AI Private Credit Research: Income Is Not Cash Collected

A private credit report can show a larger loan portfolio, rising investment income and a covered dividend while leaving the central cash question unanswered. Ares Capital's 2025 accounts illustrate the problem: total investment income rose, net investment income fell, and some recognized income was paid in kind. Those statements describe different parts of the economics. An AI research note needs to keep them separate.

For an alternative asset manager or credit research team, the useful output is a reconciliation from reported income to the next question about collection, credit quality or distributions. In AllMind, we would preserve that reconciliation alongside the source passages and prior review, so a new filing updates the analysis without erasing its definitions.

Start with the result available to each share

The following figures come from Ares Capital's full-year 2025 results. This is a historical worked example, not a statement about its latest quarter.

The annual per-share net investment income exceeded declared dividends by $0.10 in 2025, compared with $0.38 in 2024. Coverage on this measure narrowed from about 1.20 times to 1.05 times. Share growth helps explain why the per-share decline was much larger than the decline in total net investment income. These are accounting comparisons, not a forecast of the next distribution.

That distinction changes the opening of the review. “Income grew” is incomplete. “The portfolio produced more gross investment income, but less net income and less income per share” identifies the follow-up work.

Find all the PIK rows before calculating a ratio

Payment-in-kind income increases the claim on the borrower instead of requiring the same amount of current cash payment. Ares reports PIK in several ownership categories and separates interest from dividends. Its 2025 Form 10-K, consolidated statement of operations, supplies these inputs in millions of dollars:

The sum rose 5.2%. Dividing by total investment income gives approximately 15.5% for 2024 and 16.0% for 2025. The calculation must include the dividend rows; extracting only the first PIK-interest match would produce a materially smaller numerator.

The result is not a cash collection ratio. Removing $487 million from investment income does not reconstruct cash received, cash available for distribution or a default forecast. Accrual timing, fees, investment transactions and expenses need their own reconciliation. Nor does PIK alone prove borrower distress: the contract and subsequent payment history matter.

Turn the accounting bridge into a credit question

There are now three separate questions for the next review. What reduced net income after gross income increased? Which investments contributed the additional non-cash income? Did the borrower collectability assessment or contractual payment terms change?

Those questions should produce different evidence. The first needs the expense bridge. The second needs investment-level identification and ownership scope. The third needs the loan terms, amendments and subsequent filings. A report that attaches the same general portfolio paragraph to all three has not finished the analysis.

For a clearly illustrative borrower, suppose a $100 million loan earns an 8% cash coupon and 4% PIK for one year, with no principal repayment. It recognizes $12 million of interest, receives $8 million of coupon cash and ends with a $104 million principal claim before other adjustments. Calling its 12% yield “cash yield” overstates the current cash receipt by half. Calling the extra $4 million a loss is equally unjustified. The reviewer needs to assess the enlarged claim and the borrower's ability to repay it.

Keep that loan example distinct from the Ares totals. It explains the mechanism; it does not describe any particular investment in Ares's portfolio.

Where AllMind fits a recurring credit review

We recommend AllMind when a team needs the same income-quality questions answered across several managers or credit vehicles, with the supporting documents still available for review. Data Rooms can keep each vehicle's financial statements and permitted manager materials together. Grids can apply consistent questions across them, including whether a figure is interest or dividend income, cash or PIK, fund-level or borrower-level, and quarterly or annual.

The finished artifact should be a comparison that retains the calculation inputs and identifies the missing collection evidence. It should not assign an unexplained “healthy” score. For the example above, a concise conclusion would say that annual accounting coverage narrowed and disclosed PIK income increased, then specify the borrower-level review required before making a cash-quality judgment.

A loan-servicing system remains the place for authoritative payment status and transaction records. AllMind's role is to connect those permitted records with filings and research, not to replace a servicing ledger or make an autonomous credit decision. The broader manager due-diligence guide explains how to preserve the scope of a manager's answer.

What to carry into the next period

Retain the denominator and all the ownership categories. A quarterly PIK amount divided by annual investment income is not comparable to this calculation. A portfolio-wide PIK percentage cannot be assigned to an individual borrower. An annual dividend coverage figure does not establish when cash arrived.

When the next statement arrives, the useful change log is short: revised income mix, net-income bridge, share-count effect, and the status of each open credit question. That gives the committee something it can act on without turning an accounting headline into a claim about cash it has not verified.

Method

Calculations use the public 2025 results and consolidated statement of operations, checked November 10, 2026. Growth rates and coverage are our arithmetic, rounded. This analysis did not inspect private loan files or execute an AllMind product test. The borrower example is illustrative. Download the calculation inputs.