Optimizing Asset-Backed Lending with Machine Learning and Outerbounds

Setpoint: Building Trust in the Credit System

Setpoint is a leading fintech company on a mission to modernize the hidden, cumbersome processes behind trillions in daily financial transactions. Founded by industry veterans with a track record of successful startups, Setpoint provides innovative solutions that digitize, organize, and verify data, creating a real-time asset source of truth for the credit industry.

Setpoint's platforms, Asset OS and Capital OS, automate funding flows and compliance, setting a new standard for fast, accurate, and effortless credit transactions. By transforming manual workflows into efficient, automated solutions, Setpoint significantly reduces costs and enhances speed for both borrowers and lenders.

Addressing Complex Challenges in Credit Infrastructure

In asset-backed lending and credit markets, managing transactions often relies on outdated methods like email, Excel, and manual data entry, which leads to frequent errors and inefficiencies. Originators and lenders face significant challenges, such as navigating complex credit agreements that consist of hundreds of pages filled with nuanced terms, constraints, and criteria. Asset allocation, the process of distributing assets across multiple warehouse lines or funding sources, is typically done manually, resulting in suboptimal capital efficiency and increased compliance risks. The sheer complexity of these allocations creates a search space with an exponential explosion of possibilities — for example, even a simple scenario involving three warehouses and 500 assets results in more potential combinations than the number of atoms in the universe.

Moreover, manual data processing compounds these issues. Handling large volumes of documents like PDFs, Word files, and Excel sheets for data extraction and verification introduces further inefficiencies. This document-heavy process is not only slow but also prone to human error, especially when critical financial information needs to be accurately extracted, verified, and entered. The lack of automation in these areas has traditionally created bottlenecks that hamper both speed and accuracy in managing complex financial transactions.

Harnessing Machine Learning at Scale: Setpoint's Journey with Outerbounds

To overcome those challenges, Setpoint looked to integrate advanced machine learning techniques into their platform. The mission was clear: automate and optimize processes like asset allocation and document extraction to deliver unparalleled efficiency and accuracy for their clients.

Exploring the Landscape of Workflow Orchestration

Setpoint faced the challenge of finding a scalable platform to run their computationally intensive machine learning tasks. Russell Brooks, who leads machine learning initiatives at Setpoint, drew upon his experience with Metaflow from his time as principal machine learning engineer at Realtor.com. Initially, the team considered self-hosting Metaflow, but managing the necessary infrastructure would have diverted valuable resources from their core objectives.

Seeking alternatives, they evaluated other workflow orchestration platforms like Dagster and Prefect. While each had unique features, none fully aligned with Setpoint's requirements for scalability, seamless integration, and minimal infrastructure overhead.

Outerbounds offered the scalability and flexibility Setpoint needed without the complexities of self-hosting, plus the familiarity of being a managed platform built on top of Metaflow. Its ability to integrate smoothly with their AWS environment was a significant advantage, ensuring compliance and operational consistency.

"Outerbounds combined the robustness of Metaflow with the convenience of a managed service," Russell explained. "It allowed us to focus on delivering value to our clients rather than spending time on infrastructure management."

Navigating Non-Convex Challenges: Advanced Asset Allocation Strategies

With Outerbounds integrated into their workflow, Setpoint tackled a significant challenge: optimizing the allocation of assets to various funding sources while adhering to complex financial constraints and client-specific objectives. Traditional methods relied on manual adjustments in spreadsheets, which were inefficient, time-consuming, and prone to errors.

Recognizing the complexity of the task—a constrained, finite-horizon, multi-objective optimization problem—Setpoint implemented advanced algorithms to find optimal, compliant asset assignments for their clients. The problem was inherently non-convex, involving multiple competing objectives and a vast solution space that could not be efficiently navigated using traditional gradient-based methods.

To address this, Setpoint employed gradient-free techniques, specifically genetic algorithms, which are well-suited for exploring complex, non-linear landscapes with multiple objectives and constraints.

"Our optimization tasks involve balancing numerous client-specific objectives and constraints," explained Russell. "By using advanced variants of genetic algorithms, we're able to efficiently navigate the solution space to find optimal asset allocations that meet all requirements."

These models require substantial computational power to process a multitude of variables and scenarios in parallel, which Outerbounds enables with effortless scaling of computational resources, running their optimization algorithms on large clusters without the need to overhaul their existing infrastructure.

"Outerbounds allowed us to scale up resources as needed," Russell noted. "This capability was crucial for delivering timely and accurate recommendations to our clients, ensuring they achieve optimal capital efficiency while maintaining compliance."

By harnessing these advanced algorithms and leveraging Outerbounds for scalable compute, Setpoint significantly improved the efficiency and accuracy of their asset allocation process. This not only enhanced their service offerings but also provided tangible value to their clients in the form of optimized capital utilization and reduced operational risk.

Leveraging AI for Efficient Document Extraction and Verification

Another critical area where Setpoint leverages machine learning is in aiding with the extraction and verification of data from a multitude of document formats—PDFs, Word files, Excel sheets, and even images of physical documents. Utilizing large language models (LLMs), optical character recognition (OCR), and various natural language processing (NLP) and image processing techniques, Setpoint streamlines the process of creating reliable sources of truth for asset data.

By streamlining document processing, Setpoint addresses the challenges of handling vast amounts of complex financial documents that were traditionally managed through manual data entry—a process prone to errors and inefficiencies. This automation enhances data accuracy, reduces operational bottlenecks, and accelerates credit transactions. "Outerbounds gives us the computational horsepower to transform unstructured documents into structured, validated asset data," said Russell.

The Impact of Outerbounds on Setpoint's Innovation

Integrating Outerbounds into Setpoint's tech stack had a transformative effect on the company's operations and innovation capacity.

"Not having to deal with infrastructure is a game-changer," Russell remarked. "Outerbounds handles it all, allowing us to dedicate our energy to developing solutions that truly make a difference in the credit industry."

‍