Fintech Insider Sep 21, 2026 1h 1m 45m saved
With Tamara van den Ban, Managing Director of Premier Banking at NatWest Group · Jon Hart, Enterprise Sales Lead at Sage · Alexandra Rivas-Gale, Product Director at Barclaycard Payments
Eighteen AI models were asked 121 financial questions, five times each, and got the answer wrong 57% of the time. On the questions the researchers classed as harder, the error rate was 88%.
The panel's disagreement was not about the number. It was about what the number is being compared with: an advice market that most people cannot afford at all.
"So even if it did make mistakes 57% of the time, if there was a failsafe mechanism or a way for like those mistakes to be made, good, but it meant that the rest of the time people were actually getting advice."
This was the 1,100th episode of Fintech Insider, recorded live in front of an audience at Village Underground in Shoreditch, with a panel that runs premier banking at a UK high-street bank, sells embedded accounting software to banks, and builds payment products for British merchants. Three news stories, one live game, and an audience member who admitted on a microphone that she asks a chatbot which savings account to use.
The full episode is covered here so you can skip it. 61 minutes of audio, 16 minutes of reading.
Here are the 8 arguments that matter.
Key Takeaways
Nubank enters the US with 140M customers behind it and says it will grow on word of mouth — while sponsoring Inter Miami and a Formula One team
Its Brazilian market share is around 60%, and one to two million people in the US are directly connected to that community
The launch breaks Nubank's own playbook: Mexico and Colombia got a credit card alone, the US gets a card, a savings account and remittances
Free transfers home are the part most likely to spread, because the alternative is banking in one place and wiring money in another
Monzo's Aura is the first UK credit card that invests your cashback for you, at £15 a month with perks the bank values above £360
Automated is not the same as intentional — the habit forms, the understanding may not
Aura is built for people who are not rich yet, which one panellist read as fintech aging out of its first customers
AI answered 57% of financial questions wrong, and 88% of the hard ones, with one pension answer worth a £17,500 charge
26% of UK consumers already trust general-purpose AI with financial advice, per the FCA's own research
The danger is not the error rate but the confidence — the wrong answer arrives in the same tone as the right one
1. Nubank Lands In The US
David Brear opened with the Forbes story: Nubank is entering the US, its fourth market, after building a base of 140 million customers across Brazil, Mexico and Colombia. The opening offer is a credit card with no annual fee and 1.5% cash back, a savings account paying 3.5% APY, transfers to Brazil and Mexico, and crypto trading. A stablecoin-based multi-currency account is due next month, with transfers to 35 countries across Latin America.
Kate Moody's first reaction was that the marketing claim does not survive contact with the launch venue.
The word-of-mouth story was announced from a stadium it sponsors
Oh, we're going to entirely rely on word of mouth whilst announcing this from the Inter Miami football stadium
Kate Moody
That was her account of the announcement, made from the ground of a club Nubank sponsors, by a company that also sponsors a Formula One team. The underlying position is strong either way. Moody put Nubank's share of the Brazilian market at around 60%, and estimates of the Brazilian community in the US at one to two million people directly connected, before counting Mexico or Colombia. The company already has provisional sign-off on a US bank charter and is launching alongside a partner bank rather than waiting for it.
Alexandra Rivas-Gale, the American on the panel, said the country will not respond as one market.
The US is several markets with different appetites for a new bank
Some states maybe more in the Midwest are going to really be stuck to the place that they know, somewhere they feel safe, reliable.
Alexandra Rivas-Gale
California, Los Angeles, Miami and New York are where she expects the swing, and she thinks rewards and stablecoin transfers will get attention while the customer experience decides it.
2. Earning A Main Relationship
Brear pressed on the gap between being used and being the primary relationship. Rivas-Gale said that gap is the whole problem, and that it is why the word-of-mouth claim is a bigger commitment than it sounds.
Word of mouth cannot be bought
You can't buy word of mouth word of mouth is people recommend it when it's something that is really solving a problem for them.
Alexandra Rivas-Gale
Her point was that financial services is not what people discuss over dinner, so a product that gets recommended anyway has to be solving something real. One cool feature does not earn a main banking relationship; being reliable, simple and present when it matters does, and it takes time.
