Why artificial intelligence still needs a human touch

This article is an on-site version of our #fintechFT newsletter. Sign up here to get the newsletter sent straight to your inbox every Monday

Using artificial intelligence to improve fraud detection is becoming one of the hottest trends in the insurance industry, but also one of the most contentious.

US insurer Lemonade has quickly become a case study of the potential rewards and reputational risks of the technology. 

Lemonade’s promise to speed up and simplify renters’ insurance and home insurance with an AI-powered app helped make it one of the most successful large IPOs of 2020. But earlier this year it provoked a social-media backlash amid concerns over how its algorithms operate.

The company’s app uses text and video to collect information from consumers. “No forms to fill out. Just speak to the camera,” says its promotional video, describing the process of making a claim.

The model has certainly wowed investors. After a big one-day share price “pop” at listing last year, it has grown to a near-$6bn valuation. 

However, a Twitter thread from the company’s account in May provoked concerns about what exactly is being recorded and analysed when customers “just speak to the camera”.

Lemonade tweeted that its AI would scour customers’ claims videos for indications of fraud, picking up on “non-verbal cues”. The thread alarmed those who have warned about the biases in many AI systems.

After the criticism, Lemonade deleted the “awful” thread. Noting the “biases [of AI] across different communities”, it said in a blog post that it did not let AI automatically reject claims and said it has never used phrenology or physiognomy (assessing someone’s character based on their facial features or expression).

The company said it was not possible, ethical or legal “to deduce anything about a person’s character, quality, or fraudulent intentions based on facial features, accents, emotions, skin-tone, or any other personal attribute.” The term “non-verbal cues” was simply used to describe its facial recognition software, Lemonade said, designed to catch the same person making claims under different identities. 

As AI is put to work across the insurance sector, Lemonade is far from alone in trying to use it to detect fraud. 

Root Insurance, a motor insurtech that also listed last year, uses machine learning throughout its operations. 

Root can point human claims handlers toward discrepancies, such as comparing data it has gathered with a customer’s description of when and where an accident took place. Like Lemonade, however, Root says it never declines a claim based entirely on AI. 

Then there are third-party providers such as France’s Shift Technology, which offers insurers a fraud detection service that has analysed hundreds of millions of claims since it launched in 2014. 

Using information gleaned from the customer and its wider data pool, Shift can identify, for example, if the same person had a similar claim elsewhere, if the same photo is being used in multiple claims, or if a picture was taken prior to the supposed insured event. 

A generous interpretation of the Lemonade episode would be that it was simply a poorly-worded piece of marketing.

But some think there are bigger questions to be asked about how the sector is using voice and video data. Duncan Minty, an independent consultant on ethics in insurance, says insurers are “very opaque in relation to what they’re actually really doing with technology”.

Eiopa, the EU’s insurance regulator, last month recommended that insurers make “reasonable efforts to monitor and mitigate biases from data and AI systems”.

“Insurance firms should develop their approach to fairness and keep records on the measures put in place to ensure fairness and non-discrimination,” Eiopa said in a report from its working group on digital ethics.

It concludes that human oversight is key when it comes to claims. Automated processes may bring benefits such as reduced costs and better consumer experiences, according to the paper, but can also lead to a “refusal of a legitimate claim, inaccurate repair estimates [and] negotiation of unfair settlement prices for damages”. In detecting fraud, human handlers can ensure “sensible” resolution of disputes and effective monitoring of how well the AI is working.

It is clear that the data we create in every connected moment is now a powerful vetting device for insurers. But companies are likely to face growing pressure to reassure customers, and regulators, that the rules guiding speedy and frictionless processes are fair to all.

Fintech fascination

More stories from the industry that caught our eye this week

Wise becomes UK’s largest-ever tech listing The fintech formerly known as TransferWise completed its long-awaited public float this week. It became the first tech group to pursue a US-style direct listing in the UK — meaning it started trading on the London Stock Exchange without selling any new shares. The UK government has been trying to encourage more fast-growing companies to join the London market, with mixed success. Food delivery app Deliveroo was described as “the worst IPO in London’s history” earlier this year. Fears that Wise might be another high-profile flop were allayed when the company started trading at a valuation close to £8bn, and the stock has only climbed further since. Some are hoping its success will encourage other fintechs to follow suit, though one person who worked on the direct listing warned that UK investors remain wary about “pre-profitability” companies, which includes most of the UK fintech sector.

Herds of unicorns The term “unicorn” used to describe private companies worth more than $1bn was supposed to be an indicator of their rarity, but a combination of surging demand for digital services and a proliferation of deep-pocketed investors means the mythical ungulates are easier to spot than ever. The total includes growing numbers of fintechs — Dutch digital bank Bunq and Danish corporate card provider Pleo were the latest to join the stable this week. Pleo chief executive Jeppe Rindom talks more about the business in this week’s Quick Fire Q&A, below. For more on the broader trend, go deeper with this story from FT venture capital correspondent Miles Kruppa, and this column from our west coast editor Richard Waters.

Trussle hopes for Better days after acquisition Regular #fintechFT readers will know it has been a busy few weeks for mergers and acquisitions in the European fintech space. Most of the deals have been celebrated as a sign of an industry maturing, but not everyone has emerged a winner. The FT reported last week that Better, the US online mortgage lender, is buying British mortgage broker Trussle. That’s good news for Better, as the SoftBank-backed group marks its first international expansion. But on the other side of the table, Goldman Sachs will receive a fraction of the money it invested in Trussle three years ago after the broker’s valuation tanked.

Crypto Corner Investors in cryptocurrencies and special purpose acquisition vehicles have gone through similar experiences recently. Rapid growth helped make some people very rich, attracting celebrity endorsements on the one hand and regulatory attention on the other. The limelight has led to some precipitous price falls of late, but enthusiasts haven’t been put off, and this week saw several crossovers bringing the two groups together: stablecoin company Circle announced a combination with the blank cheque company chaired by former Barclays boss Bob Diamond, and a day later Peter Thiel’s crypto group Bullish Global said it would merge with a spac led by the former president of the New York Stock Exchange.

Quick Fire Q&A

Stay up to date with up-and-coming disrupters. Each week we ask a fast-growing fintech to introduce themselves and explain what makes them stand out in a crowded industry. This week we spoke to Jeppe Rindom, founder of corporate card specialists Pleo.

When were you founded? Pleo was founded in Copenhagen in 2015.

Where are you based? Our main office is in Copenhagen, Denmark, but we also have offices across Europe in London, Stockholm, Berlin and Madrid.

Who are your founders? Myself and my colleague, Niccolo Perra, founded Pleo after meeting at our previous company, Tradeshift.

What do you sell, and who do you sell it to? We are an out-of-the-box spending solution offering smart payment cards, automated expense reports and invoicing to businesses and their employees.

How did you get started? As a former CFO, Pleo was born out of my frustrations surrounding centralised spending, procurement teams, and outdated financial tools.

How much money have you raised so far? Pleo has raised $228m to date.

What’s your most recent valuation? Following our recent $150m Series C investment round, Pleo is now valued at $1.7bn.

Who are your major shareholders? Bain Capital Ventures, Thrive Capital, Creandum, Kinnevik, Founders, Stripes and Seedcamp.

There are lots of fintechs out there — what makes you so special? We’re passionately customer-centric in our approach and believe in bettering the spending experience by empowering employees with the right tools.

Recommended newsletters for you

#techAsia — Your guide to the billions being made and lost in the world of Asia Tech. Sign up here

Trade Secrets — A must-read on the changing face of international trade and globalisation. Sign up here

Most Related Links :
newsbinding Governmental News Finance News

Source link

Back to top button