Out Of The Loop

The big change in financial services comes not when banks get AI, but when customers do.

Jan 22, 2023

Dateline: Woking, 22nd January 2023.

Back in 2019, I wrote a piece for Wired predicting that in time robots will take over in banking just as they have done in manufacturing. Was I predicting that you would be served by a machine when you go to the bank ten years from now? Of course not. That would be ridiculous. For one thing, you won’t be going to a bank ten years from now under any circumstances — whether in the metaverse or the universe — and nor will anyone else.

(You'll be explaining "going to" a bank to baffled youngsters just as you were explaining "dialling" a phone to the first generation of internet children ten years ago.)

Going to a bank? Hilarious. No way. That is just the sort of boring and time-consuming thing that robots should be doing instead of people which is why I said that the big change in financial services will come not when banks are using AI, but when customers are.

This post is public, please feel free to share it.

A couple of years ago in Forbes I doubled down on that prediction, saying that the battle for future businesses will take place in landscapes across which their bots will roam to negotiate with their counterparts (i.e., other bots at regulated financial institutions) to obtain the best possible product for their "owners". At the time, I thought this was a 3-5 year trend and I was using it in workshops with advisory clients and boards to help set the outlines for medium-term strategy in the digital financial services sector.

That was before ChatGPT.

Time for a Chat

Unless you are hibernating for the winter you will have noticed that ChatGPT (GPT stands for “generative pre-trained transformer”) has blown up. It got its first million users signed up in five days! An open-source, end-to-end dialogue system built on top of the GPT-3 language model, it is designed to enable developers to quickly build and deploy conversational AI agents for chatbots, virtual assistants, and other interactive applications. It is so named because it is powered by the GPT-3 language model, which is a powerful natural language processing (NLP) model that can generate human-like text from a few words of input.

Before we continue to explore the implications, I must first point out that just as “the blockchain” was misleading labelled the truth machine, ChatGPT should not be seen as an oracle. I saw it described this way: The philosopher Harry Frankfurt defined bullshit as speech that is intended to persuade without regard for the truth. In that sense, ChatGPT is the greatest bullshitter ever. It is a Large Language Model (LLM). Such models produce plausible text but not true statements, since they cannot evaluate what is true or not. That is not their purpose.

Talking about bullshit, fortunately ChatGPT itself told the BBC that it would not take over the work of human writers (e.g., me), but would help them by providing suggestions and ideas leaving the human writer to create the final product.

(As an experiment to test this, I wrote one of the paragraphs for this article using ChatGPT. I’ll tell you which one at the end.)

Chat and Win

I saw on Twitter that DoNotPay, “the world’s first robot lawyer”, which was created by Stanford University student Joshua Browder to help people to contest parking tickets, has created a browser extension where customers can type something like “negotiate my cable bill down” and have a GPT-powered bot engage with their cable provider on their behalf.

Give a gift subscription

Browder has already tweeted a screen-capture video of DoNotPay’s ChatGPT bot negotiating a customer’s $10 per month discount at Comcast. The bot took the questions from a Comcast customer support person (at least, it seemed to be a person) and fed them into ChatGPT then passed back the answers.

He notes that the cable providers will also be using GPT and its ilk, so the inevitable trajectory is indeed bots negotiating with each other! This will certainly level the playing field between the data rich supplier trying to optimise pricing and the customers sharing anonymised data to negotiate services. I can easily imagine (indeed, I would pay for) a bot that could continually discuss the availability of frequent flier reward seats with British Airways and who wouldn’t want a bot to exploit account-opening incentives to switch savings across institutions.

Since ChatGPT has already been used to dispute a parking ticket (and win), I think he may be onto something and parking tickets and cable bills are only the start. Something that I thought was a couple of years away is happening right now and it is absolutely fascinating! Mortgages and pensions are next.

Loyalty Bonus

In this coming environment, the key question for customers will become a question of which bot they want to work with, not which bank.

My view, and I have to say that it is a tangential and uninformed view since I know almost nothing about marketing and even less about artificial general intelligence, is that people will tend to choose bots on the basis of moral and ethical frameworks. We will see Green Bots, Christian Bots, Pro-Choice Bots and who knows what other kinds of bots capable of informing choices within such frameworks. I used the examples of the AARP Automaton, the Buffett Bot and the Megatron Musk, all of which sounded plausible to me!

Once customers have chosen their bots, why would they risk making suboptimal choices around their financial health by interfering in the artificial brain's decisions? Think about it. As things stand, I am required to make decisions about my financial health while I am the least qualified entity in the loop!

Here is an obvious case study. Like most middle-class fintech bloggers, my pensions are by far and away the most important financial product in my life. I am also a reasonably intelligent person, yet in any conversation with my financial advisor, I very quickly sink into a miasma of government rules about pensions, complex taxation arrangements and risk-related investment decisions. I have absolutely no reason to believe that I can make the right decisions and would much prefer a super-intelligent bot to work with my financial advisor (well, my financial advisor’s bot) to plot the correct course through these difficult financial waters.

Decisions, Decisions

I got a letter from the tax authorities a few days ago. It is about an overdue payment of a couple of thousand dollars. I was due to pay it on 16th November, but I forgot all about it. The tax calculation is complicated, so I have no idea whether it is correct or not. I just assume that QuickBooks fed in the right figures. The tax payment is boring, since I have to go to the website and log in to do it, so I wish that my financial advisor bot could check the calculation and pay it for me. Human beings should not be required to go through these calculations and then log into their bank to send the money.

I got another letter from a parking enforcement services company because I’d parked at a nearby railway station and had not paid, so I was being fined. I was enraged. My phone was smart enough to know that I had parked my car and it knew where I was parked but it was not smart enough to pay for the parking and since payments and parking are both so boring I can scarcely bring myself to write about them, and I was thinking about more important things such as work, I had forgotten to use my phone app to pay as I caught the train. Why didn’t a supercomputer at the other end of my phone do it for me? Human beings should not be required to think about paying for parking.

It seems to me these examples illustrate my theory that in day-to-day life there are two categories of retail financial transaction that might better be handed over to agents with access to our wallets. There are the transactions that are too boring for people to do and transactions that are too complicated for people to understand. In both cases, the bots stand ready to help and given the rapid advances in technology, I don’t think I’ll have to wait too much longer.

(If you are curious, by the way, I used ChatGPT to write the paragraph describing what ChatGPT is. We are down the rabbit hole, for sure!)