Should a Private Credit Fund Build Its Own AI Tools or Buy Them?
Dr. Leigh Coney
Founder, WorkWise Solutions
November 10, 2026
6 min read
TLDR: A private credit fund of 10 to 40 people deciding whether to build or buy AI should configure first, buy second and build last. Configuring means setting up Claude or ChatGPT on your own templates and screening criteria, with no code, at $20 a seat a month billed annually. Buy specialist software for a job that needs one shared record, such as covenant tracking. Build only the piece nothing on the market covers. WorkWise Solutions does the configuring on its AI Operating Partner retainer, from $10,000 a month for firms of up to 25 people.
On the Core tier of my AI Operating Partner retainer at WorkWise Solutions, $10,000 a month for firms of up to 25 people, one build slot is always open for a configured tool such as deal screening, built on your own criteria and templates. A connector to a system with no ready-made one is a Custom Build, from $75,000, fixed once scoped. I sell no licenses and take no referral fees.
Table of Contents
1. The Short Answer
For a private credit fund of 10 to 40 people, the order is configure, buy, build. Configure Claude or ChatGPT on your own templates and screening criteria first, buy specialist software for a job that needs one shared record, and build only the piece nothing on the market covers.
Configuring means writing down how your firm does a job (the screening criteria, the memo template, the spreading layout) and saving it inside an AI assistant such as Claude or ChatGPT, so the assistant works your way every time. Nobody writes code.
It is cheap to try. A business plan of Claude or ChatGPT costs $20 a seat (one person's license) a month, billed annually, so a firm of 20 pays $4,800 a year.
The question usually arrives as build or buy. A lender with no engineers has a third route, and it comes before both.
2. Three Routes Compared
The three routes on published prices:
| Route | What it is | What it costs (published) | Who keeps it working | Fits when |
|---|---|---|---|---|
| Configure a general assistant | Claude or ChatGPT on a business plan, with your criteria and templates saved as instructions | $20 a seat a month billed annually: $2,400 a year for 10 people, $9,600 for 40. Plus setup time | One named owner on your team | The job is reading, checking and drafting from documents, and an analyst checks the output |
| Buy specialist software | A product built for one credit job, such as covenant tracking, portfolio monitoring or loan administration | By quote. No vendor in these categories publishes a price | The vendor, while your team keeps the data current | Many people and the auditors rely on one record, with history and alerts |
| Build custom | Software written for your fund alone, such as a read-only connector from Claude to your loan system | From $75,000, fixed once scoped | Whoever built it, or engineers you hire. Agree who before the build starts | Nothing you can configure or buy covers the piece |
Read the fourth column before the third. A tool with nobody to keep it working stops matching your documents the first time a template or a credit agreement changes.
Covenant tracking shows where configuring ends and buying begins. Claude can list each covenant and recheck a compliance certificate, but it keeps no test history, so tests that must run across the whole book with history and alerts belong in covenant software. When Should a Private Credit Fund Move Off Spreadsheets? lists the signs that it is time to buy.
3. The Deal Screening Example
Suppose teasers and CIMs arrive faster than your team can read them, and you want a tool that screens each one against your criteria. For a small fund, that tool has three parts:
- Your criteria, written down. Minimum EBITDA, a ceiling on debt to EBITDA, check size, and the industries you avoid.
- A one-page output. The same headings every time, each criterion marked pass, fail or not stated, and the page each figure came from.
- An analyst's check. Every figure against its page before the screen reaches a partner.
The assistant already does the reading and the writing. What makes the tool yours is the criteria, and those fit on a page your partners could dictate this afternoon.
In Claude, a saved set of instructions like this is called a skill. What Are the Best Claude Skills for Private Credit Firms? explains how to write one.
Screening suits this route because it passes the test in PwC's Global Private Credit Survey 2026: "If a task is document grounded and verifiable, AI is usually ready now." Buy a screening product only when screens must feed one pipeline record that many people share.
4. What the Evidence Says About Building
No survey splits build against buy for private credit managers, or for financial firms of 10 to 40 people. The nearest evidence comes from larger samples.
MIT's Project NANDA interviewed 52 organizations for The GenAI Divide, a July 2025 report it labels preliminary. Tools bought from an outside partner and customized with its help reached deployment about two thirds of the time, against about one third for tools built internally. The figures are self-reported, and the report says the pattern "does not necessarily prove causation."
Small financial firms that use AI mostly run general-purpose tools. In the Bank of England and Financial Conduct Authority's 2026 survey of UK firms, 73% of small firms (fewer than 100 employees) reported using AI, and 58% of those mainly use general-purpose language models, the technology behind Claude and ChatGPT.
Building needs staff. In Allvue's 2026 GP Outlook Survey of 102 private markets firms (Allvue sells fund software), 36% named "Availability of technical staff to implement AI" as a factor blocking adoption.
The credit managers that have described their choice are far larger than you. Qualitas, an Australian real estate lender with about 140 professionals and an officer hired to lead AI, built its platform in-house; it has published design capacity and a margin target, with no measured saving. Golub Capital bought: in September 2026 it said it would invest $5 million in F2, an AI vendor, and deploy its software, with no result published yet.
5. When Building Is the Right Call
Build when the piece you need is one nobody sells and nobody can configure.
