Bonds are a critical part of diversified investment portfolios. Their steady income stream and relatively low risk compared to equities make them an especially important component of pension and retirement planning.
Bonds are issued by different entities such as governments, municipalities, and corporations. Each type of issuer presents a different level of risk and tax treatment. The issuer’s creditworthiness also affects a bond’s risk and yield, with higher-rated bonds considered safer but with lower returns, while lower-rated bonds offer higher yields and a correspondingly higher risk of default.
To protect employee benefits and guard against mismanagement of funds, the US government in 1974 established the Employee Retirement Income Security Act (ERISA). It’s a federal law that provides protections for individuals in private-sector retirement and health plans. Employers and plan providers must comply with ERISA’s stringent requirements, especially when it comes to the investments chosen for these plans, such as bonds. This can be a complex and resource-intensive process, often involving extensive manual audits and detailed risk assessment processes.
“Not every bond has to comply with ERISA, but benefits administrators use ERISA to ensure that bonds meet internal rules,” says Soundarapandian. A, head of the data science practice at Hexaware. “Bonds need to be ERISA-compliant to be purchased by institutional investors.”
Verifying that bonds are ERISA-compliant has historically been a manual process involving a thorough review of documentation and legal history. Bond prospectuses can range from 100 to 400 pages in length and must be reviewed against an extensive checklist of ERISA requirements. Auditing a 401(k)-plan document manually can cost up to $7,500 and consume about 45 hours of a skilled professional’s time. Yet despite the critical nature of compliance verification, there are few automated solutions that streamline the process.
Generative AI is changing the equation, though. It has emerged as a transformative force in managing bond investment risks by automating and optimizing complex analytical tasks. Gen AI models can sift through vast quantities of unstructured data related to bonds, identify critical information, and compare it against risk parameters.
“It knows exactly where the compliance information is in a 300-page document,” Soundarapandian says. “It will fetch the relevant portions, check against the compliance checklist, and deliver a recommendation in less than 20 seconds.”
Hexaware’s BondReco is an innovative application of Gen AI in this area. The software automatically classifies bonds into ERISA-eligible and non-eligible categories to ascertain their investment safety. That can slash thousands of dollars in audit fees. Using technologies like Azure Form Recognizer and Azure OpenAI, BondReco not only enhances accuracy in digitizing data but also crafts reports that justify investment classifications.
BondReco is built on top of a highly secure enterprise large language model accessed through Microsoft’s Azure cloud. Hexaware conducted extensive fine-tuning to adapt the model to different industries and company types. “We have covered all the sectors and vertical markets,” Soundarapandian says. “The application explains every decision and produces an audit trail on the back end.”
With BondReco, the cost for ERISA clause detection in a document can be reduced to approximately $500, demonstrating the financial and operational benefits of embracing AI-driven automation in the landscape of financial risk assessment.
As powerful as AI is at reducing the time and cost of ensuring ERISA compliance, Hexaware goes an extra step to ensure quality. “We always have a human in the loop to check the output,” Soundarapandian says. That should give bond-holders an extra measure of confidence in automation.
To find out more on how to gauge the safety of your ERISA bonds through BondReco, please visit Microsoft Marketplace.
Artificial Intelligence