How LPs and Fund-of-Funds Use AI on Fund Documents

LPs and fund-of-funds use AI to read the documents their GPs send (quarterly reports, capital account statements, capital call and distribution notices, LPAs, side letters, DDQs and K-1s), pull the figures and terms into one place, and answer questions across the whole program: look-through exposure, fees, key-person terms, cash flows and manager performance. Kepler, which we build, does that work end to end: every GP document read in full, every fund, vehicle and portfolio company resolved into one model of your program, and every number computed in code and cited to its statement. Any AI tool can summarize a single report; the work LPs need is reading every document from every GP, in every format, and producing numbers that reconcile.

The sections below note where Kepler fits each job, and the last one shows how it runs.

The volume is large and the formats vary. A fund-of-funds manager may receive information from dozens or hundreds of underlying funds, each reporting in its own layout. Reports also arrive late by design: ILPA’s suggested timeline gives direct funds 60 days after quarter-end and fund-of-funds 120 days, so look-through numbers are months old by the time they’re assembled. LPs expect AI to matter. In Coller EQT’s Summer 2026 Barometer, a survey of 108 LPs, 67% said they expect AI to widen the gap between the best and worst funds, and the report notes that LPs are adopting AI in their own organizations.

This guide covers the documents, the work LPs do with them, why generic AI struggles, and what a tool needs to get right. If you’re choosing a product, our list of AI tools for LP and fund-of-funds due diligence compares the options.

The documents LPs work with

Document	What’s in it	What LPs pull out
LPA	The fund’s governing terms	Fees, carry and waterfall, key-person, investment period, removal rights
Side letter	Terms negotiated by one LP	MFN rights, fee terms, reporting rights, excusal and co-investment rights
Quarterly report	Fund and portfolio update, often with unaudited financials	NAV, holdings and valuations, portfolio company metrics, commentary
Capital account statement	The LP’s own position	Commitment, contributions, distributions, fees, NAV, unfunded commitment
Capital call notice	A request for cash	Amount, due date, split between investments, fees and expenses
Distribution notice	Cash returned	Amount, return of capital vs gain, recallable portion, withholding
DDQ	A GP’s answers to a standard questionnaire	Team, strategy, track record, operations, ESG, data security
Schedule K-1	The LP’s share of income for US tax	Income, gains, deductions and credits by category
Audited financials and AGM materials	The audited year and the GP’s annual view	Audited NAV, fees and expenses, strategy and team updates

Industry standards help at the edges. ILPA released an updated Reporting Template and a new Performance Template in January 2025, the first update since 2016, and its FAQ describes a gross-to-net reconciliation of management fees. ILPA recommends the Reporting Template for funds still investing in Q1 2026 or launched from January 2026, and the Performance Template, for funds launched from January 2026, first delivered after Q1 2027. Its DDQ was last updated in 2021. Funds outside the templates’ scope, and managers that don’t adopt them, report in their own formats.

How LPs use AI on fund documents: the main jobs

Look-through exposure

An LP holds fund interests, but its risk sits in the companies underneath. Look-through reporting answers “how much of our money is in this company, sector or country?” by taking each fund’s positions, applying the LP’s share of that fund, and adding across funds, co-investments and separate accounts. By hand, that means collecting the latest schedule of investments from every quarterly report, matching company names each GP spells differently, adjusting for ownership, and noting each fund’s as-of date. On Kepler, the answer is one number with a row per vehicle, each row citing the schedule of investments it came from and showing its as-of date. Our guide to look-through exposure works through the calculation with an example.

Cash flows, NAV and reconciliation

Each capital call and distribution should tie to the next capital account statement: contributions in, distributions out, fees and expenses charged, NAV rolled forward. ILPA’s Principles 3.0 ask GPs to show the management fee calculation, offsets included, in each call and to give LPs at least 10 business days to respond. Operations teams reconcile by hand across dozens of GPs, and a break (a call booked to the wrong vehicle, a distribution coded as income when it was return of capital) can sit unnoticed for quarters. On Kepler, each notice is matched to the vehicle it belongs to as it arrives, and a break shows up the day the statement disagrees with it. Our guide to processing capital calls with AI covers the workflow from notice to reconciliation.

