Machine Learning: The end for America’s Most Hated Company?

AI enabled alternative credit scores: A threat to Equifax?

Machine Learning: The end for America’s Most Hated Company?

 

What do low cost airlines, The Weinstein Company, and for-profit colleges have in common? Unfortunately for Equifax, this is not the beginning of an off-color joke. In 2018, Equifax beat each of these companies out for the dubious honor as America’s most hated company [1].

 

Years of consolidation in the credit monitoring space has left consumers with little choice but to accept the status quo when it comes accessing credit. Three companies, (Equifax, Transunion, and Experian) maintain credit histories on approximately 200 million consumers and process nearly two billion items of information each month[2]. Traditionally, a proprietary linear regression is applied to the data in this file to come up with a FICO score, which is then used in conjunction with the file, as the basis for lending decisions. This consolidation of power has been good for business, while perhaps bad for consumers. As of this writing, Equifax ($12.1B)[3], TransUnion ($11.93B)[4], and FICO ($5.5B)[5] have a combined market capitalization over $29 Billion.

 

How can a company, and an industry with such low satisfaction in the eyes of consumers be so profitable? My assertion is that the cost of gathering and maintaining the data files provides an effective moat, protecting the entrenched players from competition from innovative challengers. To this point, in 2017, Equifax had $1.48 Billion Cost of Goods Sold on $3.36 Billion Revenue[6].

 

As data has proliferated, and computer processing power has become dramatically cheaper, Machine learning techniques pose an existential threat to Equifax. With alternative scoring methods, the cost of entering the market as a low-cost competitor has decreased to the point of eliminating the moat around a tired and stodgy business.

 

Again, unfortunately for Equifax, they seem to fail to understand the magnitude of the threat. Instead of revamping the entirety of their operation, they are making small and incremental steps to improve the process that they already use, without regard to the customer experience, cost, or satisfaction levels. Conventional wisdom holds that there are “two main reasons to use artificial intelligence to derive a credit score. One is to assess creditworthiness more precisely. The other is to be able to consider people who might not have been able to get a credit score in the past.”[7] In this vein, “using machine-learning techniques has led to a performance lift of as much as 16.6 percent for bank card originations and 12.5 percent on auto originations of consumers with dormant credit files.”[8]

 

Machine learning drives these gains by considering many variables simultaneously, looking for nonlinear interactions instead of considering one variable at a time.[9] These increases are without a doubt a benefit to Equifax, and demonstrate the power of using machine learning in assessing credit risk. However, celebrating these gains without restraint, would be much like Blockbuster celebrating a commensurate increase in DVD rental income as Netflix and Hulu built streaming businesses. A quick review of CB insights reveals that funding for AI startups reached record highs in 2016 and include businesses across the spectrum of the financial services industry. Of the top 100 startups ranked by CB insights, 14 were in the credit scoring or direct lending space.[10]

 

Rebuilding the moat

 

Equifax should recognize the barbarians at the gate and look inward. As cost to collect data decreases, they will have to compete on consumer satisfaction. Where do they have a sustainable competitive advantage and what has driven consumers to their abysmal approval rating of their performance? Before a startup can steal market share, I would refocus on the customer experience. For existing customers, I would use the informational advantage from the existing data set to provide the most accurate scoring possible, while using AI to find and correct errors in users files. I would grow by addressing the underbanked population in the United States, while establishing goodwill by displacing predatory payday lending services.

 

Open Questions

 

  • What areas is a startup most likely to pry into Equifax business? (I.E. are auto, payday, mortgage, consumer, or education loans most susceptible?)
  • What is Equifax’s biggest competitive advantage in the face of consumer sentiment and decreasing costs of accessing data?