For a small lender, the likeliest such piece is a connector, which lets Claude read a system such as your data room or loan system directly.
Ready-made connectors already exist for Lumonic (portfolio monitoring), Datasite and Ideals (data rooms), and credit data from Moody's, S&P Global and PitchBook; the guide to Claude connectors for private credit firms covers them.
Other systems, among them Allvue, WSO, iLEVEL and Hypercore, are not listed in Anthropic's connector directory. Start with exports by upload, which cost nothing to build. If exporting becomes a daily chore, the next step is a custom connector to the system's API (the access a vendor gives other software), built read-only first so Claude can look up a loan and cannot change one.
Before any build, ask the vendor whether the system has an API and whether your license covers using it.
WorkWise Solutions, which publishes this guide, does the configuring on its AI Operating Partner retainer. Core is $10,000 a month for firms of up to 25 people ($12,500 for 26 to 60), with a three-month minimum, weekly office hours and one build slot for a configured tool such as a deal screen.
A fixed-price first step is the AI Readiness Sprint, $12,500 for firms of up to 20 people, one to two weeks. A connector to a system with no ready-made one is a Custom Build, from $75,000, fixed once scoped.
WorkWise sells no licenses and takes no referral fees. Not a fit if you already employ engineers who can build and maintain your tools.
6. Where to Start
- List the jobs you were planning to build or buy a tool for.
- Mark each one configure, buy or build, using the table's Fits when column.
- Configure the first one this month: buy business seats (never personal accounts), write the criteria and the one-page output, and test on three closed deals.
- Ask for quotes only on jobs that need one shared record, and test each vendor on your own credit agreements.
- Hold any build until the configured version has run for a quarter and you can name the missing piece.
The scorecard for step 4 is in How Should a Private Credit Firm Choose AI Software?
"Strategic partnerships are twice as likely to succeed as internal builds."
MIT Project NANDA, The GenAI Divide: State of AI in Business 2025, preliminary findings from interviews with 52 organizations (July 2025)
- •For a private credit fund of 10 to 40 people, the order is configure, buy, build: configure Claude or ChatGPT on your own templates and screening criteria, buy specialist software for a job that needs one shared record, and build only the piece nothing on the market covers.
- •Configuring a general assistant costs $20 a seat a month billed annually on Claude Team or ChatGPT Business, which is $2,400 a year for 10 people and $9,600 for 40, plus the hours to set it up.
- •A deal screening tool for a small fund has three parts: the fund's criteria written down, a one-page output with the page each figure came from, and an analyst who checks every figure before the screen reaches a partner.
- •MIT's Project NANDA, which interviewed 52 organizations, reported that tools bought from an outside partner and customized reached deployment about two thirds of the time, against about one third for tools built internally.
- •Building is the right call for a piece nobody sells, such as a read-only connector to a loan system that is not listed in Anthropic's connector directory, and no neutral source publishes what such a build costs at a fund of this size.
Frequently Asked Questions
Should we build our own AI deal screening tool or buy one?
For a credit fund of 10 to 40 people, configure one first. Write your screening criteria and a one-page output into Claude or ChatGPT on a business plan, at $20 a seat a month billed annually, and have an analyst check every figure against its page. Buy a screening product only if screens must feed one pipeline record that many people work from, and test it on your own closed deals before you sign. Build only a piece no product covers, such as a read-only connector to your loan system.
How much does a custom AI tool cost for a credit fund?
It depends on which of three things you mean. A general assistant configured on your own criteria costs $20 a seat a month billed annually on Claude Team or ChatGPT Business, which is $4,800 a year for 20 people, plus the hours to set it up. Specialist credit software is sold by quote, and no vendor in the category publishes a price. Software written for your fund alone is priced per project, and no neutral source publishes a typical figure for a small fund, so ask for a fixed price in writing and the yearly cost of maintaining it.
Who can build AI tools for a small private credit fund?
A fund with no engineers has three places to go: the vendor of a system it already uses, a software developer it hires for the project, or an independent AI advisor. Ask each for a fixed price in writing and who maintains the tool afterward. WorkWise Solutions, which publishes this guide, configures tools such as deal screening on its AI Operating Partner retainer, from $10,000 a month for firms of up to 25 people, and builds read-only connectors as a Custom Build from $75,000, fixed once scoped. It sells no licenses and takes no referral fees.
Related Guides & Articles
AI Operating Partner Retainer
Weekly office hours and a build slot for configured tools such as deal screening, built on your own criteria and templates. Core is $10,000 a month for firms of up to 25 people.
AI: Build, Buy, or Partner?
The same decision for any investment firm: when off-the-shelf is enough, when a custom build earns its cost and when to partner.
How Should a Private Credit Firm Choose AI Software?
A scorecard and the questions to ask vendors, tested on your own credit agreements and borrower reports before you sign.
What Are the Best Claude Skills for Private Credit Firms?
Six skills to write first, from a covenant catalog to credit memo drafts, each in your own format with an analyst checking it.
Not sure whether to configure, buy or build?
Tell me the job you were planning to build or buy a tool for, and I'll tell you which of the three routes fits and what it costs. On the AI Operating Partner retainer, Core is $10,000 a month for firms of up to 25 people, with one build slot for a configured tool such as deal screening. A connector to a system with no ready-made one is a Custom Build, from $75,000, fixed once scoped.