowledge of the industry know if AI is being employed in this way, or is it likely that the agent was mistaken?

Woodstok2000

Modern agentic AI is non-deterministic, but that doesn't mean that all AI systems are (it's a very broad definition) and it also doesn't mean they are liable to behave differently every time. A well written system performing appropriate tasks is certainly capable of 99%+ accuracy, but you still need to have human oversight built in to ensure you're catching the errors.

I'm not sure why you'd need agentic AI to process address updates, but if they are using it this sounds like a failure of governance.

subjecttocontract

I smell a whiff of compensation due. None of us would need that sort of hassle/aggravation.

wmb194

https://forums.moneysavingexpert.com/discussion/6685144/banking-processes-managed-by-ai

By any chance was this Lloyds/Halifax/BoS and do you have accounts with more than one of these brands? If so and long before “AI” I remember people saying they’d had problems updating contact details: its systems try to automatically propagate updated details between its various banks and their systems but it doesn’t always work correctly, including reverting back to previous details.

Chief_of_Staffy

If AI was the problem then it was badly trained. The technology itself is fine. More likely it's the archaic processes the banks still use. Very little is updated in real time; instead information is updated periodically like a 1970s mainframe batch job (which I don't doubt some still are). And periodically often means five times a week. My address change took four days to show up on my account. The phone number change on my credit card is coming up for five years. I have to keep my old phone active purely to receive SMS texts from the bank on my credit account. Nobody can fix it because the developers of the banking software are long since dead and nobody else knows how they work.

surreysaver

https://forums.moneysavingexpert.com/discussion/comment/82155307#Comment_82155307

I don't know where you get that AI is fine from. As far as AI is concerned, it stands for Artificial Incompetence. Its okay to use to start looking for an official source of information, but to rely on it is to fail.

eskbanker

https://forums.moneysavingexpert.com/discussion/comment/82155324#Comment_82155324

To pigeonhole AI on the basis of one specific use case (of punters looking for official information) would also be to fail…

Chief_of_Staffy

https://forums.moneysavingexpert.com/discussion/comment/82155324#Comment_82155324

On what basis do you make that statement?

Consumer40

Modern agentic AI is non-deterministic, but that doesn't mean that all AI systems are …

You make a good point about the distinction between a specialised AI and an LLM such as Gemini or Chat-GPT. I gather that specialised AIs can do a better job than medics at identifying a tumour on an X-ray. So I can see that a specialised AI might have a role in, say, credit scoring, where there is no right answer and traditional systems are already opaque. However, for something like a name & address update, I agree with you that it does not seem appropriate.

I smell a whiff of compensation due.

Normally I would ask as I think failure should incur a cost for the service provider and so act as an incentive to improve. However, the complaint handling was excellent. In a few minutes the agent had updated the system properly, reviewed the secure messages, agreed that my experience was unacceptable and spent time documenting feedback for the relevant team - including incorporating further points that I wanted him to to add. If AI was the problem then I guess the staff have a vested interest in exposing it.

By any chance was this Lloyds/Halifax/BoS …?

It was the Co-operative Bank. I believe they are working on refreshing their systems but I hope they are not tempted to take any risky shortcuts.