Europe’s fintech sector is rethinking its enthusiasm for generative AI as some of its most vocal proponents acknowledge the technology’s limitations in financial services.
The AI agent vertical has attracted VC investors in their droves in the last year. They’ve allocated close to €2bn in capital to startups building AI agents, including those specialising in the financial services. But a reversal in attitude towards the technology’s potential within fintech seems to be underway.
After saying that the technology was ready to do the work of 700 customer service agents at Klarna, in May CEO Sebastian Siemiatkowski admitted the company had relied too much on automation.
“As cost unfortunately seems to have been a too predominant evaluation factor when organising this, what you end up having is lower quality,” he said at Klarna’s Stockholm HQ. “Really investing in the quality of the human support is the way of the future for us.”
A spokesperson for Klarna rejected suggestions the company was dialling down its reliance on AI, telling Sifted: “What we’ve learned is that as AI takes care of the majority of interactions, the small number of cases that require a human become even more important.”
Last month fintech conference Money 20/20 also went back to basics a year on from a big AI love-in, signalling a broader shift in the sector about the technology’s potential in such a regulated market.
“Frankly, a lot of companies got caught up in the hype. Klarna moved fast, but possibly too fast to fully test the downstream impact, especially in customer-facing areas,” says Gabriel Le Roux, the CEO of ecommerce payments fintech Primer. ”Across fintech, there’s been a rush to roll out AI, often driven by cost-cutting or the pressure to ‘look’ ahead of the curve.”
Changing tune
In the years following the release of OpenAI’s ChatGPT, in 2022, there was a rush by fintechs to roll out AI products. In 2023, London-based compliance fintech ComplyAdvantage launched an AI tool to identify and curb fraud. Chief execs at neobanks such as Monzo also waxed lyrical about the technology’s potential to assist in fraud prevention and make internal processes smoother.
Regulators — typically more conservative in approaching emerging technologies — were also keen.
“It’s something that banks and financial institutions are really investigating,” Jessica Rusu, chief information data and information officer at the Financial Conduct Authority, which regulates the country’s financial services industry, said on stage at Money20/20 last year. “Chatbots are going to be really good for any kind of exchange of information between the customer and the business.”
Now those claims are being reassessed. UK neobank Starling’s chief information officer, Harriet Rees, tells Sifted it’s important for the company to retain a human-run customer service division, that runs 24/7 and is free to customers.
And while it’s developed a way to use AI to summarise post-customer interactions for agents, there’s no plans for AI agents to take over from humans yet.
“We’re definitely not going to be handing over any key decision making to any AI models at this point,” she says.
Cost-cutting
Rees’s comments underscore the divergence in AI adoption between fintech and other industries, in which companies have kept their teams super lean by leveraging the technology to do tasks that once required a larger workforce.
That’s difficult to do in the heavily regulated financial services industry, where regulators such as the FCA have in recent years introduced rules requiring fintechs to prove they’re acting fairly and delivering “good outcomes”.
Industry watchers say this means it’s unlikely we’ll see a day where AI can make decisions in financial services without human oversight. It also has slowed the pace of innovation in the sector, as companies have to be mindful of the impact any customer-facing AI tool might have on them.
“It’s really important the human element remains absolutely there,” Rees says. “We won’t compromise on that.”
“Klarna may have shaved costs initially, but maintaining customer satisfaction and trust ultimately required rehiring humans,” says Elina Rayberg, an investor at US VC Valar Ventures, which isn’t an investor in Klarna. “The short-term AI efficiency gains weren’t enough to offset reputational or customer experience risks.”
Rayberg also believes AI isn’t reducing costs in fintech, it’s just shifting them.
“Building safe, compliant and reliable AI systems at scale requires significant investment in model infrastructure, human escalation teams, data governance and regulatory tooling,” she says.
According to a May IBM survey of 2,000 CEOs globally, only 25% of AI initiatives have delivered the expected return on investment over the last few years.
And while those investments may pay off and increase efficiency in the long-term, fintechs viewing AI as a way to cost-cut in the short-term are making a mistake, says Starling’s chief financial officer Declan Ferguson.
“You can’t cost cut your way to growth or profitability,” he tells Sifted.
Product development
That’s not to say there hasn’t been any AI innovation in fintech in recent years. Starling last month launched Spending Intelligence, which allows users to ask questions about their money, such as “how much did I spend on groceries last week?” but stops short at encouraging customers to take action based on this information.
“We’re not trying to give you advice,” says Rees. “We’re trying to give you the knowledge you need to make better financial decisions and ultimately help you to be good with your own money.”
Dutch neobank Bunq and Nordic neobank Lunar have also rolled out consumer-facing AI financial assistants in recent years. Fellow UK neobanks Zopa and Revolut are also developing similar tools slated for release later this year.
The past month has seen fundraises from the likes of Gradient Labs, a startup creating AI agents for the financial industry founded by former Monzo employees, and Barcelona-based Murphy, which creates AI agents for debt collection. Revolut is also currently hiring for an applied AI lead who will help its team to build and deploy AI solutions, such as “AI assistants and AI automation platforms”.
Still, the pace of product development is slow. Le Roux tells Sifted the company took months internally testing its AI sales assistant Tessa to ensure it had product-market fit. The tool, which can handle cold outreach, respond to inbound sales questions and book meetings, is still in its pilot phase.
“It only works because we’ve been careful about how and where we deploy her,” he says.
As both Le Roux and Valar’s Rayberg point out, a mistake by an AI agent could prove fatal to a fintech company, undoing years of brand building and possibly exposing the company to regulatory issues.
“AI takes time to train, test and make sure it actually fits your product, your users and your brand,”says Le Roux. “Especially in customer experience, the bar is high for a reason. Trust takes years to build, and seconds to lose.”
Read the orginal article: https://sifted.eu/articles/fintechs-ai-limits/
