Managing risk is core to equipment finance operations and capital. Risk — the potential of unwanted outcomes — emerges in all shapes and sizes and the industry has instinctively looked to data to help manage risk. One of the best examples of this predilection is how underwriters consistently seek additional data and analysis for large-ticket deals compared to app-only and auto approvals for high cadence, small-ticket flows. Underwriters and operational leaders analyze data, looking to identify circumstances and behaviors that precipitate unwanted outcomes.

Data is the language of risk.

AI is built on data, and AI models predict behaviors, so it is not surprising that finance leaders today are looking to AI as next-gen data use in risk management strategies. To understand how AI can help, consider the three ways lenders fundamentally manage risk:

• Avoid it: “We only do the good deals” or “We only do ‘A’ credits”
• Transfer it: “We can share this deal with the friendly lender down the street,” or “We need to reduce our exposure here and sell these deals to someone not as exposed.”
• Engage and manage: “We will work with riskier credits, but we add structure and/or collateral to reduce it,” and “If our exposure gets too high, we will find ways to balance or hedge to get back within guidelines.”

Many readers are probably most comfortable with, and practice ‘Avoid’ and ‘Transfer’. Business operators are usually taught risk management in the context of project management: managing risk by identifying and avoiding it. Advanced courses then add “Have a Plan B” for mitigation because not all risks are known at initiation or easy to identify, so “it will happen.” But the adventurous, the entrepreneurial and highly motivated, look for risk to engage because they seek the higher rewards associated with that risk. The good news is that the availability of data serving these strategies continues to grow and now AI is augmenting that data to provide improved execution of all three strategies.

Avoid it

Avoidance strategy involves eliminating business risk by avoiding the activities or situations that carry it. UPS implemented a case study example of a risk avoidance strategy when they changed their truck routing algorithms from shortest distance to no-left-turns in an attempt to reduce accident liability and costs. No-left-turns avoided crossing traffic and avoided collisions but also had the advantage that the routes saved time and money by removing left-turn waiting and the corresponding fuel. A crucial aspect of this story is that data is the key to identifying both the risk and the associated activities. Accidents are detrimental to a delivery business and are readily apparent in the data. However, the crux of the no-left-turn policy was identifying the primary activity associated with the undesired outcome in both operational and incident data. UPS gathers tremendous amounts of data from both the package delivery and driving operations, so they had the data needed to identify the accidents and the associated activity to avoid.

In equipment finance, the outcome to avoid is default. The associated activity is delinquency — the failure to make timely payments. Companies typically try to avoid default by ranking borrowers’ ability and desire to pay using a variety of scoring parameters based on historical payment data, including credit scores like PayNet and FICO. But this data and the associated scorecards are backward-looking, and while they accurately reflect past behavior, everyone knows the disclaimer “past performance does not guarantee future results.”

Payment and its converse, delinquency, are behaviors and as such present a distribution of results across both the population and for any given member of a cohort. AI prediction machines are built for this specific problem — predicting the probabilities of a range of behaviors. Data identifies the payment-behavior outcomes as well as the circumstantial parameters enabling AI models to provide the probability of each of those behaviors. As a result, lenders can avoid such circumstances or borrowers who may not be able to pay. If the AI predicts the probability of the risk, delinquency, is non-zero, “avoiders” can avoid the interaction.

Transfer it

One might infer that transference is a reactive risk management strategy because the transfer or sharing of risk often occurs post-origination. Indeed, lenders typically transfer risk in two ways: they share the risk by syndicating the lease amongst multiple participants, and they reduce risk exposure in a portfolio by syndicating deals that create unwanted exposure or no longer conform to the overall credit guidelines. But transference does not have to be reactive. The enterprise can have a transference strategy that leverages its ability to originate deals and/or its relationships with capital markets. Such strategies can be very proactive and growth-centric as opposed to simple risk mitigation.

