Revolutionizing personal credit with machine learning

Avant uses machine learning to open access to credit to thousands shunned by traditional banks

When it comes to personal finance, FICO scores are a key part of the conversation. A simple summary credit score (between 300 and 850), it can determine, for instance, whether one may qualify for that trendy new rewards card or not. For others, it can make a difference in the interest rate they may have to pay on a home improvement loan. For the roughly 100 million Americans with FICO scores under 670 (considered “subprime”), however, it means much more: often the difference between having access to (any) credit at a traditional bank, or not [1][2]. Avant, an online personal loan lender, is out to change that, and machine learning is at the heart of what it does.

Simply put, Avant has used machine learning to create its own credit scoring system, an alternative to FICO scores which are otherwise the industry standard. Particularly, Avant looks far beyond the handful of factors, such as credit history and utilization, that determine FICO scores, instead using advanced algorithms that consider more than 10,000 variables in evaluating a loan application [3]. This helps Avant identify, and serve, customers among the “subprime” FICO pool that otherwise exhibit “prime” behavior (through the additional variables tracked). Furthermore, the data-driven and online nature of the process makes the loan application as easy as “hailing a car with Uber”, with approvals provided instantly or within a business day [4][5]. Since inception in 2012, Avant has thus used machine learning to make >$4 billion loans to 600,000 customers [6].

But that wasn’t without its challenges. In 2016, the very credit scoring model at the core of Avant’s business came under fire for not being as accurate as expected. For instance, ~$300 million loans generated by Avant and sold to investors in April 2016 exhibited 14.5% net losses in first 11 months, significantly higher than the 10.6% projected loss rate [7]. Besides financial loss, this also shook investor confidence, causing the company to slow down lending and focus instead on tightening the credit standards of its machine learning algorithms. Performance improved with loans generated the following year, and the company is expected to generate a profit in 2018 [8].

In the medium term, competition poses a risk as other players develop similar algorithms and catch on to the use of machine learning in banking previously underbanked “subprime” customers. In particular, the incumbents of the personal lending space, traditional banks, may seek to expand into Avant’s market and hence pose a threat. The company has anticipated this risk by instead initiating partnerships with traditional banks, offering its proprietary credit scoring algorithm as a SaaS product under the “Amount” brand [9]. HSBC announced in October 2018 that it will soon start using the Amount platform to offer personal loans of up to $30,000, citing the expansion as “an area of opportunity … outside of our footprint” [10]. The expansion from incumbent banks into this space thus represents both a threat and an opportunity, and active management by the company of this risk will be critical in converting it into an opportunity.

Looking ahead, regulatory compliance is likely to be a key area of focus for the company. Emerging from the 2008 financial crisis, traditional banks cut back on “subprime” lending, tightening access to credit for loan applicants that today form the business opportunity and customer base for lenders such as Avant. Marketplace lenders such as Avant are currently regulated more lightly than traditional deposit-taking banks [11]. However, this is because machine-learning based lending is a nascent industry, and the regulation for it is only now developing. In 2016, the Treasury released a white paper on marketplace lending, requesting information from 28 online lenders including Avant [11]. There is a risk of enhanced regulation in the future for Avant, especially if such marketplace lenders are eventually held to the same standards of consumer protection and financial stability as traditional banks. I would recommend the management to anticipate and prepare in advance for this risk, with similar foresight as it showed with regards to competition by developing Amount and partnering with banks.

Today, the use of machine learning in lending is still largely untouched territory. As the use of machine learning in financial services expands beyond the first movers we are seeing today, I wonder how competition would evolve: particularly, what barriers to entry would exist (if any), and what values among the customer offering would firms differentiate themselves on.

 

(762 words)

 

  1. “US Average FICO Score Hits 700: A Milestone for Consumers”. FICO. https://www.fico.com/blogs/risk-compliance/us-average-fico-score-hits-700-a-milestone-for-consumers/ July 10, 2017. Accessed November 13, 2018.
  2. “What Are the Different Credit Scoring Ranges?” Experian. https://www.experian.com/blogs/ask-experian/infographic-what-are-the-different-scoring-ranges/. Accessed November 13, 2018.
  3. Strauss, Karsten. “Loan Star: Is Avant A Future Billion Dollar Unicorn?” Forbes. https://www.forbes.com/sites/karstenstrauss/2015/04/15/loan-star-is-avant-a-future-billion-dollar-unicorn/#2174543736d0. April 15, 2015. Accessed November 13, 2018.
  4. Crowe, Portia. “This 25-year-old data engineer is helping disrupt the world of finance”. Business Insider. https://www.businessinsider.com/avant-robert-krzyzanowski-on-disrupting-finance-2016-3. March 25, 2016. Accessed November 13, 2018.
  5. Faux, Zeke. “Will This Online Lender’s Risky Business Model Hold Up?” Bloomberg. https://www.bloomberg.com/news/articles/2015-10-29/online-lender-avant-wants-to-become-amazon-of-finance. October 29, 2015. Accessed November 13, 2018.
  6. Avant. https://www.avant.com/about_us. Accessed November 13, 2018.
  7. Daniels, Steve. “Online lender Avant’s reboot still hasn’t yielded profit”. Chicago Business. https://www.chicagobusiness.com/article/20170527/ISSUE01/170529902/online-lender-avant-sees-slow-recovery-after-reboot. May 27, 2017. Accessed November 13, 2018.
  8. Daniels, Steve. “Online lender Avant looks to grow again”. Chicago Business. https://www.chicagobusiness.com/article/20180518/NEWS01/180519863/avant-a-chicago-based-online-lender-shows-signs-of-life. May 28, 2018. Accessed November 13, 2018.
  9. PR News Wire. https://www.prnewswire.com/news-releases/avant-rebrands-its-financial-technology-saas-business-unit-as-amount-300718128.html. July 30, 2018. Accessed November 13, 2018.
  10. Levitt, Hannah. “HSBC U.S. Partners With Web-Based Avant to Offer Personal Loans”. Bloomberg. https://www.bloomberg.com/news/articles/2018-10-22/hsbc-u-s-partners-with-web-based-avant-to-offer-personal-loans. October 22, 2018. Accessed November 13, 2018.
  11. Lux, Marshall and Chorzempa, Martin. “When Markets Quake: Online Banks and Their Past, Present and Future”. Harvard Kennedy School. https://www.hks.harvard.edu/sites/default/files/centers/mrcbg/files/73_final.pdf. April 2017. Accessed November 13, 2018.

