As cities, states, and the nation reopen, small business’ access to capital will be a critical ingredient in recovery. The colossal pandemic of COVID-19 has had a wide-ranging impact on public health, civil society, and the economy. In the world of financial services and financial technology, companies lending to small businesses were among the hardest-hit. Most lenders severely curtailed their operations, with many pausing non-government-backed lending completely. This historic pandemic has revealed countless truths, among them the insufficiency of credit data and the need for small business lenders to adopt a deeper, more data-driven, and more personalized approach to risk management. In this post, we outline several analytical principles for recovery, with data sources and metrics that are likely to be useful for lenders seeking to restart with confidence.
Dramatic Impact, Uneven Recovery
It’s impossible to overstate the depth or breadth of the pandemic’s impact on our world. To date, in the United States, 24MM jobs have been lost, and over one-hundred-sixteen thousand people have died, more than the number of Americans who perished in World War I. It is still too early to tell the longer-term impact on society, government, and the nature of work.
The economic impact of COVID-19 has been sharper and more severe than that of a typical recession. To avoid the devastating health effects of a global pandemic, people are staying home, events are cancelled, and businesses are shuttered. Many small businesses were legally not allowed to operate, and those that had the capital to survive are only now beginning to reopen.
Metrics For The Recovery
This unique disruption has caused many small business lenders to reexamine the way they operate. As lenders ramp up again, automation, efficiency, and trusted data have taken on even higher importance.
It is critical for lenders to develop a nuanced understanding of customers’ business dynamics and future prospects. At Ocrolus, we offer an extensive set of analytical metrics for use in small business underwriting and are now augmenting and enhancing them with insights specifically geared towards recovery. In particular, we can summarize the themes as:
- Data: Increase the breadth of data sources used to evaluate a small business;
- Duration: Widen the timeframe used for evaluation; and
- Detail: Incorporate greater detail about each prospect and use automation to scale a personalized approach.
Data- Increase Breadth of Sources
Small businesses are inherently complex, and each business is different. It is not sufficient to rely on traditional credit data. In a volatile world, the value of past experience as a predictor of future performance is diminished.
Bank data is particularly important, as lenders will want to understand cash balances, revenue, and expenses. Several important questions can be answered with this information. How have cash balances trended over time, and has a business’ cash cushion become more robust or depleted? What is monthly and annual revenue, and is there seasonality to these figures? What are the main sources of revenue, and what fraction of them are recurring vs. episodic? Are revenue sources overly concentrated or diverse? What are monthly expenses and net cash flow; are these figures trending up or down; what costs are fixed vs. variable; what are the largest expense categories?
In addition to bank data, lenders may also seek to examine payroll information, tax documents, financial statements, and personal/corporate identity verification documents. For certain business types, specific information on rent and inventory may be relevant. While there are many potential sources of valuable information, this data is often complex, disparate, and fragmented. It is difficult to access, resource-intensive to manage, and full of noise. Each lender needs to focus on the datasets that contain pertinent information for their particular customer segment and determine how to extract valuable signals in an automated and reliable way.
History- Extend the Timeframe
Ask any analyst or risk manager how much data they want, and you will receive a predictable answer: “as much as possible!” Of course, there has always been a balance between completeness of customer information and the technical friction required to obtain it. While many business lenders have historically relied on three months of financial data, they are wise to now consider six months or more as the minimum, particularly as automation reduces the marginal friction of additional history.
Observing performance over a longer time period helps reveal seasonal patterns in business cash flow. It can also make it easier to detect anomalous events or longer-term trends in a business’ upward or downward trajectory. For example, it would be relevant to compare revenue and expenses pre-COVID and post-COVID and understand the extent to which a business was affected, what – if any – stimulus relief it received, and how quickly it has been able to recover momentum. 3 months simply won’t provide enough information.
In addition to obtaining data at the point of underwriting, forward-looking lenders will look to maintain insight across the customer lifecycle. This is particularly important for more persistently-accessible financial products, such as a line of credit that allows a customer to draw and repay multiple times, a credit/charge card, or various insurance products. For longer-term customers, it is useful to understand revenue growth, profitability, and cash balances. In addition, lenders will want to know how quickly and effectively their capital has been utilized, as well as if the customer has obtained other sources of credit. Failing to monitor a customer’s cash flow and use of credit post-approval may lead to a miscalculation of debt capacity and lack of ability to detect emerging risks in a portfolio.
Gathering more data over a longer period of time requires an investment on the part of lender and borrower alike. Lenders and data service providers will need to work together to offer low-friction solutions to provide this data in a cost-effective way. At Ocrolus, we accomplish this using machine-learning-powered automation, with human-in-the loop intervention where necessary. Our algorithms learn over time, reducing turnaround time and friction as they gain more experience. Through Ocrolus+, we are also able to combine data collected from documents with persistent financial account connections in a single API.
Detail- Deeply Understand Business Dynamics
Each small business is unique, and lending to this segment of the market requires a nuanced appreciation of the financial and operational dynamics of particular businesses. It’s not enough to concentrate on a limited set of data points, and it doesn’t make sense to evaluate all applicants in the same manner.
Consider, for instance, a retail store and a local law firm, each with $1MM in annual revenue. Would you evaluate them for credit in the same way? Of course not! The retail business likely has a high cost of goods sold, as well as major fixed expenses such as rent and payroll. The law firm likely has a much higher gross margin, as its major cost is the time of the partners and other staff. By examining both businesses at a transaction level, it is possible to classify expenses as recurring vs. episodic, understand the seasonality of revenues, and determine the categories driving a company’s spending. Upon examining this type of data a lender would likely make different decisions on loan/line size, pricing, and eligibility for businesses with different operating models.
A bank transaction at first glance contains a limited amount of information: a date, an amount, and a description that could scarcely be considered human-readable. However, models trained on billions of transactions allow us to discern a variety of useful information from each record:
- Who is the merchant/payee?
- Is this transaction recurring?
- Should this transaction count as revenue?
- What category tag(s) can we assign?
- Is this a non-sufficient-funds or overdraft transaction? Is it a fee transaction pertaining to one?
- Does this transaction indicate a relationship with an alternative lender? How can we use this to understand a business’ use of capital and its capacity to service debt?
- Is this transaction a transfer from one of the business owner’s other accounts, or is it actual revenue from a customer?
- Is this revenue from a merchant processing transaction?
- Do the numbers from a set of transactions tie out perfectly, or is there a suspicious inconsistency?
As one might expect, detailed, transaction-level data as described above can also be stitched together to understand patterns across transactions, bank accounts, time periods, and entities. This information can be used to understand the intricate financial dynamics of a business. It can also form a profile that can be used to measure the likelihood of successful payback, predict delinquency, or identify fraudulent behavior. Lending is a business where money is made or lost at the margin, and in times of volatility, it is essential to make risk decisions with full visibility into the nuance of customer dynamics.
Small businesses account for over 60% of net new jobs in the United States. The economy depends on the health and vitality of business owners and their companies. As the crisis caused by COVID-19 hopefully abates, small businesses will require capital to form, to operate, and to grow. As lenders begin to scale up their activity, forward-thinking leaders will be wise to consider how they can evaluate borrowers in a more complete and nuanced way. This is essential for making capital available for business while also managing risk in an ever more volatile world.
Through analyzing an expanded set of data, over a longer period of time, in greater detail, lenders can make high-quality decisions. If they can do this in an efficient and automated way while maintaining a pleasant customer experience, significant value will be created for lending companies, small businesses, and the overall economy.
Ocrolus helps financial services companies make high-quality decisions in an automated and efficient way with trusted data.