--- title: "AI for Private Credit Funds in New York | By The People Tech" description: "AI for New York private credit and direct lending funds: credit agreement and covenant review, due diligence, portfolio monitoring and LME exposure." url: https://bythepeople.tech/industries/private-credit --- [Industries](/industries) Private credit “AI can’t read a credit agreement.” # Every covenant lives in a _definition_. Read that first. By The People Technologies works with private credit and direct lending funds in New York on the documents a loan is made of: credit agreements and covenant definitions, diligence on a new borrower, quarterly compliance certificates, and the LME exposure in amendments. Credit teams learn a cited first-pass method in a [workshop](/workshops) on their own agreements, and a [forward deployed engineer](/forward-deployed-engineer) can build covenant monitoring across the book with them. The credit officer still signs. By [Tim Miner](/about-us/tim-miner), founder and CEO · Published 21 October 2026 What a loan is made of 1. At origination Term sheet and draft agreement What the definitions let the borrower do 2. In diligence Quality of earnings Add-backs against the cap 3. Each quarter Compliance certificate The ratio, by this deal’s definition Where we usually start 4. When it happens Amendment or waiver What it changed, and what it left valid 5. Monthly Borrower reporting Headroom before it becomes a default Credit is where a model’s confident wrong answer costs the most, because the value sits exactly where a deal departs from the market: the negotiated definition, the carve-out, the basket, the test date one quarter later than usual. A model answering from the general case misses every one of them. The agreement ## How can AI review a credit agreement without missing the _covenant definitions_? An analyst pastes a credit agreement into the firm’s AI tool and asks whether the leverage covenant is breached. The answer is fluent, confident and wrong, because the agreement defines EBITDA with six add-backs the model never saw. The analyst catches it, and stops trusting the tool on anything they could not check themselves. The fix is in the request. The whole agreement goes in, with the definitions schedule, every amendment and the compliance certificate. The model reads the definition before it computes the ratio, cites the section for every term it uses, and answers “not in the document” where the papers are silent. Credit teams do this in the [workshop](/workshops) on their own agreements. The [accuracy guide](/guides/ai-accuracy-investment-firms) gives all five fixes. Due diligence ## How do credit funds use AI for _due diligence_ on a new borrower? Under deal pressure the term sheet gets read closely and the rest gets sampled. A model used well reads all of it, against the fund’s own credit questions, and hands the team a first pass with every finding tied to its clause. ### Add-backs against the cap Each adjustment in the quality of earnings report, set against what the draft agreement’s EBITDA definition allows and where it caps them. ### Baskets and carve-outs The investment, debt and restricted payment capacity the definitions create, added up across the documents rather than read one clause at a time. ### Unrestricted subsidiaries What the agreement lets the borrower move outside the lenders’ reach, and on what terms. ### What the data room lacks The schedules, consents and side letters the agreement refers to and the data room does not contain. The credit team still writes the memo and makes the call. What changes is that the first week goes on judging findings instead of finding them. Monitoring ## Can AI monitor _covenant compliance_ across a loan book? Covenant verification is checked against the document that defines the covenant. The definitions are negotiated deal by deal, and the check recurs every quarter for the life of the loan. That makes it reading at volume with a judgment at the end, which is where a model earns its place. ### 1. The certificate arrives With the borrower’s figures and its own calculation of each ratio. ### 2. Recompute by the deal Each ratio recomputed from that deal’s definitions, with the amendments that are still in force. ### 3. Flag the difference Every gap between the borrower’s number and the definition’s, cited to the clause. ### 4. The officer reviews A person reads every flag and decides what to raise with the borrower. Covenant monitoring across a credit book is one of the processes a [forward deployed engineer](/forward-deployed-engineer) rebuilds with the team that runs it. The team builds it and owns it. LME exposure ## How should a credit fund review its book for _LME exposure_? A liability management exercise turns on a few words in the debt contracts. A private credit team measures LME risk by knowing which borrowers’ documents leave room for one: unrestricted subsidiary capacity, investment and transfer baskets, and whatever the definitions allow. The exposure usually sits outside the main credit agreement, in the amendments, side letters and subsidiary documents. For one private credit team, the read used to take a week per contract. Now it runs as a structured first pass across the book, every finding cited to the clause it came from, so counsel starts from the questions. Our [LME review](/insights/lme-review) sets out