Small Businesses Turn to AI Lending Without Human Oversight - Quartz
Small Businesses Turn to AI Lending Without Human Oversight – Quartz

When Xiomara Rosa-Tedla needed a small loan in February 2020 to fund her e-commerce startup Unoeth, she didn’t contact a venture capitalist or bank officer. She asked for an algorithm.

Rosa-Tedla founded Unoeth with her father in 2015. The company sells leather handbags and other handmade accessories in Ethiopia, and Rosa-Tedla was sitting on a backlog of unsold goods. She needed a few thousand dollars to buy ads on Facebook and Instagram so she could show her products to the right customers and sell her supply.

Rosa-Tedla didn’t want to sell part of her family business to a venture capital firm or take the steps to get a loan from a bank where, she jokes, you have to “give up your house, your car, and your firstborn” as collateral. Instead, she downloaded an app from a loan company called Clearco and gave it access to sales, revenue and traffic data from Unoeth’s website on the Shopify e-commerce platform.

Within minutes, an artificial intelligence algorithm analyzed Rosa-Tedla’s activity and presented her with three financing offers: one for $5,000, one for $10,000 and one for $18,000. She took the $5,000 offer and had the money in less than two days, never negotiating with a human.

Clearco is part of a growing industry that offers loans to small businesses, especially e-commerce startups, with virtually no human intervention. In 2020, Clearco distributed $2.5 billion in funding to 5,500 companies, relying entirely on algorithms to decide which companies to give money to, how much to offer and what the terms of the deal should be. .

Payments and e-commerce companies join the debt party

E-commerce platforms like Shopify and payment companies like PayPal and Square now also offer their own versions of the service. These companies have mountains of granular data about their customers’ activities, which allows them to train algorithms to predict which companies are safe bets to lend money to and which are riskier investments. The algorithms can then adjust the terms of the deal accordingly, increasing the cost of capital to account for the greater risks. (Clearco says humans never review its algorithms’ decisions, but Shopify, PayPal, and Square employ human reviewers to review certain offers above a certain size.)

AI-powered lenders say they fill a funding niche not served by venture capitalists or traditional banks: they offer rapid injections of capital under a sharing model revenue that doesn’t require founders to give up equity, cultivate personal relationships with members of Silicon Valley’s elite, or go through bank due diligence processes. Lenders like Clearco also say their funding method can distribute investments to a more diverse set of founders overlooked by traditional funding methods. Some founders say AI-approved loans give them a faster, easier way to cover day-to-day operating expenses and grow.

AI loans expand their footprint

AI lenders are attracting a growing number of customers. Shopify Capital has loaned $2 billion since its launch in 2016, and half of that total was loaned out last year. Square Capital says it has lent $9 billion since its launch in 2014. Clearco’s total lending skyrocketed between 2017 and 2020, and the company now expects to lend more than $1 billion in 2021.

While these numbers pale in comparison to US venture capital ($130 billion invested in 2020) or small business loans ($23 billion issued in 2019), AI lending is growing much faster than the either of these two more traditional models.

Zavain Dar, a partner at venture capital firm Lux Capital, says the rapid growth of AI lending platforms is a promising sign. “If you look at the traction these [AI lending] companies got it shows that there was a need for some form of financing where it wasn’t a loan against your house to create a bodega, and it wasn’t ‘I’m going to go build a top 100 billion-dollar venture tech start-up,” he said. “A lot of companies needed something in the middle, and the market had overlooked that big middle.”

AI decides differently than humans

The algorithms determining which companies are funded are designed to be narrow-minded. While human bank officers or venture capitalists can make lending decisions based on who started a business, where they went to school, or what kind of products they sell, companies that deploy these algorithms say they only consider a limited set of sales data. Square Capital says its model considers only a handful of data points, including “processing volume, payment frequency, and customer mix.” Clearco’s model primarily looks at sales figures, but it also takes into account data on a company’s margin profile (a measure of the profit that sellers make on each sale), sales growth and the number customers who browse the company’s online store each month.

“We wanted to start from the first principles of what we thought would make an e-commerce business work,” said Clearco President Michele Romanow.

