Original article can be found here (source): Artificial Intelligence on Medium
How SMB Lending Banks can Transform Credit Underwriting using AI and ML
Author: Vijaya Kumar
Small and Medium sized Business (SMB) lending banks, credit unions and fintech/online lenders in the US are going though challenging times. Uncertain economic and regulatory conditions and changing needs of SMBs, are some of the broad challenges they face. Add to it the pressure of risk management that arises from diverse SMB customers and a sheer number of loan applications.
A lower interest rate is no longer a differentiating factor for banks, but a seamless customer experience is. SMBs now need intuitive, ‘agile’ banking to secure funds to grow their business — they need
- Faster processing of application
- Fair lending terms and
- Transparency during the underwriting process.
Accordingly, many SMB lending banks have automated and fine-tuned their credit underwriting processes.
But, are they able to accurately model credit risk for all types of SMBs?
Many of them still lack effective systems for data, analytics and the right platform to accurately assess small business risk. This points to a large market that is left unserved. According to Biz2Credit, the approval percentage for small business loan applications at big banks was 28.2% in December 2019 — a record high. And the approval rate at small banks, which often are SBA-approved lenders, also climbed one-tenth of a percent to 50.6% in December. This still means nearly 50–70% of all SMB loan applications that banks receive are being rejected! Fundamental flaws in approaching credit risk assessment can ultimately threaten their bottom-lines and competitiveness.
How can SMB lending banks provide a better customer experience while managing risk effectively?
SMB lending banks in the US now need to improve their lending process across the credit life cycle. They should reduce risks by integrating non-traditional data sources on borrowers to arrive at credit scores that are more holistic and better reflect creditworthiness. This is mainly because small businesses do not fit into traditional credit scoring categories. Usual criteria like frequency and amount of borrowing, credit scores, repayment history and so on are not enough to assess them.
Interestingly, in late 2019, five US financial agencies issued a joint statement stating that they “recognize alternative data’s potential to expand access to credit and produce benefits for consumers.”
Banks stand to gain significantly by using artificial intelligence (AI) and machine learning (ML) powered underwriting solutions to assess the risk in loan applications and deliver a smooth customer experience.
Let’s deep-dive into the three areas that need to be addressed to transform credit underwriting using AI and ML.
Most SMB lending banks have limited integration with data from credit bureaus and alternative data providers. In many cases, data points are entered manually compounding the challenge. In addition, data gets captured into multiple systems which remain unintegrated. This makes getting a holistic portfolio of SMBs very difficult.
To improve the quality of underwriting, data from alternate sources must be integrated to create a more inclusive snapshot of an SMB. Only then can banks optimize their lending terms and avoid risk. A 2019 CPFB study revealed that using alternative data and ML approved 27% more applications than a traditional lending model and yielded 16% lower average Annual Percentage Rates (APRs).
Let’s take a look at the different types of data that should be integrated:
- Financial data: Ratios, Tax returns, Cash flow statements
- Macro-economic data: Sector risk (failure rate), Macro-economic indicators
- Data from credit bureaus: D&B, SBFE, Owner’s credit score
- Size and age: Total assets, Age bands (early life, mature)
- Alternative data: LexisNexis, Thompson Reuters, Social Media reviews
- Compliance: Court actions, Secretary of State (SOS) filings
Gathering and integrating this data is a significant exercise in itself. This effort will only be useful if you glean powerful insights to make drastically better lending decisions.
This brings us to the next important part of the solution:
It will not help if there is no systematic method for feature definition, extraction and calculation. This will result in sub-standard decisioning.
ML and AI scoring algorithms can transform SMBs’ data into meaningful insights. The ideal solution should be able to recognize patterns in financial and non-financial data, parse ambiguous data points or those whose context is unclear, and connect millions of data points to make an accurate risk assessment. This can potentially expand the lender’s customer base even to the “credit invisible” or “thin-file” category (those who have no credit history with Equifax, TransUnion, or Experian).
As time passes, the self-improving algorithms should be able to develop patterns from the added data on each loan and its repayment, and be able to make sharper predictions.
The ideal analytics solution needs to use effective risk segmentation schema based on:
- Industry segments
- Obligator types
- Collateral type and value
- Guarantees, and
- Lien position/sensitivity of claim.
It should utilize an ensemble of statistical techniques capable of
- Random Forest / Logit / GBM / RNN, LSTM
- Deriving hidden correlations and non-linear patters
- Isolating leading indicators of SMB behaviour, and
- Champion/Challenger models or strategy design.
The tool must help banks in optimal credit decisioning and effective monitoring of their:
- Decision to approve and amount of new exposure
- Extension of limits to existing relationships
- Optimal sectoral concentration, and
- Control limits / sensitivity analysis.
Hands down, ML-based models are better at assessing risks and pricing than human underwriters or merely digitized underwriting processes.
The credit underwriting platform must be capable of effective strategy management — the ability to run and manage multiple concurrent strategies, and multiple handoffs. It should also be able to integrate decisioning systems into LMS and LOS workflows. In addition, the system should have batch and real time decisioning capabilities that integrate seamlessly into the bank’s existing lending workflows.
This workflow diagram sums up the ideal credit underwriting platform:
This will result in an improved lending process throughout the credit life cycle from data ingestion to monitoring. The picture below explains this further:
To sum up, an AI and ML-powered small business credit underwriting platform will be the biggest opportunity and competitive advantage for SMB bank lenders.
It will augment the entire credit life cycle — finding prospects, measure creditworthiness and manage the loan portfolio. It will also reduce the cost and time involved in loan applications, and help banks offer more loans confidently. By offering increased convenience and faster loan decisions, lenders will be able to reach markets they have never serviced before.
Scienaptic’s cutting-edge platform Ether does just that. It is an advanced AI powered underwriting platform with prebuilt features which can be deployed within 4–6 weeks and deliver drastically better decisions. Talk to us to know more.