Not all AI-powered and Agentic services are the same. This is especially true when managing complex loan models and workflows associated with commercial, corporate and private credit.
Key differences frequently derive not from the particular model being used but from the underlying software and systems with which the model is interacting. Having an MCP server is an important step but far from the last in enabling AI and Agentic services for loan operations and portfolio management. When working with AI tooling, the underlying loan models and workflows matter and matter more than may be expected.
Band-Aids and Gaps
While AI tools, skills and plug-ins can be useful to put band-aids on gaps in the underlying software managing a commercial, corporate or private credit loan book, two material problems begin to emerge. First, cost. When a complex loan structure is not supported directly in the underlying software, like for example, a PIK, unitranche or non-pro rata funded loan, the use of AI tools to stitch together the functionality for a system to act “as if” that loan structure is supported becomes expensive quickly. Every interaction with that loan requires meaningful expenditure of tokens to maintain, manage, query and update that loan. From there, running reports, taking loan actions, or engaging in routine, let alone more sophisticated, portfolio management, require constant and material token usage to return proper values. Second, indeterminate outcomes. AI tools on top of AI tools on top of other AI tools can become brittle and are more likely to return values for reporting or taking loan action that are probabilistic in nature rather than explicit. Worse, at scale, across 100s (if not 1000s) of different credit facilities each with a series of AI-dependent tooling, increases the risk of errors — errors that are difficult to locate and may be impossible to completely remove. Ultimately, this can drive a failure of confidence that can result in removing AI tooling altogether as the prototype does not scale to production.
LoanStreet CLS Complete Loan Model and Workflows
LoanStreet CLS supports the most complex loans for commercial, corporate and private credit along with intelligent workflows matching the real-world behavior for loan operations and credit administration teams. Our loan model natively supports these loan structures and workflows upon which Agentic (or AI powered tooling) can be leveraged. Because the Agentic (or AI tool) is interacting with explicit models and workflows, permissions, data governess, confidence in deliverables and actions enable scale and low-cost and efficient token usage. The AI agent is working directly against an MCP server for the explicit loan values and workflows minimizing token expense at every turn. The result: LoanStreet CLS is the premier solution for efficient token usage — delivering a powerful and cost effective Agentic servicing for commercial, corporate and private credit loans.