where the exposure sits and what a full read takes. Tool or people ## Should a credit fund buy an AI tool for credit or _train its people_? Buy whatever the firm’s compliance will approve, then train the people who read the agreements, because the tool is rarely the gap. A specialised wrapper decides what you can ask it and changes when its vendor decides. A credit analyst who can specify the task, give the model the definitions and check the citation can do that on whichever model the firm uses next. Most funds begin with a [workshop](/workshops) for one credit team and decide from there. The [retainer](/retainer) carries the method across the rest of the firm, measured on how many people use the tools well. **A credit fund with tools bought and left, and one whose credit teams use them well** | | Tools bought and left | Credit teams who use them well | | ---------------------------- | ------------------------------------------ | -------------------------------------------------------------------------- | | The leverage covenant | Answered from what covenants usually say | Computed from this deal’s EBITDA definition, each add-back cited | | Add-backs in diligence | Checked on a sample under deal pressure | Every adjustment set against the definition and its cap | | The compliance certificate | The borrower’s calculation, taken as given | Recomputed by the deal’s definitions, differences flagged for the officer | | Amendments and side letters | Skimmed, or not found | Linked to the base agreement, with what each one changed | | LME exposure across the book | A week per contract, so rarely done | A structured first pass across the book, every finding cited to its clause | | When the model is wrong | The analyst stops trusting it | The citation shows where, and the instruction gets fixed | We have delivered over 150 workshops at more than 100 financial services firms. We are in the Claude Partner Network, Select tier. ## Questions private credit funds ask us ### How can we use AI for private credit due diligence without missing covenant details or add-back caps? Give the model the whole draft agreement with its definitions and every schedule, and the quality of earnings report beside it. Ask it to set each add-back against the EBITDA definition and its cap, cite the clause for every finding, and say what the documents do not answer. The credit team reads that first pass against the sources and makes the call. ### Why does ChatGPT get the numbers wrong on our credit agreements? Usually because it was never shown the definition. Asked about a covenant without the definitions schedule and the amendments in the conversation, a model answers from what covenants usually say, and the value in credit sits in where this deal differs. Give it the documents, make it cite the section for every term, and let it say “not in the document”. ### Are there AI consultants that specialise in private credit funds? By The People Technologies is a New York AI adoption consultancy for private equity, credit and hedge funds. Credit teams work on their own agreements and compliance reporting in our workshops, and our forward deployed engineers build processes such as covenant monitoring across a credit book with the team that runs them. ### Can AI monitor covenants across our whole loan book? It can do the reading every quarter: recompute each ratio from that deal’s own definitions and amendments, and flag every difference from the borrower’s certificate with the clause cited. A credit officer reviews each flag and decides what to raise. The fund’s own team builds it, so it keeps running after we leave. ### Does AI replace the credit officer or counsel? No. Every finding is linked to the clause it came from, so the officer and counsel review answers rather than searching for them. They still sign off every flag, and the fund still makes the call. ## Sources * [Anthropic, “Reduce hallucinations”, Claude documentation.](https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations) * [Paul, Weiss, “Appellate Review of Uptier Transactions: Serta and Mitel Decisions Reversed on Appeal” (31 December 2024).](https://www.paulweiss.com/insights/client-memos/appellate-review-of-uptier-transactions-serta-and-mitel-decisions-reversed-on-appeal) * [Anthropic, Claude Partner Network directory: By The People Technologies, Select tier.](https://partnerhub.claude.com/directory/partner/by-the-people-technologies-partner) ## Read next * [Counsel read the credit agreement. The exposure is usually in the rest.](/insights/lme-review). Where LME risk sits in the debt contracts, and what reading every one of them takes. * [Making AI accurate on investment documents](/guides/ai-accuracy-investment-firms). The leverage covenant example in full, and the five fixes. * [Forward deployed engineers](/forward-deployed-engineer). The engagement that builds covenant monitoring with the team that runs it. * [AI for private equity firms](/industries/private-equity). The same method on the CIM, the IC memo and portfolio reporting. ## Bring us one agreement. The one the tool got wrong, and the person who caught it. We will show you on that agreement what a cited first pass looks like, and if we are the wrong firm for it, we will say so. [Book 30 minutes with Tim](https://calendly.com/timminer-1/chat-with-tim) [hello@bythepeople.tech](mailto:hello@bythepeople.tech)