# AI Agents Cut 3-Day Portfolio Builds to Minutes — Jack McDonald, Apollo & Michael Manapat, Rowspace

## Executive summary

Apollo's structured credit and CLO management unit, Redding Ridge, demonstrated how AI agents are transforming highly manual, data-intensive financial processes. Previously, tasks like building mock portfolios or reconciling weekly trades required days of manual effort across multiple people. By implementing AI platforms that own and reconcile fragmented, proprietary data from dozens of sources, these processes are now automated and completed in minutes. The discussion emphasizes that successful AI adoption in finance requires deep domain expertise, 'boots on the ground' engineering, and building proprietary platforms rather than simply plugging into generic LLMs like ChatGPT.

## Key takeaways

- Portfolio Construction Time Reduction: Building mock portfolios for CLOs, which involves reviewing deep indenture documents and complex rules-based waterfalls, used to take portfolio managers 2–3 days in manual Excel models. With AI assistance, this process is reduced to minutes, allowing PMs to focus on the final 10% of judgment and refinement.
- Automating Trade Reconciliation: The weekly trade reconciliation process, which historically required multiple individuals to manually review bank statements and settlement data, is now automated and repeatable in minutes, shifting the team's role to quality assurance (QA).
- The Need for Proprietary Data Ownership: Generic LLMs (e.g., Claude, ChatGPT) are insufficient for mission-critical financial decisions because they lack access to proprietary, fragmented, and often conflicting internal data. Trust requires building a platform that owns the data and validates information from multiple sources (e.g., trade order management system, performance system, Bloomberg).

## Technical details

- CLO Portfolio Construction: CLOs are highly rules-based, with money flowing through specific waterfalls detailed in indenture documents. The process involves aggregating data from document libraries, trade systems, and real-time pricing to build complex, multi-credit portfolios.
- Data Fragmentation and Reconciliation: Financial data is often fragmented across various documents and systems. The solution involves using AI agents to act as data engineers, reconciling information from multiple sources (e.g., 20 different sources) to establish a single 'golden source of truth' for decision-making.
- AI-Native Platform Architecture: Apollo's approach is to build an 'AI-native' platform, moving beyond point solutions. This requires embedding engineers with the front office to understand the exact information flow, calculations, and data validation rules of the business.

## Practical implications

- For builders applying AI to finance, the most critical step is 'boots on the ground' time: spending ample time with front-office professionals to map complex, nuanced workflows and identify the true 'golden source of truth' for data.
- AI should be leveraged not just for efficiency (reducing OpEx), but to structurally create a durable competitive advantage by providing superior, validated information access.
- The focus must be on building proprietary, closed-loop data platforms that govern and reconcile information, rather than relying solely on general-purpose LLMs.

## Topics

AI Agents, Financial Technology (FinTech), Structured Credit, CLO Management, Data Engineering, Process Automation, LLMs, Rowspace, Apollo, Redding Ridge Capital

Source: https://www.youtube.com/watch?v=BOWYujhIQuo