{ "video": { "id": "BOWYujhIQuo", "title": "AI Agents Cut 3-Day Portfolio Builds to Minutes — Jack McDonald, Apollo & Michael Manapat, Rowspace", "duration": 1137, "upload_date": null, "channel": "AI Engineer", "source": "AI Engineer", "event_format": "" }, "analysis": { "video_id": "BOWYujhIQuo", "title": "AI Agents Cut 3-Day Portfolio Builds to Minutes — Jack McDonald, Apollo & Michael Manapat, Rowspace", "one_liner": "At Apollo's Redding Ridge (about $50B in CLOs), Rowspace's agents cut CLO mock-portfolio builds from 2–3 days of manual Excel work to minutes. Trust came from working out, with the actual teams, which of up to 20 conflicting sources is the golden source of truth, not from a general-purpose LLM.", "summary": "Michael Manapat (Rowspace co-founder) interviews Jack McDonald (Apollo principal and COO of Redding Ridge Capital, Apollo's captive structured credit and CLO manager) about moving AI in finance from nice-to-have to mission-critical. McDonald gives two concrete cases: CLO portfolio construction and weekly warehouse trade reconciliation. He argues that trust comes from owning your own data, with a vendor sitting with the teams daily to work out the correct calculations and the authoritative source for each data point. He thinks finance is still very early in AI adoption, and that AI becomes a durable competitive edge only if it is adopted properly, by giving teams accurate information at their fingertips.", "key_points": [ "Redding Ridge is a separate operating company captive to Apollo. It manages about $50B across structured credit, mainly CLOs, with roughly thousand-name loan pools and thousands of data points a day. That data is private and scattered across documents, unlike public equities or fixed income.", "Portfolio construction: CLOs are rules-based, with tests and payment waterfalls buried in indenture documents and legal agreements. PMs used to read those documents and build mock portfolios of 200–400 single credit names in Excel, which took 2–3 days. With Rowspace it takes minutes using real-time data, getting to 80–90% completion, and the PM uses their judgment on the last 10% to 'make it theirs'.", "Data for portfolio construction is pulled from doc libraries, trade and compliance systems, and real-time pricing.", "Warehouse purchase activity: the weekly trade reconciliation used to be manual. Staff went into bank accounts and statements to see where trades settled, then emailed a recap to management, taking many hours a week from 2–3 people. The same report now runs in Rowspace, can be regenerated in about a minute, is automated, and the team only QAs it daily.", "McDonald says general LLMs like Claude and ChatGPT help with research, new ideas and getting up to speed on a product or investment quickly, but they work off general information, not your own data.", "Trust in Rowspace came from 'boots on the ground': Manapat's team sits with the PM and finance teams daily to learn the calculations and decide which source overrides the others. Sources include the trade order management system, the performance system, Bloomberg and pricing indices.", "Manapat says Apollo was AI-forward early, before Claude Code and while Cursor was still in the 'tab completion era'. Its stance was to rebuild the firm in an AI-native way, with a platform rather than point solutions. McDonald credits a flat, open, innovative culture.", "McDonald thinks finance, especially asset management and investment banking, is still very early. He calls model creation and research aggregation 'lower value add' for now, and expects less junior support over time for functions like those." ], "takeaways": [ "When applying AI to finance, embed with the end users (PMs, finance teams) daily to learn their workflows and calculations. McDonald credits this for Rowspace being ahead of roadmap and expectations.", "Build the data layer before automating: for each data point, decide with the domain experts which of the many sources is the golden source of truth and should override the others.", "Aim for agents to produce about 80–90% of the work product (mock portfolios, recurring reports) and keep humans for judgment on the last 10% and for daily QA.", "Assume that odd-looking structures in financial workflows probably exist for a reason. Understand them before redesigning them.", "Use the time saved on routine tasks to question why processes exist and whether to change them or bring in new technology." ], "topics": [ "ai in finance", "agents", "data engineering", "data quality", "clo portfolio construction", "reconciliation", "enterprise ai adoption", "forward-deployed engineering" ], "tools": [ "Rowspace", "Apollo", "Redding Ridge Capital", "Claude", "ChatGPT", "OpenAI", "Claude Code", "Cursor", "Bloomberg", "Excel" ], "quotes": [ { "text": "for one piece of data that might come from 20 different sources and there's only one that's actually right", "at": "08:19", "url": "https://www.youtube.com/watch?v=BOWYujhIQuo&t=499s" }, { "text": "that allows these PMs to kind of build something to call it 80 90% completion and then kind of use their judgment and their thought to further refine that portfolio that last 10% to make it theirs", "at": "04:59", "url": "https://www.youtube.com/watch?v=BOWYujhIQuo&t=299s" }, { "text": "typically things might look a little bit novel or strange and out of place, but it's probably actually there for a reason or structured in a specific way", "at": "17:54", "url": "https://www.youtube.com/watch?v=BOWYujhIQuo&t=1074s" }, { "text": "no given firm now has an advantage because they use Excel", "at": "13:39", "url": "https://www.youtube.com/watch?v=BOWYujhIQuo&t=819s" } ], "words": 3932 }, "summary_url": "/#BOWYujhIQuo", "transcript": { "html": "/transcripts/BOWYujhIQuo.html", "txt": "/transcripts/BOWYujhIQuo.txt", "vtt": "/transcripts/BOWYujhIQuo.vtt" } }