Tamara van den Ban came at the same question through the US card market, which she divides into groups. There are what she called the reward hunters, moving from offer to offer; there are people comfortable with brands they already associate with cash back on holidays and shopping; and there is a generational shift underneath both.
The question younger customers ask is what the card builds
Does it help my kind of build wealth or run around and go on holiday or do all of these different things that actually give you the best outcome?
Tamara van den Ban
Multiple cards per customer, each held for a different reward structure, is now normal, which makes entry easier and loyalty harder.
3. The Remittance Play
Jon Hart read the launch as a deliberate break from the way Nubank entered its previous markets, where it led with a credit card and little else.
The money transfers are the part that spreads
if you are banking with a banking provider, but you're having to send your money back with a Western Union or Moneygram. If you can do that all in one place, that is going to spread the word of mouth.
Jon Hart
Moody's supporting evidence came from her own research on the show's home turf: reading back through Monzo's early public question-and-answer sessions, she found the bank reporting a customer acquisition cost of zero in April 2018, when people were sending each other small payments simply because they could. She also likes that Nubank's US offer has no fees and nothing a customer has to remember to cancel after six months, which she traces to the founder's stated focus on removing complexity.
The competitive frame Brear reached for was two heavyweight fintechs arriving in each other's markets at once. Rivas-Gale laid out Revolut's side of it: a full banking licence approved in Colombia that week, what she called fairly substantiated speculation that it will buy a bank in Argentina, and payments licences in Peru.
Both giants are busy in each other's home markets
So it's kind of like whilst they're both of them are distracted chasing new markets, is their competitor going to sort of sneak in the back and pinch share in the market that they had dominance in?
Alexandra Rivas-Gale
She does not expect either to drop its core market, but she is watching for it.
4. Monzo's Auto-Invest Card
The second story, from This is Money, was Monzo's new Aura credit card, which Brear described as the first UK credit card that automatically invests cash back as customers spend. It pays 1% back on groceries and 0.5% on everything else, routed into funds and ETFs through Monzo's investment product. It costs £15 a month and bundles perks the bank values at more than £360 a year, and it sits alongside Monzo's existing Flex Build and Flex Purchase credit products.
Moody's case for it starts from a national problem rather than a product.
Britain does not invest early enough
we know you when you look at the statistics you can see that the UK as a society underinvests. We don't have enough people investing early enough.
Kate Moody
The barrier she hears in customer interviews is not cost, it is not knowing where to start or how much to start with. What she wants to see is whether Monzo builds the follow-on journeys — whether a customer sweeping five pounds of cash back a month is then shown what topping it up would do, and how the money is allocated.
This is newer in the UK than elsewhere. Moody pointed to Trust Bank in Singapore, which pays out in fractional shares or ETFs instead of cash, and said Tiger Brokers is doing something similar. Brear's own test was to ask the room how many people hold a cash-back credit card; a few hands went up, one of them wearing the glow stick as an earring.
5. Automated, Not Intentional
Van den Ban's support for the product came with a condition attached, and she repeated it twice.
Automation builds the habit but not the understanding
automation makes it really simple but it's not always intentional
Tamara van den Ban
What she wants alongside the habit is confidence: whether the customer understands what they are invested in, what it is worth and what the risk is. Her own framing is that spending is not a route to building wealth, so the mechanism has to teach something as it runs.
Brear compared it to hiding vegetables in a bolognese, then asked the question the product is really built on: whether customers who come for the credit and stay for the investing eventually put in more than the cash back. Van den Ban called that the training wheels, and said it works only with the right structures and guidance around it.
Rivas-Gale, a Monzo customer, wanted the same idea without the credit product attached.
She would rather this sat on a debit card
I would love to have seen something like this without you having to take out a credit product
Alexandra Rivas-Gale
She called Monzo's current debit cash-back offer poor, pointed again at Trust Bank in Singapore, where customers choose between cash back, stock back and air miles, and noted that Nubank offers unlimited cash back in its own markets. Brear's answer was that you have to move to Mexico for that one.
Van den Ban's related observation is that Monzo has not made the auto-invest a choice. Elsewhere, dynamic cash back is a feature the customer selects; here it is the default behavior of the product, and what that does to customer behavior is the thing worth watching.
6. Built For Henrys
Hart was blunt that the card is not aimed at him, and read the design as a statement of who it is for.