Fees, expenses and offsets

Management fees, offsets, partnership expenses and carried interest are set by the LPA and changed by side letters. ILPA’s principles recommend that portfolio company fees be offset 100% against the management fee. Checking what was charged against what was agreed means reading the governing terms, the fee pages in each report and the notices, then recomputing. On Kepler, that recomputation runs in code on the fee terms as your side letter amended them, with every input cited.

Terms: key-person, MFN and side letters

Investment and legal teams ask term questions across the whole portfolio: which funds have a key-person clause naming a partner who just left, which side letters carry MFN rights, where the LP holds excusal or co-investment rights. ILPA recommends that a key-person event automatically suspend the investment period, becoming permanent within 180 days unless a supermajority of LPs votes to reinstate it. The answer to a term question has to be a complete list, with the clause text from each agreement. Kepler returns that list from every LPA and side letter it has already read, with the clause text and page for each.

Manager diligence and monitoring

Before a commitment, an LP reads the DDQ, the track record, the data room and years of quarterly letters. After it, the LP watches for drift in strategy, team, valuations and pace of deployment. AI is useful for comparing what a manager said in 2022 with what it says now, and for checking that track-record figures in a pitch match the reported numbers. On Kepler, every claim in a manager comparison links to the letter or DDQ answer it came from.

Tax documents

For a calendar-year partnership, the return and each partner’s K-1 are due March 15, and an automatic six-month extension pushes that to mid-September. K-1s arrive in volume and late. AI can extract K-1 line items and compare them with the GP’s estimates, while most LPs keep final tax work inside their tax provider’s process.

Why generic AI struggles with fund documents

• Every GP formats differently. There is no standard layout for a quarterly report or a capital account statement in practice, even with ILPA templates in circulation, so a tool built on fixed templates breaks on the next manager.
• Many documents are scanned. Notices and statements are often PDFs of printed pages, with tables that lose their structure in naive text extraction.
• The numbers must reconcile. A summary can be roughly right. A capital account roll-forward can’t: one transposed digit breaks the tie-out.
• The answers span documents. Look-through, fee and key-person questions need the LPA, the side letters, the notices and the reports read together, across every fund. A chat tool that answers from one uploaded file or a few retrieved passages can’t produce them.
• Names don’t match. The same portfolio company appears under a legal name, a brand name and an abbreviation across GPs. Without entity resolution, exposure is undercounted. Resolving those names is the job of an ontology.
• Time matters. A NAV is as of a date, a restated quarter replaces an earlier one, and a side letter can amend the LPA years later. The tool has to know which version was true when.
• Access must be controlled. Fund documents are confidential under the LPA, and different teams may be cleared for different funds.

What to look for in a tool

1. Reads every page of every document, including scans, and keeps each figure beside its passage.
2. Resolves funds, vehicles, GPs and portfolio companies into one record each, across spellings.
3. Computes exposure, performance and reconciliations in code, with formulas and inputs shown.
4. Returns complete lists for term questions, with clause text.
5. Shows both values, with sources, when documents disagree.
6. Runs on its own as documents arrive, with alerts that quote the source.
7. Mirrors your permissions and logs every query.

Together, these describe what we call enterprise intelligence: one model of the firm that AI can reason over, prove and act on. We built Kepler to meet all seven on an allocator’s documents, and the next section shows how.

How Kepler does this

We built Kepler for this document problem, and the work starts with a model of your program.

One model of your program. Our engineers build your firm’s ontology from your own sources: GP portals, the inbox where notices land, your commitment tracker, your accounting system and years of archived reports. Every fund, parallel vehicle, feeder, GP, key person, commitment and portfolio company resolves to one record. The same company held through three funds under three spellings becomes one entity, and your definitions of net IRR, unfunded commitment and exposure are written once and applied everywhere.

Every page read on arrival. Extraction models we train read each document in full when it lands: quarterly reports, capital account statements, notices, LPAs, side letters, DDQs and K-1s, scanned or native. Each figure keeps its value, scale, unit, period and passage. Because the reading happens at ingest, a question across 300 LPAs is a lookup over facts already filed, and “which agreements have a key-person clause naming this partner” comes back complete.

