Is AI Worth It for a Small Private Credit Fund?

1. The Short Answer

On the published evidence, AI is worth it for a small private credit fund on document-heavy work: pulling figures out of borrower financials, spreading them into your template and writing the first draft of a memo. That evidence says almost nothing about a firm of 10 to 40 people, because the case studies come from large managers, banks and software vendors.

As of November 2026, no private credit firm of 10 to 40 people has published a measured AI result. A search of trade press, vendor customer pages and smaller lenders by name found none.

So treat the published examples as a reason to test, then measure one workflow at your own fund for one quarter.

2. The Published Examples

Read the last column of the table first: who reports a result tells you how much weight it can carry.

Four of the seven rows carry a before-and-after number. Two of those four describe banks. The other two describe lenders, and both were published by the company that sold the software.

The two lenders that reported for themselves, Qualitas and Ares, gave no measured saving.

What the rows share is the kind of work: reading documents, pulling figures out and drafting from them. PwC's Global Private Credit Survey 2026 gives the same rule: "If a task is document grounded and verifiable, AI is usually ready now."

3. What the Evidence Does Not Show

The published record lacks three things a small fund would want:

• A fund your size. The nearest named lender is Qualitas, with about 140 professionals and a senior officer hired to lead its AI work. The Bank of England and FCA's 2026 survey of UK financial firms found that "73% of small firms reported using AI" (fewer than 100 employees). It counts use and reports no return.
• A checked number. A case study is written by the company selling the software, about a customer it chose. StackAI presents its figures as estimates, and OHA's return came without its dollar inputs or formula.
• A pilot that became a return. MIT Project NANDA's 2025 report, The GenAI Divide, found that "Just 5% of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact." That finding covers custom tools built for one task, in a sample of 52 organizations.

The closest small-firm evidence comes from a survey of all industries. Among AI users in the Federal Reserve Banks' 2026 Report on Employer Firms, a survey of 6,525 firms with 1 to 499 employees, "71% said its use led to increased productivity," while "the vast majority of firms that use AI experienced no change in their labor costs because of AI."

That is the likely shape of a good result at your fund: the same team gets through more work, and payroll stays where it was. PwC describes credit firms using AI as "reviewing more deals with the same headcount."

4. How to Measure It at Your Fund

Decide what you will count before anyone opens the tool. A Federal Reserve Bank of Atlanta working paper from March 2026, built on a survey of nearly 750 corporate executives, found "a productivity paradox, in which perceived productivity gains are larger than measured productivity gains."

Three counts are enough for a small fund:

1. Hours per borrower package, before and after. Time the same analyst spreading and covenant-testing this quarter's packages by hand, then next quarter's with AI. The "after" includes every minute spent checking figures against their pages.
2. Errors caught in checking. Log each figure the analyst corrects, in the AI's spread and in the hand-keyed one. A faster spread with more corrections has saved nothing.
3. Deals screened per month. Count how many new deals get a first read from the same deal team.

No survey publishes hours per workflow for direct lenders, so your own stopwatch is the only baseline.

A large saving on one task can be a small share of the week. OHA's figure is up to 75% off review time on some workflows. Across the whole week, a Federal Reserve Bank of St. Louis survey of workers who use generative AI found "an average time savings of 5.4% of work hours," about 2.2 hours in a 40-hour week.

At quarter end, price the hours given back at what that person's time costs your firm, and set the total against what you spent. The Credit Memo ROI Calculator does that arithmetic for memo drafting, and the checks for a spreading test are in Can ChatGPT or Claude Spread Borrower Financials?

5. What It Costs to Find Out

Licenses are the small line. On Claude Team and ChatGPT Business, the plans each vendor sells to companies, a seat (one paid login per person) costs $20 a month billed annually, or $25 billed monthly. Billed annually, that is $2,400 a year for 10 people, $4,800 for 20 and $9,600 for 40.

A one-quarter test needs fewer seats: five, billed monthly for three months, come to 5 x $25 x 3 = $375. At $20 a month, a seat that gives its user back one hour a month has cost $20 for that hour.

Specialist monitoring software needs a quote, because no vendor in that category publishes a price.

The larger cost is time: an owner's hours each week, the analysts' hours learning and checking, and any outside help. Put all three in the "spent" column.

6. Where to Start

1. Pick one workflow the team repeats every quarter, such as spreading the borrower package.
2. Time it by hand on this quarter's packages and write the hours down.
3. Buy business-plan seats for the test team, never personal accounts, and write a one-page rule on borrower files using Is It Safe to Put Borrower Financials Into ChatGPT or Claude?
4. Run next quarter's packages with AI, an analyst checking every figure, and log the hours and the corrections.
5. Put the last day of the quarter in the calendar to keep, change or stop.

Your number will be worth more to you than any row in the table above, because it will be the only one measured at a fund your size.