Previous:

Fighting Organizational Inertia in Aerospace

Next:

IBM Watson for Cybersecurity Takes Center Court at Wimbledon!

6 thoughts on “Revolutionizing personal credit with machine learning

  1. Interesting topic and very accessible analysis of the competitive landscape for “subprime” lending. I have always felt like the stats that feed FICO scores seemed a bit arbitrary–a better measure of knowing what affects your score than being creditworthy–so it’s interesting to see some evidence that alternative metrics can be effective.

    To your question about barriers to entry and customer offerings, I believe the main barrier to entry in financial services will continue to be building trust, but it may be that technology can help build that trust faster than the long track records that financial institution have needed in the past. I expect that the average person will become more literate in machine learning concepts in the near future, so marketing this approach will become easier, but also much more competitive. Ultimately it may be a single technology company that dominates the lending space with a truly revolutionary advancement in machine learning capability.

  2. Great article! Today, credit scores are imperative to someone’s quality of life in America. They dictate if you can buy a house, go to school and work in a financial institution. However, credit scores are easily hacked and only measure a small part of someone’s past ability to pay, which may mean nothing about their future ability to pay.

    In terms of barriers to entry, I think you hit the nail right on the head. Regulation is one of the most cumbersome parts of the financial industry, and having the money to deal with all of the regulatory barriers is one of the issues barring companies from entering the banking industry.

  3. Interesting article on using machine learning to provide an alternate take on the FICO score. One possible barrier to entry for competitors is having access to the same kind of data Avant uses for predictions, but if the “10,000 variables” actually use a lot of info that can be found publicly/users can easily grant access to (e.g. info on a Facebook profile) then this isn’t likely to be more than a speedbump. Regulation, as cited, is more likely to pose a significant barrier to entry as the industry ramps up. Unfortunately, once competitors get past barriers to entry, I don’t see much the firms can differentiate on – it sounds like Avant is quite similar to traditional banks except for how they determine who to lend to, and the traditional banks don’t have much to compete on except possibly ‘frills’ like good customer service. “Money is a commodity”, and if pricing (aka interest rates) becomes the main differentiator it could spark a race towards lower and lower rates and thus lower margins.

  4. Interesting article here! There are a few other trends that have implications for the future of a fintech startup like Avant. First, I’m interested in getting a bit more under the hood surrounding the actual data sources that they use in their machine learning algorithms in order to do proprietary credit underwriting. Based on my personal experience in the space (working with international banks as an alternative credit scoring vendor, similar model to “Amount”), several key datasources come to mind, including existing transaction and other account data at the consumer’s bank, their digital footprint from social media or their mobile network. With regard to existing bank transaction data, it is unclear how an Avant can maintain its unique value proposition as banks themselves acquire fintechs or develop their own data capabilities. An interesting extension of your essay would have explored financial data APIs and aggregators like Plaid.com and the role that they play in the ecosystem. With regard to seemingly nonfinancial data, it would have been interesting to look at the owners of social media and e-commerce data as potential players in the lending space within the next 5 years. One example is a former competitor of mine that build their entire credit scoring product on Facebook API data and whose entire business model was subsequently thrown into crisis mode when Facebook released a patent on social credit scoring and clawed back access to its graph API for such commercial purposes. Amazon and Google are reported to have similar designs. Another interesting development is regulatory change at the Federal level, where the OCC is working towards creating a new Fintech charter that would allow for non-accredited banks to work nationally for the first time. Loved your essay, its a really fascinating space!

  5. Great article! I think Avant and similar lenders fill an important need in the market that traditional lenders cannot primarily because of scope. The situations that Avant addresses are fringe cases in the lending market and because of this as well as heavy regulations, banks tend to focus their data science efforts on better pricing business within their current target consumer space as well as marginal expansions around this range. It is simply too expensive and inefficient to test and learn in this space and assume charge offs of ~15% that are at least over 3x that of traditional banks’ current books.

  6. Very succinct and easy-to-understand summary of a fascinating business! I find Avant to be such an interesting example of a business that can make a positive social impact while also being a financially viable enterprise. Providing access to a part of the financial system traditionally not available to this group of people enables personal household growth and growth opportunities for financial institutions, essentially growing the pie for everyone.

    However, my concern with Avant is the sustainability of its competitive advantage. The same way that Venmo has seen their competitive advantage eroded with the introduction of Zelle in this space, it feels likely that resource-unconstrained financial institutions will invest in the same technology Avant has pioneered. Avant may be able to protect against this by continuing to improve their AI and using their head start to stay ahead of what other players can create, but this may be challenging over the long term.

Leave a comment