The tension between digital innovation and manual credit underwriting
PropTech innovators are changing the way property is bought, sold and financed, bringing rapid change to an industry not previously known for its agility. Over the past decade, the sector has grown from a mere handful of notable companies like Zoopla or Rightmove, to a red-hot source of innovation and disruption attracting large scale investment.
The fintech revolution within property lending means that today real estate backed loans, including mortgages, are no longer the prerogative of banks. The burgeoning alternative finance sector comprises a wide array of alternative debt providers. Many of them are well-performing principal lenders neither organised as challenger banks nor as P2P platforms or marketplaces. And today, all property lenders – young and old – are asking themselves what role technology can and will play within their businesses.
The dilemma: It is generally accepted that a lender that doesn’t embrace technology will not be able to scale its business. This is true across all lending verticals – SME loans, consumer loans, real estate loans. At the same time, lenders also fear that where credit decisions are taken by machines, large credit losses will follow. Whilst they recognise the need to scale, they are also mindful that ill-conceived technology without sufficient manual oversight will scale the mistakes, too.
Interestingly, it seems property lenders are disproportionately wary of the pitfalls in automating credit underwriting – more so than their colleagues in SME and consumer lending verticals, it seems. And rightly so.
With tech lending platforms like Zopa, the world’s first and oldest loan marketplace, now in its 12th year (and notably in the process of becoming a bank), it makes sense to analyse what impact the advent of high tech has had within consumer and SME lending. Since fintech lending against property is a younger asset class than consumer lending or SME lending, we can then ask ourselves whether property lending will follow a similar path as it matures.
Within SME and consumer loans, many sophisticated tech lenders principally rely on automated systems to perform their credit assessments. Many use a statistical approach; intelligent systems first gather borrower data, then smart algorithms interpret those to assess credit integrity and price the loans. Where the underlying loans are large in number and small in size, there is safety in numbers: the statistical approach is supported by the granularity and homogeneity of the underlying asset class. More often than not, the alternative approach – to manually underwrite each loan – is somewhere between too costly and impossible.
At the most sophisticated end of the consumer / SME lender spectrum, credit assessment systems operate like a traffic light. Many have developed a ‘red light’ and an ‘amber light’: their system automatically rejects loan applications that fail to satisfy pre-determined eligibility criteria, while automatically forwarding all other applicants to manual underwriting for human review. The most developed credit assessment technologies are already capable of assigning ‘green lights’ as well: they automatically accept a loan application – and then price and pay out a loan – without any human involvement at all.
Of course, the willingness to let a machine decide could be the result of a platform’s tolerance of risk, rather than its technological sophistication. And maybe that is because there is safety in numbers. However: it is nonetheless remarkable that today – before the real impact of the artificial intelligence revolution is felt – there are already several lending platforms in the US and Europe with sufficient confidence in their fully automated credit systems to allow them to price and pay out thousands of loans, reviewed and assessed by software only.
As time goes by, platform investors across all lending verticals increasingly rely on credit assessment technologies, and less so on manual underwriters. Mortgage banks and other traditional property lenders are also acutely aware of this phenomenon; they have no choice but to ask themselves to what degree credit underwriting can be automated. Having realised that ‘Silicon Valley is coming’, as JPMorgan CEO Jamie Dimon once put it, banks are increasingly looking towards fintech innovators with amazement and respect; and with an open chequebook. More specifically, NatWest recently confirmed it is hoping to introduce an automated lending process aimed at allowing NatWest to approve certain commercial real estate loans in less than an hour.
Today, even the most advanced credit underwriting processes in property are more manual than their consumer and SME lending counterparts. The key question is whether, as time goes by, property lending will remain more manual or whether this vertical is just a few years behind the curve. I have discussed this question with senior property lending professionals at traditional banks and fintechs; their responses were remarkable. All bankers were at pains to highlight the importance of learning from fintechs, to accelerate their credit decisions with automation. Conversely, fintech property lenders are emphasising the strength of their manual underwriting teams – many of them ex-bankers – stressing the “old school” nature of their approval processes. This dichotomy highlights the tension between high tech and old school underwriting, specifically within real estate lending.
To understand this phenomenon more fully, it helps to look more closely at key differences between real estate lending on the one hand, and SME and consumer lending on the other. In our experience, there are three critical differences, which fundamentally impact the role tech can play within property lending.
i.Digital availability of data
The most advanced consumer and SME lending platforms have managed to automate credit and underwriting processes where borrower data, such as bank account balances and transaction history, became digitally available. Smart lending systems are only able to produce the deep insights they require because a plethora of sources like borrower banking history, credit ratings, and (within SME) information on a company’s directors can be pulled together by a machine, not a human.
These lenders utilise proprietary technology to gather and cross-reference digitally available datasets. Digital availability is a prerequisite to assessing the borrower’s credit integrity. If implemented correctly, the speed, accuracy and completeness of these data gathering processes is remarkable. Machines will outpace and outperform the capabilities of any human-led process by a most significant margin.
ii.Loan book granularity
Consumer lending platforms such as Zopa or LendingClub produce highly granular loan books for their investors, made up of thousands of individual loans. Investors expect defaults to occur – they just want the rate at which they occur to remain constant. By comparison, property lending revolves around writing significantly larger loans, and far fewer of them.
To put things into proportion: an average size consumer loan on Zopa is c. £10,000; the average LendInvest loan size exceeds £1,000,000. The concentrated nature of real estate loan books means a single loan “going bad” will have a large impact on the loan book performance. Good real estate loans are backed by hard assets that don’t easily drop in value. Where a property loan is secured on sustainable real estate with prudent loan-to-value ratios, credit losses are rare to come by. Where granular consumer portfolios, for example, offer safety in numbers, real estate loans offer safety through asset backing.
The prospect of a single property loan failing is problematic for real estate lenders. Scaling mistakes through ill-conceived technology, even temporarily – for example in a test phase – is deemed unaffordable. Where SME and consumer lenders can tolerate a (stable) proportion of their loans to default, real estate lenders are not afforded the same privilege.
iii.Property loans are less homogenous
Compared to property loans, consumer and SME loans are relatively uniform. Each property is also unique. Even adjacent and identical houses in the same area can be very different in value, for example where one house overlooks the river and the other the local school or a housing estate. An important degree of protection stems from the fact that property loans benefit from independent asset valuations. However, since valuations tend to be an art and not a science, the question will always be whether a property’s features have been correctly adjusted for by the valuer. Allowing a machine to make this assessment is not straight forward, and many property lenders argue it never will be.
With the above in mind, real estate lenders are right to remain careful before handing over significant credit controls to machines, which are unable to account for gut feel factors and other small-but-meaningful differences. For the foreseeable future, it appears property loan underwriting will continue to require significant human involvement, more so than consumer or SME lending. Does this mean there will be less tech in property lending? Not necessarily: maybe it means other areas of property lending, and not credit underwriting, will see the most relevant tech innovation.
Whilst the above analysis sets out the key challenges in automating credit underwriting within property lending, we have yet to identify those areas in which technology can (and is being) utilised by lenders to innovate. This analysis – Part II of this series – will have to wait until next time.