First time right applications for mortgage lenders – Expectations vs Reality

In May 2020, the Bank of England warned that the UK economy is facing the sharpest downturn since 1706. The mortgage sector has been one of the worst-affected sectors by the COVID-19 crisis. Pre-COVID, the UK mortgage origination is viewed as complex and lengthy. There are many “pain points” in the process, taking 18-40 days (HOA, 2020) and costing the industry £5K-£7K per loan.

Ninety percent of UK mortgage originations involving manual processing of (on average) 100 different types of paper documentation. Also consider that for the 100 banks in the UK, each has 4 bank statements, resulting in 400 different bank statement variants. Most lenders manually read paper-versions, or scans, and validate the data before underwriter assessments can begin.

What’s the problem?

Socially distanced and work-from-home origination has further exacerbated the problem. An increase in erroneous filings leads to higher costs in terms of money, time, and reputation. With the COVID-19 crisis, physical documentation validations have reduced, which has increased re-submissions (sometimes four times before they are ready before underwriting).

The underwriting assessment involves a largely manual process as unstructured data residing in documents is difficult to extract. Mortgage credit, as a result, is denied for those borrowers most in need. During the COVID-19 crisis, 50 percent of applications have been denied (Butterfield, 2020).

First-time home buyers, vulnerable and underprivileged areas and SME businesses have been greatly affected. UK Finance states that the two million repayment holidays have increased the cost of operations and lenders have withdrawn almost two-thirds of their products representing over 75 percent LTV (Williams, June 2020).

The UK uses, on average, 100 types of documentation in mortgage applications. 90% of lenders manually read paper-versions or scans and validate the data before underwriter assessments can begin. As a result, in specialist lenders, applications are submitted four times on average before they get it right. Socially distanced and work-from-home origination has further exacerbated the problem.

AI-enabled automated data extraction and validation of mortgage documents will address this problem and accelerate an urgent sustainable economic recovery from Covid-19.

A ground-breaking innovation with one shot learning and first-time right applications for mortgages

The innovation will combine computer vision, machine learning and NLP technology. The innovation will work in two steps:

1. Extract data with 95%+ accuracy from mortgage documents and minimal training datasets enabling one-shot learning leveraging mortgage versions of techniques such as BERT and GPTs which have yielded great results in AI for natural languages, combined with SOTA computer vision models.

2. Validate the extracted data in real-time against document and cross-document conditions to enable first-time-right applications at scale. Cross-document level validations are complex requiring significant natural language processing and understanding techniques e.g., extracting data to identify one-time credits in a bank statement and co-relate it to a free form letter supporting the credit provided.

To perform this in real-time and at scale, cloud production architecture and software engineering leveraging massively parallel processing will be employed.

The innovation will drastically improve on the current state-of-the-art by achieving 95%+ accuracy of extracting and validating data from mortgage documents with one-shot learning in real-time and at scale for first time-right applications. One short learning (OSL) algorithms and models leveraging the newly curated Mortgage BERT framework will be applied.

Massive Benefits

Digilytics RevEL will help deliver mortgage loans that are more affordable and delivered quicker. In a post- COVID economy this will help the UK mortgage industry reduce the cost of operations and time taken to disburse funds to borrowers requiring urgent access to credit.

The innovation will create social benefits by allowing human expertise to focus on the complex underwriting, by automating the simple validations with data extracted accurately from documents.

Furthermore, it will make the loan origination process paperless thereby accelerating a sustainable solution for an urgent recovery from the Covid-19 crisis.

Conclusion

Going paperless will lead to environmental benefits. With a simpler and streamlined process, employees can be redeployed to where more human interaction is necessary, providing social benefits, greater customer satisfaction and lower reputational risk as well as favourable views during regulators' on-site examinations. Reduction in transaction costs, should lead to more affordable credit availability and financial stability.