Case Study – How Square used AI

This case study is about how a fintech startup challenged the big banks by lending capital to traditionally underserved small businesses. Using machine learning on its retail transactions dataset, Square Capital can proactively offer small loans to businesses before they even ask for them. Square’s delinquency rate is half of the industry average.

Square and Square Capital

Square is a financial services company focused on the needs of small businesses. It was founded by Twitter founder Jack Dorsey in 2009, and went public in 2015. Based in San Francisco, it had a $26 billion market cap in 2019 and employs 2,300 people. Square’s first product was a credit card reader that plugged directly into a smartphone to allows mall retailers and professionals to accept credit card payments. The reader hardware was distributed for free without the need for a monthly subscription, kick-starting Square’s reputation among small retailers. The hardware’s iconic design and steady flow of referrals also contributed to cement Square’s reputation (international expansion has been limited so far).

In its first years of operation, Square was focusing on a niche of small businesses with less than $125,000 in turnover; these businesses were badly served by banks and conventional payment processors. Square’s low fees and easy-to-use website helped drag entrepreneurs into what was growing into a complete financial ecosystem. In the years leading up to its public listing, Square complemented its services by launching a peer-to-peer payment app, a customer relationship management (CRM) and marketing platform, and point-of-sale hardware for larger merchants.

In 2014, Square launched Square Capital, another component of its financial ecosystem. As small businesses expand, they often develop cash-flow issues that could be relieved with additional working capital. However, conventional banks are often not sensitive to the needs of these customers, as the small amounts don’t justify the over-head involved in marketing and processing the loan. Square Capital uses its point-of-sale transaction data to proactively offer small loans to its customers. For customers,the experience is seamless, as they can accept and manage the loan directly from the same management website that they use for the other services in the ecosystem. Loans don’t have a specific repayment schedule: funds are automatically withheld from the credit card transactions processed by Square. The combination of frictionless experience and existing customer base has allowed Square Capital to expand rapidly: 200,000 merchants have borrowed more than $3.1 billion through the platform in the years 2014–2018 alone.

While credit-risk models have been a staple of the financial industry for many decades, Square has access to a much more extensive dataset, including seasonality and timing of each individual purchase. This allows Square to build models with great visibility into the cash-flow position of each business and appropriately size the offered loan. Together with straightforward repayment options, this pushes the loan delinquency rate down to 4%, half of the industry average.

Criticism and Competition

From a financial standpoint, the loans offered by Square Capital are nothing new. Lenders have been offering merchant cash advances (MCAs) to cover businesses’ short-term cash needs since time immemorial. Compared to MCAs, which don’t have a prescribed time limit, Square Capital loans must be paid back within 18 months.This makes it possible for Square to pass off the loans to a wider array of financial institutions, which have a harder time dealing with MCAs without a known repayment date. Square operates only as an originator of loans, rather than a lender, maintaining an attractive balance sheet that’s comparable to other tech companies.

While most financial institutions use machine learning to evaluate credit risk, the current crop of fintech companies like Square have also been criticized for removing too much human oversight from their credit application processes. For example, peer-to-peer lending startup Prosper came under scrutiny in 2015 after lending $25,000 to the terrorist couple behind the San Bernardino (California) shooting,which left 24 people dead.

While the technology community is excited about fintech being able to disrupt and scale traditional banking through the use of machine learning, some communities are not as optimistic. For example, the very same small businesses that enjoy fast turn-around time for Square Capital loans are also vocal about perceived unfairness and lack of transparency when their loan applications end up rejected. Because Square does not publish any details about its risk-evaluation models, some merchants are left to reverse-engineer algorithmic decisions. From a technical standpoint, this is linked to issues o fboth accuracy (expressed in terms of true/false positives) and bias of the model.

Thanks to its integration with the Square ecosystem, Square Capital can offer attractive features compared to traditional banks and MCAs, which have less direct access to a merchant’s transaction flow. However, other services such as PayPal Working Capital also have access to this source of data, making it harder for Square Capital to stand out. Surveys conducted on small business owners have suggested that they usually choose a lender based on the perceived chance of being approved, rather than on loan amounts or terms. This means that Square Capital might have to focus on user experience and integration with the rest of the ecosystem, rather than optimizing risk-evaluation models.

 

Leave a Reply

Your email address will not be published. Required fields are marked *

New

Winter AI Program

quick contact

Office Hours

Outsource your AI functions to industry-specialized AI experts, so you and your team can dedicate your time and resources to your core mission.

©2020, accessible ai. All Rights Reserved.

WeCreativez WhatsApp Support
Our customer support team is here to answer your questions. Ask us anything!
👋 Hi, how can I help?