At first, the Clearco team guessed what a successful e-commerce business looked like: they coded their algorithm to only lend money to businesses that met a certain threshold of sales, profit margins, and traffic. website. Their initial guesses turned out to be pretty ugly. “In our early cohorts, we were losing about 20% of our money,” Romanow said. But over time, Clearco used data from past transactions to train a machine learning model to come up with its own rules, and it has since gradually refined the algorithm. The AI ​​examines the results of its past lending decisions to learn to spot patterns and more reliably predict which companies will be able to repay their debts.

After the first year, the AI ​​has improved enough to make a steady profit on its loans.

Repay loans by revenue sharing

A key difference between algorithmic lenders and other lenders is how loans are repaid. Banks and credit card companies typically charge monthly interest payments, while venture capitalists take an equity stake in a company. But AI lenders are focused on revenue sharing: Companies repay some of their debt each time they make a sale.

The terms vary but almost all work the same way. Lenders give a business a lump sum upfront, say $10,000. Then the company gradually pays that amount back by giving the lender a small share of each sale made, which can range from 1% to 20% of each sale, depending on the terms of the agreement. If a business makes a sale for $50, it will need to send the lender between 50 cents and $10.

The startup continues to repay the lender bit by bit until it has repaid the initial amount, plus a lump sum, which is typically something like 6-12% of the amount the company borrows, depending on the agreement. In this example, the startup would end up paying between $10,600 and $11,200.

The ideal recipient has a low sales volume. The faster a company sells its products, the faster it pays off its debt. A company with very strong business can repay its loan, plus a flat fee of 6%, in one month. But paying a 6% fee in one month equals an annualized interest rate of 72%. In this scenario, take out a small business loan (typical interest rate: 3-7%) or even have credit card debt (typical interest rate: 15-18%) could be cheaper.

But the AI ​​approach has other advantages. None of the AI ​​lenders report transactions to the credit bureaus, which means that if a startup doesn’t repay their loan in full, it won’t affect the owner’s credit. They also do not require any form of collateral, which means that if a business goes bankrupt, its owners can get out of their debt without penalty.

Can AI funding reach diverse entrepreneurs?

Self-reported data from Clearco seems to suggest that AI loans can help reduce some of the biases that prevent women and people of color from accessing traditional forms of finance. Venture capital and small business lending are marked by clear racial and gender disparities: His harder for women and people of color to get investments. When U.S. lawmakers approved $659 billion in emergency small business loans to help businesses survive the pandemic, 83% of loans went to white-owned businesses, compared to just 2% for black-owned businesses. Fair 1% of US venture capital goes to black-owned startupsaccording to data from risk-tracking firm Crunchbase.

On the other hand, Clearco announced in April that 13% of its funding went to black or Latino founders, well above those who receive funding from banks or venture capitalists. Clearco also claimed that it funds “eight times as many female founder-led companies as it does traditional venture capital firms” and that the majority of its funding goes outside of the traditional tech hubs of California, New York, of Texas and Massachusetts which typically absorb the lion’s share of venture capital funding.

The findings earned cautious applause from Jeanna Matthews, a computer science professor at Clarkson University who studies the ethics of AI systems. “If you see the impact of a deployed system and you see it helping them avoid bias, that’s a good sign,” she said. But Matthews cautions that the data does not guarantee that AI lenders’ algorithms are free from bias. Even if they excluded data on the identity of the founders or the nature of their products, innocuous data points such as sales and revenue figures could become proxies for the identity of the founders. “Often the bias is in the data even with those columns removed,” Mathews said, “and if you’re not careful, you can end up rediscovering those same data points through proxy variables.”

Ultimately, Matthews says, perhaps the best thing about AI lender algorithms is simply that they’re able to process more funding applications from more business owners faster than anyone else. what a human. As a result, they are able to accept funding applications from virtually anyone, anywhere.

“Maybe what we’re saying is it’s better to say yes to a lot of people,” she said. “Maybe there are great ideas from women, people of color, people from many states, that no one was picking up on before, so when you say yes to people in those areas, you get value that d ‘others haven’t.’


Please enter your comment!
Please enter your name here