The perks are sized for someone on the way up
the inclusion of airport lounges feels a little bit kind of jet setter, but you only get two of them a year because you're not earning enough money yet to use anymore.
Jon Hart
He would not move his American Express or his stocks and shares over, but he thinks the behavioral design is clever and the long game is real: get people at the start of their investing lives and keep them. His analogy was the piggy bank a high-street bank gives to children in the hope they turn into customers.
Rivas-Gale said the analogy holds because it worked on her: she banked with HSBC for a decade because of something she was given at school. Then she named the target.
The customer is a high earner who is not rich yet
They're high earners, not rich yet, and they're waiting until they start to build that wealth.
Alexandra Rivas-Gale
By the time that customer has money, she said, Monzo will have a full catalog of products to serve them from one app. Brear found the implication less comfortable: fintechs are no longer aiming at the people who grew up with them. Moody put the same point historically — the industry spent years being told banking did not serve millennials, built products for them, and is now facing a generation whose language she does not recognize.
7. AI Gets 57% Of It Wrong
The third story, from IFA Magazine, was research by the financial technology firm Saturn, which tested 18 popular AI models against 121 financial questions and put more than 10,000 questions through them in total. The models were wrong in 57% of answers on average, rising to 88% on the questions classed as harder. The failures ran from calculation errors and missing risk warnings to out-of-date tax information and rules that do not exist. One wrong answer about a pension tax rule, the report estimates, could have left the consumer with a £17,500 charge.
Moody's response was to ask what the alternative is.
The comparison is not with good advice, it is with none
the alternative to AI that we had before was that so many people just didn't get any kind of financial support full stop
Kate Moody
Eight years of customer research at 11:FS, she said, kept producing people who wanted advice and guidance that the regulated industry had no commercial reason to give them. Something has to give; AI is part of the answer and is not right yet.
Rivas-Gale did not soften it.
The error rate is frightening because of what it costs
It's not as if you're buying the wrong book and it's like, well, too bad throw it away. It might be your future that's at risk.
Alexandra Rivas-Gale
What alarms her more than the error rate is the usage rate. The show cited figures putting one person in three at a weekly use of these tools for their finances, and the argument from the panel was that banks should step into that gap with guardrails rather than leave it to a general-purpose chatbot.
Brear then went to the room and found a live example: an audience member who uses a chatbot for basic research before doing her own deeper work, picks her own stocks, but asks it which savings account to put money into. She said it had not steered her wrong yet, which Brear noted is a different claim from not having been steered wrong.
8. Trust And The 26%
The savings-account question is the one Rivas-Gale wanted to take apart, because the tool cannot know what it is being asked.
"Best" is meaningless without the rest of the customer's life
It could give an right answer for one person, but it might be the wrong one for the other 99.
Alexandra Rivas-Gale
A model can summarize factual information that exists on the internet, she said; what it cannot do without proper modeling, safeguards and a view of the whole situation is decide what is right for one person.
Brear brought the trust question back to something the industry already did to itself, comparison sites that ranked whoever paid the most at the top, then gave the FCA's own figure: 26% of consumers trust general-purpose AI tools for financial advice.
Moody's answer is that the judgment has to stay with the reader, the way it does with a news article written from someone's perspective, and that the confident tone is what makes that hard.
A confident wrong answer reads exactly like a right one
it'll answer it so confidently. It will tell it to you and it'll be absolutely sure of that answer and you'll read it. You go, "Oh yeah, brilliant stuff."
Kate Moody
Her line for where the boundary sits: indecision may be worse than no decision, but a wrong financial decision is an expensive mistake, and nearly right is wrong. Brear's version was that the models are built to please — he wants one that tells him to pay off the mortgage and stop buying shoes rather than recommending another credit card.
Van den Ban described what a bank doing this responsibly looks like from the inside: experimenting heavily, a research office specializing in responsible use, starting with spending insights that customers can ask questions about in plain language, and, as she noted at a live event in London, models that can handle Scottish, Irish and northern English accents.
The human stays, on both sides of the product
it's not AI replacing the human. It's actually how do you combine the strength of both
Tamara van den Ban
That includes pointing the technology at the bank's own staff so they can serve customers faster, with human oversight on the output, because the consequences land on someone's money and their children's.
Rivas-Gale closed the story on the product side. She called AI one of the best things to happen to how software gets built, with faster code, faster shipping and products in customers' hands sooner, and said all of it has to stay anchored to the same test.
