Tusk Private: AI on Your Firm's Own Credit Deals
Tusk Private turns a firm's own credit agreements into a private, queryable intelligence graph: plain-language answers, cited to the firm's own clauses.
The Answer Is Somewhere in Your Deal Files. That Is Not the Same as Having It.
The deal files are a firm's proprietary dataset, though hardly anyone at a credit firm thinks of them that way. Ten or fifteen years of agreements the firm negotiated line by line, plus every amendment it signed and every definition it argued about, sit on a shared drive somewhere, and between them they describe how this firm trades terms with these sponsors more accurately than any vendor product will, for the obvious reason that vendors were not in the negotiations. What the files will not do is answer a question, at least not without an analyst and most of a week.
If one had to dig into this maze only once or twice a year, this is fine, but what investors are asking for now, is at a different cadence. Once sponsors began using unrestricted subsidiaries and debt baskets to move collateral out of reach, the dull machinery of an agreement became the part that decides who gets paid, and deal done way past in 2017 becomes your new go-to reference. The documents have not helped by getting longer and less alike, with every sponsor's counsel now drafting its own. So the questions multiply, and they multiply on several desks at once: the investment committee wants precedent before it signs off, risk wants to know how much of the book carries the provision that just blew up in somebody else's deal, and the LPs who used to ask about marks now ask about documentation and want the answer as of a date. All of it is portfolio-wide, and the files are organized one deal at a time, which is how a reasonable question turns into a week of an analyst's life.
What a Simple Question Costs
Take a question any credit committee might ask: which of our sponsors has negotiated the loosest restricted payment baskets since 2024?
Somebody has to answer it. The first step is remembering which deals might be relevant, which depends on who is in the room and how long they have been at the firm. Then the folders open. Each candidate agreement runs to hundreds of pages, and the restricted payments section cannot be read on its own, because the basket in Section 6.04 leans on definitions in Section 1 that an amendment may have quietly restated since closing, and that amendment is rarely filed anywhere near the agreement it modifies. The analyst finds the term, records it in a spreadsheet, and moves to the next deal, where the same concept hides under different language in a different section. A week later there is an answer, accurate as of that Tuesday, in a format that will be rebuilt from scratch the next time a similar question lands.
Eric Ball, formerly Treasurer of Oracle, put a number on this kind of work in his interview with us: finance teams spend over 90 percent of their time assembling data rather than analyzing it. He was describing treasury systems, where the data at least arrives in fields. In credit the numbers have to be read out of prose first, which is why the assembling is measured in weeks and the analysis suffers purely because of shlep.
Why the Obvious Fix Never Worked
Pointing software at the deal files is not a new idea. Four problems have kept it from working, and each one defeats a partial solution.
The first, and the most underestimated, is aggregation. Most firms do not have a dataset; they have a filing system, one where executed agreements sit beside drafts that were never signed, waivers survive as email attachments, and the full sequence of amendments to a facility is known only to whoever papered it. Before anything can be asked, every document has to be identified as operative, tied to the deal and the chain it belongs to, decomposed into comparable attributes, and held in one structure. Without that foundation, search is just faster browsing.
The second is what credit documents are. An agreement is not a linear text. Its meaning is assembled from a definition in one section, a basket in another, a test in a third, and a side letter that modifies all three. A keyword search returns the original definition with full confidence, unaware that amendment seven changed it. Tools built for ordinary documents fail here quietly, which is worse than failing loudly.
The third is trust. In most industries a roughly right answer is useful. In credit, approximately correct is same as definitely wrong. A misread basket is discovered when the borrower does something the firm believed it could not do. A system that cannot show exactly where its answer came from is not saving work. It is adding risk.
The fourth is privacy. These documents are confidential, and usually restricted by their own terms. They cannot be pasted into a consumer chatbot, and they cannot feed a pooled model where the firm's negotiating history quietly educates the market. For many firms this problem alone ended the conversation.
Tusk Private: The Whole Book, One Question Away
Tusk Private takes the engine we built for credit, trained on $5 trillion of debt, and points it at your firm's own documents. The hard work of re-positing thousands and thousands of pages into a query-able format - Executed versions are separated from drafts, amendment chains are put back in sequence, definitions are resolved to the version in force, covenants are mapped as logic rather than prose, and every provision is connected to the deals, sponsors, and sectors it belongs to - precede the first question. Credit analysts verify the extraction before it enters the graph, and new amendments and waivers are ingested as they arrive, so the answer on Friday accounts for the consent that circulated on Thursday. This work takes, thankfully, only a few days at CredCore's end, and after that, anyone in the firm can ask a question in plain language.
Every answer traces back to the firm's own clause, with the source language one click away, so the analyst verifies rather than trusts. Simple lookups come back in seconds. The multi-dimensional screens that used to consume a week come back in minutes.
The graph is the firm's alone. Nothing is pooled across clients, every question and answer is auditable, and CredCore holds SOC 2, ISO 27001, and ISO/IEC 42001 certifications. When a comparison against the public market is useful, Tusk Private can put the firm's own terms next to public comps drawn from the corpus behind Tusk Liquid, which covers the deals across the entire liquid market.
What Teams Ask of Tusk Private
A credit committee, mid-debate, wants to know which sponsors in the book have taken the most aggressive covenant packages since 2024, and gets a ranked answer with clauses attached before the meeting ends. A portfolio manager, reading about someone else's blocked dividend, asks which of the firm's own agreements leave the most room for leakage through restricted payment baskets. Ahead of a new deal, an analyst screens the whole book for agreements that are simultaneously cov-lite, weak on MFN protection, and carrying more than $500 million of free-and-clear capacity, a question nobody would have commissioned when it cost a week.
When a question gets answered in minutes, teams ask questions they never used to ask. The screen that was too expensive to run becomes the screen that runs before every IC. And when the next deal arrives and the sponsor grid lands on the desk, the firm's entire negotiating history is already in the room.
Your firm has spent years writing the best reference book on its own market. Tusk Private is how it finally gets read.