 

Word Count: 776

 

[1] https://www.usatoday.com/story/money/business/2018/02/01/bad-reputation-americas-top-20-most-hated-companies/1058718001/

[2] Smith, L Douglas, PhDStaten, Michael, PhDEyssell, Thomas, PhDKarig, Maureen, MBAFeinstein, Jeffrey, PhD; et al. Financial Services Review; Atlanta Vol. 27, Iss. 1,  (Spring 2018): 1-28

[3] https://finance.yahoo.com/quote/EFX/?p=EFX

[4] https://finance.yahoo.com/quote/TRU/?p=TRU

[5] https://finance.yahoo.com/quote/FICO?p=FICO&.tsrc=fin-srch

[6] https://www.marketwatch.com/investing/stock/efx/financials

[7] Crosman, Penny.ISO & Agent; New York Vol. 6, Iss. 2,  (Mar 1, 2017): 42.

[8] VantageScore White Paper Highlights VantageScore 4.0’s Use of Machine Learning Manufacturing Close – Up; Jacksonville (Dec 16, 2017).

[9] Crosman, Penny.ISO & Agent; New York Vol. 6, Iss. 2,  (Mar 1, 2017): 42.

[10] https://app-cbinsights-com.prd1.ezproxy-prod.hbs.edu/research/ai-fintech-startup-market-map/

Previous:

Additive Manufacturing in Medical Device: Can Johnson & Johnson stay ahead of the curve?

Next:

Can Machine Learning predict, prevent and diagnose Diabetes effectively?

Student comments on Machine Learning: The end for America’s Most Hated Company?

  1. I wonder if there is an ethical reason why machine learning is not used to generate a credit score. The company would have to explicitly forbid the algorithm from taking things such as gender, race, height, and other personal information into account when doing the calculation. Further, it would seem that Credit Scores have to be easily explained. I wonder if the “black box” nature of some machine learning algorithms would make this difficult.

  2. I think the challenge in the credit score space is that the end consumer is not the individual’s being rate but the business/banks/3rd parties who want to assess the credit worthiness of an individuals. To successfully serve that end business, you’d need a huge amount of coverage (e.g. if you can only use machine learning to measure credit scores for a small fraction of the population, then it’s less useful) as well as the integrations with all relevant parties (e.g. right now credit checks are instant because of the systems that have been setup to connect the different parties). Because of the uphill infrastructural challenges, I can see why the credit agencies are slow to move on new technologies like machine learning.

  3. Very interesting! I understand that public perception of Equifax is extremely negative. But I wonder what the lenders who actually use these scores to make lending decisions think of it. Is there truly a real incentive for them to look for a different solution?

  4. Really interesting article. It seems to me though that competitors will struggle to get their hands on a comparable data-set. I appreciate that this advantage is probably not unassailable forever but it’s probably worth noting that they have a huge lead on potential competition.

  5. Great piece of analysis. I can see a real threat to lenders bypassing the traditional credit scoring system altogether, as FICO scores become increasingly divergent from their own models and seen as outdated and imperfect. LendingClub has seen this pattern in its own loan approvals as it has refined its model based on its increasingly large cumulative loan book. However, the FICO score still provides a “language” that the industry understands, and a standard that is measurable across all consumers. It would be challenging for a start-up to replace the industry standard in the short term, even if they do have a superior algorithm.

  6. I would be interested in knowing how ML and past data trends especially around the biases of how the credit system is stacked up today will translate into this new credit system. There may be an ethical dilemma such as the one we saw with Amazon and its hiring AI which was systematically rejecting female candidates.

  7. Super interesting topic. In answer to your second open question, I believe Equifax’s greatest advantage is the depth of its existing data on consumers. My opinion is that Equifax would be best-served in growing its business by partnering with the new machine-learning driven credit assessment companies in the world. Transunion and Experian, for example, have started to form partnerships with East Africa based alternative credit data start-ups to supplement their existing data sources and scorecards. These companies jealously guard their performance data (i.e., the incremental efficacy of using more quantities and types of data), but the rapid adoption by banks indicates a positive outlook.

Leave a comment