Last verified 2026-07-12

Fee & carry verification against LPA terms

Can AI verify management fees and carried interest against the LPA? Yes for the reading — extracting the fee basis, offsets and waterfall terms from a dense agreement and setting up the check. No for the number itself, which you want computed deterministically, and no for signing off a distribution to an LP.

The pain	Management fee and carried-interest terms live in the LPA and the side letters — a fee basis that steps down after the investment period, offsets for transaction and monitoring fees, a distribution waterfall with a preferred return and a catch-up. Checking that what was charged and distributed matches those terms means reading the agreement carefully and re-deriving the numbers, and the errors are quiet: a missed offset or a mis-tiered catch-up is money in the wrong pocket, found late if at all.
What AI does today	Read the fee and waterfall terms out of the LPA and side letters into a structured form a human can check, and flag where a charged fee or a distribution looks inconsistent with those terms. The boundary — sharpest in the whole library: AI reads the agreement into the model's assumptions; a deterministic model computes. A fee or carry figure computed probabilistically is a liability, not a feature.
Proof it's real	Operator assessment — for pure-AI verification. What demonstrably exists is the adjacent service category: Colmore's FAIR (a Preqin company) validates fees, expenses and carry for LP clients by reviewing every LPA and side letter and modelling expected-versus-actual — a technology-plus-people service, which itself tells you where the market thinks the human belongs. A named deployment of AI-only term-extraction feeding fee verification: not found. See To verify. Newest evidence: 2021 (Colmore/Preqin), category ongoing.
What it can't do	It cannot certify the distribution — that is where the human sits: fund accounting owns the computed figures, and a carry calculation goes to LPs and gets audited. A term the model half-read — a catch-up rate, an offset percentage, a hurdle definition — mis-computes every period after and compounds quietly.
The real alternatives	The administrator calculates, the auditor checks — the market default: the admin runs the waterfall, the annual audit samples it. Wins as the baseline everyone already pays for; its known gap is that auditors sample rather than re-derive every term. A fee-validation service (Colmore-class, mostly LP-side) — a specialist re-derivation of expected-versus-actual across a portfolio of funds; wins for allocators with many GP relationships. An internal term-reading step on approved AI — fund-finance teams with an internal LLM gateway can extract the LPA terms themselves into the waterfall model they already own; no new vendor, same verification discipline. The Excel waterfall and a careful reader — the in-house route; entirely sound for a GP checking its own single fund. AI's slice in any route is reading bespoke terms into structured assumptions faster — the verification logic itself stays rules.
What you need in place	The LPA, every side letter, and the charged/distributed history in one place; a deterministic waterfall model whose assumptions the extraction feeds; human verification of every extracted term against the agreement before reliance; and ownership with fund accounting (GP side) or the investment-ops team (LP side). In a group, anything touching investor economics needs central finance and compliance in the loop before any tool does — assume the approval path is longer than the build.
Effort & cost	Weekend script — an LLM reading the LPA's fee and waterfall clauses into a term sheet (basis, offsets, hurdle, catch-up, tiers) you verify against your model; free, and genuinely useful today. Off-the-shelf — validation services priced per fund or per portfolio (the service model); document-extraction tools at five figures a year. Real project — extraction feeding your deterministic model with verified terms; five figures. For a GP with one or two funds, the careful reader plus the auditor is the honest answer — the case is strongest LP-side, across many funds' terms.
What to watch	Verify every extracted term against the agreement before the model relies on it — a misread offset or hurdle produces confident, wrong "matches" every period. And keep the calculation deterministic: if the fee or carry figure varies run to run, that is a defect, not intelligence.

Questions operators ask

Can AI check our carry calculation?

It can read the terms that drive the calculation out of the LPA and side letters, and flag charges that look inconsistent — a real time-saver on dense agreements. The calculation itself belongs in a deterministic model, and the sign-off with fund accounting; nobody credible sells it otherwise.

What is fee validation, and who buys it?

A specialist service — mostly bought by LPs and allocators — that re-derives expected fees, expenses and carry from every LPA and side letter and compares them to what was actually charged. It exists because auditors sample rather than re-derive; AI is making the reading step cheaper, but the category today is technology plus people.

To verify

Named AI deployment in fee/carry verification — no public production example of AI-only term-extraction feeding fee verification found as of 2026-07-12; the graded evidence upgrades when a validation provider or administrator names one.

Changelog

2026-07-12 — published.