Lend Me Your Eyes | TWIF UK & Europe

👋 Welcome to the new readers who’ve joined us since last week. You’re joining 176,000+ other subscribers who love fintech.

Hello fintech friends,

Another fun week across the world.

Jensen brings Key Man risk to a new level as Doomers state Western-civilisation itself seems to have escaped from Huang’s pants, not once but TWICE!

Spain ruined my game of padel by having a nationwide mega STRIKE.

Adam Neumann mogs Elon Musk. And last, and most importantly, Manchester United manager Michael Carrick has been sacked.

OK… back to some fintech musings and the usual news.

Today, after about two years of thinking I could get a better deal than I'm currently on with my Barclaycard, I asked ChatGPT, "What should I do?" ChatGPT said, "If you're a Revolut customer, it's probably a good idea to get a personal loan. The interest rate will be cheaper. You can use that to pay off your card. Hey presto."

Now, I've known about this for a long time, but I've just been too lazy, inert, as the credit industry likes to refer to call me. I began doing it on Revolut, but now that I live in Spain, the first hurdle was that Revolut needs me to have a Spanish number, or I can't start the loan process. I fell at the first hurdle, and I'll figure out another way to do it.

This little obstacle is an insight into how a large part of the credit industry actually works, especially if you're issuing a card. Most people need to have a huge incentive to seek alternative finance facilities, namely because of the hassle involved in switching. An agent does not need a huge incentive to do the same and so will continuously seek the best rate possible for its master. Brands or entities depending on the goodwill (or the laziness) of a customer , stand to suffer greatly.

And yeah… underwriting is getting sharper. Revolut says its in-house foundation model, PRAGMA, is 2.3 times better at identifying credit default risk than the models it runs in production today. It was trained on billions of events from customers’ own banking histories, not some arcane, dusty bureau file.
What a shock! Reading the sequence of someone’s financial life, not just a three-digit summary of it, turns out to matter.

But a better model is only an edge until everyone has one. The smartest lenders oy yesteryear are at risk of becoming the animal feed of tomorrow’s AI models. Give it a few years and strong underwriting becomes table stakes, the way a mobile app did. The way a smooth onboarding journey did. Kids of today will never have to face the problem of taking a day off work to apply for an account!

So what is left to compete on? Distribution will be through LLMs so that doesn’t matter. Claude helpfully says proprietary data and the cost of funding.

But I don’t think so. Proprietary data matters today. But in the future it will be more akin to a superfluid. Uncontainable, it will leak through the imperceptibly small gaps in all our technology much like each new model finds an order of magnitude more security vulnerabilities. Maybe speed? Perhaps this will be the competitive edge. Whilst data can leak, it probably can’t do it instantly. And so speed of decision could be all that matters.

Then we’ll see the fight that we saw in the early 2000’s as hedge funds battled to have the closest connection to the exchanges.

But what I really want to know is — whose gonna buy me my eSIM for my Spanish Loan?

Which AI will handle the deluge of sidequests that intersect with any buyers journey? Because ultimately, he who beats the side quest beats the competition.

- Dan

Mumpsimus — a custom or notion obstinately clung to however unreasonable it has become

Thanks for reading!