Once again, data plays a central role in a risk transfer strategy. Deal participants and the buy-side of any syndication require hundreds of columns of high-integrity data per contract for any investment made or pool purchased. Again, data is the language of risk. How they analyze the data and make their buying decisions is commonly referred to as their “credit box.” Buyers have various ways to define their credit box, including different data types, as well as different methods for analyzing data to meet their risk-reward objectives.

“Every private equity deal is unique. There’s not a template.” said Eli Sethre, CFO at Channel in a recent ELFA article on private capital entering the equipment finance space. As more and more private funders (buyers) become capital partners this approach to defining buying criteria has resulted an exponential growth of the matching targets for practitioners of risk transference.

This is where AI converts a reactive approach to one of strategic intent by predicting which lender partners will be interested in either participating or buying a given contract. A proficient originator with data on past deal fundings can build AI models predicting which of the many “credit boxes” of the associated funders will fit a deal pre- or post-origination. The AI predictor will provide the probability of funding with each candidate, empowering the seller to use matching probabilities to judge the best partner for sharing or reducing their portfolio risk.

AI predictors identified and quantified delinquency risk during the macro effects of COVID stimulus enabling risk transference.

The AI models can also incorporate reactive risk management into the transfer strategy by predicting emerging delinquency and default probabilities within the portfolio due to economic circumstances. The figure above shows how one portfolio manager identified an asset exposure risk, a specific kind of trucking asset, and then reduced exposure by packaging a pool of those deals with a known buyer whose risk-reward profile matched the pool. The AI predictors forecasted the increase in delinquency risk throughout the COVID stimulus period, learning from the results over time (see bar charts below the curve) as a key part of a transference risk management strategy.

Engage and Manage It

In addition to those who work to avoid or transfer risk, some actively seek out and engage in risk for its rewards. Whether motivated by financial, personal or other gains, these players are confident in both their skills and judgment to enable them to navigate the best path to the desired outcome.

AI prediction enhances an engagement strategy by providing probabilities of the outcome distribution thus quantifying the risk while simultaneously identifying the engagements to avoid.

Risk takers rely on their judgment to avoid unwanted outcomes in the risk distribution. They know they can make the right choices and are not afraid of the unknown because they leverage and study their data, no matter what comes their way. Perhaps more than any other strategy, AI can help risk-takers. As the figure below shows, AI predictors both quantify the risk, i.e., the standard deviations of the outcome distribution, and identify the source. With low cost and prolific AI prediction judgment becomes even more valuable. Those who trust their judgement and have built organizations equipped with agile technology tools can quickly adapt to emerging threats as they engage the higher rewards of increased risk. One risk engagement CEO stated, “I just want to win more than I lose,” which was, indeed, a winning strategy.

Conclusion

Risk management is one of the most important planks of any equipment finance business platform. Peter Drucker said, “You manage what you measure,” so why isn’t more time spent measuring risk more effectively? Finance companies do measure almost everything, they are rich in data, and they buy even more data — Equifax, Experian, TransUnion, Oculus, Fusable, and the list goes on. However, while data analysis can provide statistics to help better understand the natural tendencies of processes and behaviors, distributions alone describe them; they do not predict. AI models quantify risk by providing statistically relevant distributions of the outcomes of each action, such as underwriting, funding, collections, and reselling. As such, they predict the likelihood of those outcomes and the best path forward for the enterprise. AI enhances the implementation of risk strategies, whether avoiding, transferring, or engaging.

Data is the language of risk. AI is telling the stories.

Sidebar:  

Fraud is a risk that everyone tries to avoid.

Fraud is a behavior that AI has modelled and helped manage for over three decades; it is still growing as a concern, so clearly AI is not a panacea. Fraud is a purposeful behavior that is counter to the goals of the enterprise as opposed to a result of circumstances or conditions. As such its form changes constantly, and it doesn’t follow the usual natural distributions of human behavior. Fraud and underwriting are almost mutually exclusive when it comes to data because bad actors will provide exactly the data underwriters, including AI agents, want to see to get approvals. AI will continue to be critical to identifying and avoiding fraud, fighting tech with tech, but fraud is a different kind of risk than payment default, so it must be added as an additional purpose-built function as part of an overall risk management strategy.