Numbers that reconcile. Code does the arithmetic. Look-through exposure comes back as one number with a row per vehicle, and each row opens to the statement and line that fed it. Contributions and distributions roll forward against each capital account statement, and when a GP’s reported NAV doesn’t tie to the activity in its own notices, Kepler shows both figures and flags the break. Run the same question next week and the figure and its sources match, unless a new document changed them, in which case Kepler shows which one.

Work that runs on its own. Agents read each notice and report as it arrives, with the same permissions as the person who set them up, and nothing written in a GP’s document can redirect them. They check calls against unfunded commitments and alert you with the exact quote when a key-person event, a fee change or a markdown shows up. The quarterly board pack drafts in your format in Word, Excel or PowerPoint with every figure cited, and the same answers reach Claude, ChatGPT, Slack and your own systems through the Kepler API.

Your playbooks, the right models. Your manager scorecard, re-up memo and board pack templates are saved once and applied every quarter, and each correction your team makes improves the firm’s own models privately. Compact models we train read the GP documents, frontier models take on the reasoning and the board-pack prose, and we move to better ones as they ship, so your team gets each improvement without changing how it works.

Rules built in. Permissions mirror your firm’s, down to the fund and the team, every query and calculation is logged, and nothing Kepler reads is ever treated as an instruction. Our engineers built this layer before at Palantir, and they sit with your operations and investment teams to connect sources and stand up the ontology in the first week.

Questions allocators ask Kepler

• “Which funds name this partner in a key-person clause, and does a trigger suspend the investment period in each?”
• “Roll forward Fund IV’s capital account from this year’s notices and show where it breaks from the GP’s statement.”
• “How has this manager described its value-creation strategy across five years of quarterly letters?”

To see Kepler on fund documents like yours, book a demo; to try it on public companies, go to app.kepler.ai. For tools beyond the LP desk, see the best AI tools for asset managers and for family offices, and how the Kepler platform fits together.

Frequently asked questions

How do LPs use AI on fund documents?

To collect and read what GPs send (quarterly reports, capital account statements, capital call and distribution notices, LPAs, side letters, DDQs and K-1s), extract the figures and terms, reconcile cash flows and NAV, and answer questions across the whole program, such as look-through exposure to one company, every fund with a key-person clause, or how a manager's story has changed over five years of letters. Kepler, which we build, does this on one model of the whole program, with every document read in full and every number computed in code and cited to its statement.

What documents do LPs receive from GPs?

The governing documents (the LPA, subscription documents and any side letter), quarterly reports with portfolio updates and often unaudited financials, capital account statements showing the LP's own position, capital call and distribution notices, annual audited financial statements and annual meeting materials, and, for US taxable investors, a Schedule K-1 each year. Before committing, LPs also review a DDQ and the fund's data room.

What is look-through exposure for an LP?

Look-through exposure is the LP's indirect holding in the companies, sectors and countries inside its funds. It's calculated by taking each fund's positions from its latest schedule of investments, applying the LP's ownership share of that fund, and adding across every fund, co-investment and separate account. The hard parts are matching the same company across GPs' different spellings and working with data that is a quarter or more old. Kepler resolves those spellings to one company, does the arithmetic in code and shows each fund's as-of date beside its row.

Can ChatGPT read capital account statements and LPAs?

It can read and summarize one you upload, and connectors from providers such as MSCI and Chronograph bring LP data into ChatGPT and Claude. On its own, a frontier model reads a handful of files or retrieved passages at a time and writes figures itself, so cross-fund answers and reconciliations are hard to trust. Kepler runs inside both, with every fund document already read and every number computed and cited.

What are the ILPA reporting and performance templates?

They are standard formats the Institutional Limited Partners Association publishes so GPs report fees, expenses and performance consistently. ILPA released an updated Reporting Template and a new Performance Template in January 2025, the first update since 2016. ILPA recommends the Reporting Template for funds in their investment period during Q1 2026 or launched from January 2026, and the Performance Template, for funds launched from January 2026, first delivered after Q1 2027.