The grounding principle does not change
our number one goal is to keep our customers secure, give them a resilient platform and ensure that their money is always safe
Alexandra Rivas-Gale
Her optimistic case is that fine-tuned models put financial information in more hands and make investing less frightening, which matters precisely because so few people get help at all.
Bonus Insights
The closing game was real headlines against invented ones
Brear ran an "and finally or artificially" round: six strange stories, glow sticks up for real and down for AI-generated. Real, and the audience mostly got them: a scientist who opened an OnlyFans account to fund marmot research and later got a crypto boost; Pope Leo's US bank hanging up on him because staff thought it was a prank call; an app called Are You Dead climbing the Apple charts, which the panel thought sends for help if you stop replying. Invented: a buy-now-pay-later firm offering instalment plans on parking fines, a digital bank letting joint-account holders round up purchases into a divorce fund, and a vending machine running an AI wellness check before dispensing caffeinated drinks. Brear said he would buy the parking-fine product.
Rivas-Gale's best line of the night
On the gap between what AI can do and what it cannot: "is it reassuring that I find that it's funny that an AI has greater than 10% chance of killing us but still can't figure our pensions rules."
The 1,100th episode, and an anniversary nobody planned
Moody's eight-year anniversary at 11:FS fell on the same day, which nobody had arranged. Hart's route into fintech was listening to this podcast at about episode eight, to prepare for a job interview at a personal-finance app. Rivas-Gale had been on before, from a different company with a different title.
Why an expat tax position pushes people to a chatbot
Rivas-Gale said her own situation as an American living in the UK is complicated enough that typing questions into a chatbot is the easy thing to do, which is exactly how people end up treating a fluent answer as a checked one.
The panel's bottom line is that the error rate in the research is not an argument against AI in financial advice so much as an argument about what it is replacing: Moody thinks a flawed answer beats the nothing most people currently get, Rivas-Gale thinks the consequences of money advice make that trade different from any other, and van den Ban's position, which is to automate the habit and keep a human on the outcome, is the one the room did not argue with.
Products, Companies & Tools Mentioned
Nubank (Entering the US as its fourth market with 140 million customers, a 1.5% cash-back card, a 3.5% APY savings account and transfers to Brazil and Mexico)
Monzo (Its Aura card auto-invests cash back into funds and ETFs at £15 a month, alongside Flex Build and Flex Purchase)
Revolut (Chasing the US and Latin America at once — a full banking licence in Colombia, payments licences in Peru and speculation about an Argentine bank)
NatWest Group (Van den Ban's employer, experimenting with AI spending insights under a research office focused on responsible use)
Sage (Hart's employer, which embeds accounting and payroll into banks' own apps so business customers stay in one place)
Barclaycard Payments (Rivas-Gale's employer, a UK payments acceptance business she has joined to run product transformation)
11:FS (Makes the show; Brear is its CEO and Moody its customer strategy director)
Saturn (The financial technology firm behind the research that put 18 AI models through 121 financial questions)
ChatGPT (The tool the audience member uses to research savings accounts, and the kind of general-purpose model the research tested)
Trust Bank (Singapore bank that lets customers take cash back as fractional shares, stock or air miles — the comparison Rivas-Gale wants Monzo to meet)
Tiger Brokers (Named alongside Trust Bank as an experiment with the cash-back mechanism)
Western Union and MoneyGram (What Nubank's free transfers home are competing with, and why Hart thinks remittances carry the launch)
American Express (Hart's current card, which Aura would not displace; Brear said he would switch away from his)
Inter Miami (Where the word-of-mouth launch was announced, alongside a Formula One sponsorship)
HSBC (Kept Rivas-Gale for ten years on the strength of something she was given at school)
Financial Conduct Authority (Its research put consumer trust in general-purpose AI for financial advice at 26%)
Books & Resources Mentioned
Nubank comes to the U.S. with a big bet on word of mouth – Forbes (The first story, and the show's source for the launch detail)
Monzo's new Aura credit card comes with AI subscription perk and cashback – This is Money (The second story)
ChatGPT and Claude get financial advice wrong 57% of the time – IFA Magazine (The write-up of Saturn's research that the third story came from)
Listen to the full episode
🔴 YouTube | 🔗 Episode page
Watch the full episode:
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: