Unlocking Greater Efficiencies: Evolving Tech Solutions for Structured Underwriting in a Hybrid Working World

by Andrew Carman Sept/Oct 2023
The shift to remote and hybrid work has rapidly transformed the way we work, including our approach to underwriting both small and large-ticket transactions. Andrew Carman explores how newer technologies can help remote working risk teams underwrite more effectively.

Andrew Carman,
Lonetown Capital

Recently, a hybrid-working leasing team had limited time to prepare an important credit memorandum. Financials had already been spread and customer interviews completed. But the sales team still had to follow up with the lessee regarding a critical customer contract, while risk needed to prepare commentary on forecasts and stresscase scenarios where debt service coverage was tight.

Despite the timing, the team felt reasonably in control, largely because they were using a new technology called PerCina Report. Designed for ease of use, it enabled the users to generate credit reports and equipment valuation curves effortlessly with just one click. The technology also facilitated real-time team collaboration while prioritizing the security of data handling.

The five team members seamlessly streamlined their report writing and analysis: risk analysts in both Mumbai and the U.S. added vital financial commentary, sales engaged with the customer directly in the tool to add their commentary and the underwriting team manager monitored real-time updates in the final report — all virtually and simultaneously. Within a few hours, a secure and comprehensive credit memorandum was created and sent digitally to their investment committee for a $9 million transaction.

An Evolving Underwriting Landscape

Tools that streamline underwriting processes are not just a matter of convenience, they are becoming increasingly indispensable. The equipment finance industry has had a rich history of embracing technology to drive efficiency, from leveraging non-industry- specific software for back-end accounting and front-end CRM systems, to developing more industry-specific solutions for pricing and credit scoring. Despite these advances, structured credit and large ticket underwriting is still regarded as the “last bastion” of manually driven processes. Even companies that have gone all-in on technology continue to process their larger credit analysis and decisioning manually. While there are excellent reasons for maintaining these approaches, a new wave of technologies focused on more advanced credit analysis and adjudication, including for larger deals, is underway.

With artificial intelligence and machine learning making leaps into the mainstream, and risk teams increasingly working remotely, the adoption of purpose-built technologies is becoming more pressing. The emergence of more accessible tools like ChatGPT, Bard AI, LaMDA and Bing AI, is also accelerating discussions about where and how to use such technologies. AI advances in credit technology are beginning to challenge the effectiveness of traditional underwriting methods and therefore their potential applications warrant better understanding. Staying informed for now is likely a good short-term strategy, but to be better positioned to leverage these exciting solutions, companies must dive deeper into what these applications can do today and what they will soon do.

New Technology for Remote Risk Teams

Hybrid working, in its various forms, seems here to stay. I surveyed several ELFA member companies, which included a variety of large and new independents, bank-owned lessors and member advisory firms that specialize in structured underwriting, to ask about remote working policies and, more specifically, if their risk teams were continuing to work remotely. While some bank-owned lessors remain relegated to following in-person working protocols, member companies are generally adopting more hybrid models with success, and their risk teams’ experience was also positive. While there is still a strong desire to have teams together whenever possible to attract and retain key talent, particularly in credit, remote working flexibility is becoming a must.

My further research also shed light on the technology impact on remote underwriting. While communication tools have improved company systems and data access for an increasing number of employees, working remotely in a variety of office settings has continued to pose challenges. These process gaps are compelling companies to rethink their technology platform strategies.

Large Ticket Underwriting on the Fast Track

Smaller ticket micro lenders have long been the vanguard of technological adoption and today many are already incorporating or considering adopting AI technologies in the solutions they use. Larger ticket underwriting, on the other hand, remains largely reliant on legacy methods. Discussions with a variety of ELFA member lessors confirmed that when it comes to large deals, they still rely on traditional methods because they work, despite being inefficient. It’s important that an analyst be able to get into the “details” of the financials to understand a deal properly. For example, some credit shops have experimented with outsourcing financial spreading to a third-party provider with mixed success. Others have adopted technologies that spread financials automatically, only to have an analyst re-spread the numbers themselves at a later point in the process. Some spreading tool technologies and larger deal scoring tools have taken hold with some companies, but for most deals more than $1 million, manual underwriting still rules the day.

This may be changing faster than we think. Promising new technology companies and solutions are beginning to focus on more of the manual underwriting processes. Established fintech lenders like OnDeck, BlueVine and Pipe, that continue to take their solutions further into mid-sized deals, are starting to share the stage with new B2B SaaS lending platforms like Liquidity Capital, Infermal.com, Hum and OakNorth. Additionally, alongside these companies exist several growing AI-credit focused solution providers like Zest AI, Ocrolus and Underwrite.ai, among others.

This newer vanguard of AI credit scoring players focuses on a range of solutions aimed at improving lending and credit decisioning processes. They leverage more powerful AI capabilities to create forecast models that demonstrate greater accuracy in predicting obligors’ risk of delinquency and likelihood of default, enabling lenders to turn complex data and insights into actions without writing a single line of code. What’s different about these AI-enabled companies is the range of the data they access and how they access it, which enables them to be more ticket-size agnostic, thereby paving the way to larger transactions.

Fast-growing, modern companies need lenders that understand new business models and can operate from the same tech level as they do. Existing smaller ticket finance companies as well as end-to-end solution providers increasingly operate this way — combining proprietary system configurations, customer systems and third-party API integration — and have created powerful solutions for the equipment finance industry.

For larger transactions, however, real-time company-specific data is required and, historically, has only been achievable with human interaction. This is changing. These new AI solution providers are making it easier for lenders to “plug in” to a wider array of customer data sources, both internal and external. Beyond credit bureaus, payment systems and bank accounts, more data is being incorporated into the analysis — ERP systems, customer accounts, AP and AR systems, inventory data and CRMs — all in real-time. Additionally, more companies are now open to providing their data to their lenders using these tools if they get what they need — that is, a faster loan decision with less work. It’s important to note that the number of AI-enabled solution models focused on risk are growing and were built this way originally. This will likely enable them to develop solutions even faster.

Newer technologies are also being designed to ingest company data from dozens of backend sources seamlessly; first, transforming historical company data to create a baseline cash flow forecast, then applying proprietary, deep learning engines to run literally billions of scenarios on all of the data factors to refine the likelihood of default based on future cashflows — at a very high probability rate. Furthermore, these systems can be embedded with customer systems to run continuously, so real-time feedback on how a company will perform is continuously generated. These models can also provide dynamic portfolio management services ranging from covenant testing to financial reporting, nonstop. To date, primarily newer companies needing capital more urgently allow this level of access to their proprietary systems. It will take a little longer for more established companies to jump on board. But when “data can substitute as collateral” no longer applies to just small ticket credit scoring models, adoption of these new technologies by established companies, and thus more tech-decisioned larger deals, will occur.

How to Think About AI & Newer Technologies

AI has ushered in a new era of automation and analytics, touching many aspects of equipment finance, including faster and more accurate credit decisions, improved customer service, predictive maintenance on equipment, fraud detection, KYC and even personalized marketing.

But how does AI work? AI, machine learning and deep learning are related, but not the same thing. Teaching a computer to play checkers was an early example, but a narrow version of AI. ML, a subfield of AI, uses algorithms to parse data, learn from it, and then make a determination or prediction about something. Rather than relying on code with instructions, the machine is “trained” using large amounts of data and examples that give it the ability to learn how to perform a task. Keeping spam out of our inboxes is an early example of this. Today, more advanced AI and DL go further and use artificial neural networks containing layers of artificial neurons. DL is an advanced form of AI utilizing more layers. Inspired by the biology of our brains, these networks possess discrete layers of neurons that are used for higher cognitive understanding — the first layer detects shapes, second layer eyes, and so on until it recognizes faces, for example. Think of a self-driving car reading a stop sign. Early versions were prone to errors because the car was not trained adequately enough. Advances in computational power and data storage now make it possible to create enormous networks and layers of neurons. It’s now possible to run massive amounts of data through a system and effectively train a machine on millions of scenarios, enabling it to learn more effectively. Now think of the car seeing the stop sign, but even in fog. This is DL.

How are companies getting started? Many financial service companies are moving forward and investing in AI, including adoption at a platform level or more simply accessing an AI-enabled tool. There are several ways to begin experiencing these AI tools:

  • Signing-Up for Existing Select AI Services is a fast way to get started. Simply sign up with an existing generative AI solution (ChatGPT, Bard AI, etc.) if general research, chat or basic analysis is something you are seeking. These are low cost, but output controls and results can be considerably limited. They rely on general internet-based information, which renders these solutions prone to reduced accuracy and inconsistency.
  • Prompt engineering is another level of adoption that uses general internet data combined with training on business-specific information. This requires hiring expertise to create and edit prompts and requires integration with your business systems to introduce this data. Output accuracy is much better, and this is still relatively affordable.
  • Building a custom model yields much stronger output accuracy. This can be expensive, requires more specialized skills and takes more time.
  • Teaming up with existing solution providers is another path that may be more advantageous in the short run. These providers have already made significant investments in their platforms and are looking for customers to integrate with. The trade-off here might be the degree to which their solutions match your businesses’ and yield the results you desire.
  • Caution when using. Despite the exciting possibilities associated with this technology, there are challenges to address. Data quality, integration complexities and ethical concerns related to bias in decision-making and perhaps who we lend to come to mind. Off-the-shelf generative AI tools learn from existing data sources. While they may be strong search engines, they also are prone to pulling incorrect data depending on the data or knowledge source they are pointing towards. Given this, the practicality of regular business use for these tools must be thoroughly vetted and understood.

Secondly, companies making serious investments in AI at a platform level that want to produce better data outcomes have elected to build their own solutions and hire skilled teams. In order to better control the AI outputs, these companies are making significant investments to create comprehensive knowledge libraries in which AI applications can be trained on data more specific to their industry. This approach requires more time and money, but output accuracy improves greatly.

Re-Thinking Structured Underwriting and Moving Forward

As exciting as these newer technologies are, we must remain mindful that larger deals, by nature, are complex and possess numerous elements that must align before a lender says “yes.” Furthermore, these deals possess numerous variables, which pose challenges when trying to capture information consistently. To close this gap and potentially accelerate tech adoption, we really should be asking ourselves, “Are we framing the technology adoption question for larger deals correctly?” Maybe the tech is not being applied to the right part of our underwriting process? More importantly, are we delineating properly between where technology can perform a process more accurately, faster and cheaper than we can, and therefore should be deployed to do more of that specific task? From there, we can begin refining human oversight in a process so human interaction can continue to be justified and increase its own efficiency. This would enable an analyst to become hyper-focused on higher value added tasks that still require expertise and where they can continually excel.

The continuous developments of AI and related technologies are both exciting and daunting. What’s interesting about AI is that it is not about replacing the human touch but enhancing the role we will play with these technologies. When balanced correctly, we can achieve even greater efficiencies.

ABOUT THE AUTHOR: Andrew Carman is the CEO of Lonetown Capital and creator of PerCina Report, a credit underwriting technology. Carman has worked in global finance for more than 25 years with companies like GE Capital, Siemens Financial Services and SQN Capital.

Leave a comment

No categories available

No tags available