Scott McKay is the CTO at Kickfurther. He specializes in building complex systems and the organizations that build the systems.
getty
My company, Kickfurther, has carved out a niche by connecting businesses in need of funding for their retail inventory with buyers of that inventory. A key component of this business model is the ability to perform financial risk assessments on these businesses to ensure that the inventory has a high probability of being sold.
While the tech world buzzes with excitement over AI and large language models (LLMs), my company has found that traditional machine learning (ML) methods, such as gradient-boosted decision trees, remain highly effective and reliable for such assessments.
Machine Learning: The Original And Proven AI
Old-school machine learning might not have the allure of the latest AI trends, but it has consistently proven its worth. To get started with ML, the adage “what’s old is new” applies: Like with newer AI models, data is invaluable for harnessing the power of ML to evaluate financial risks.
Data continues to be a goldmine for qualification and servicing, yet it is often underused. Massive opportunities remain untapped, which you can capitalize on with sophisticated data-handling techniques. To conduct thorough risk assessments, for example, it’s important to draw from numerous data sources. For us, this includes but is not limited to:
• Banking transactions
• Bookkeeping records
• Order and inventory information
• Warehouse data
Diverse datasets are crucial for creating a comprehensive picture of a business’s financial health.
Once you understand the data you need, one of the best ways to streamline data acquisition and minimize manual oversight is to have an asynchronous architecture with numerous “connectors” that feed into a data lake. This setup allows for continuous data streaming of data, enhancing efficiency and accuracy.
Using “connectors as a service” tools such as Fivetran can allow you to easily pull from various data partners. For partners not supported by Fivetran, you will need to develop custom connectors. All this data should then flow into a data lake to be stored and managed.
Building And Iterating ML Models
At the heart of our risk assessment process are ML models, which we build and deploy iteratively.
By using historical data dating back several years, you can run retrospective experiments to validate and refine your models. This historical data should be enriched with real-time inputs from the data lake, allowing you to analyze thousands of parameters and identify those with the best predictive value.
By capturing new data sources—combined with ongoing data engineering to improve model performance and keen account monitoring—both improvement or degradation show up quickly in the model, allowing you to patch an improved new version.
Our models are integral to our decision support systems. Moreover, the same data streams can automate various operational processes to ensure efficiency, reliability and early detection of problems.
These kinds of insights allow you to empower your teams or customers by presenting them with relevant data that enables informed decision-making. That said, it’s crucial to continue to monitor the performance using additional ML tools to identify and take necessary recovery actions promptly.
Navigating Change Management In Model Deployment
Deploying major revisions of risk models can present interesting challenges.
With our model, for instance, buyers create a “consignment opportunity” (co-op) to purchase inventory from a brand. Companies with previous co-ops might find themselves in different risk—and therefore pricing—buckets when new versions of the models are implemented. Since our goal is to continually identify less risky co-ops, scores tend to drift downward as we select for better and better co-ops.
One way to manage this type of concern is to create short-lived “grandfathering” policies, ensuring a smooth transition. In this case, you can retain previous customers whose good track records might not be reflected in a conservative risk model.
Despite the complexity of these operations, we achieve our goals with an unusually small team. For other small teams, this becomes possible by making significant investments in automating the data pipeline and in tools that facilitate rapid iteration. The effectiveness of those tools enables teams to more rapidly push the boundaries of what’s possible in financial risk assessment.
All this is to say, while the allure of new AI technologies is undeniable, the proven power of “old school” machine learning with remains a cornerstone of success. By leveraging diverse data sources, sophisticated integration techniques and iterative model development using proven ML techniques, you can innovate and excel in the realm of financial risk assessment.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
{Categories} _Category: Inspiration{/Categories}
{URL}https://www.forbes.com/councils/forbestechcouncil/2024/10/25/how-to-leverage-machine-learning-for-financial-risk-assessment/{/URL}
{Author}Scott McKay, Forbes Councils Member{/Author}
{Image}https://imageio.forbes.com/specials-images/imageserve/660c293df8aa6d92af1ece44/0x0.jpg?format=jpg&height=600&width=1200&fit=bounds{/Image}
{Keywords}Innovation,/innovation,Innovation,/innovation,technology,standard{/Keywords}
{Source}Inspiration{/Source}
{Thumb}{/Thumb}