How to Automate LP Reports From a Live Shadow Book
Automating LP reports requires a data source the reporting layer can trust: a live, continuously reconciled shadow ledger with a deterministic NAV. Once that exists, a report becomes a period query against numbers already computed and audited, with AI drafting only the narrative sections around figures it never calculates — and a manager approving the draft before anything reaches an investor.
[[lp-reporting|LP reporting]] is the periodic statement a fund sends its limited partners: NAV, returns, portfolio composition, and commentary. "Automating" it does not mean an AI model invents the numbers — it means the report-generation step reads from a book of record that already has correct, tied-out figures, so the only work left is assembling and narrating them. That book of record is the same one covered in shadow NAV and continuous reconciliation.
Without a reliable shadow ledger, automation just moves the manual-reconciliation problem downstream: you would be automating the formatting of numbers nobody has checked.
Why does manual LP reporting take so long?
The bottleneck in most crypto fund LP reporting is not writing the report — it is assembling correct inputs from a book that lives across a dozen exchange accounts, several wallets, and a spreadsheet trying to reconcile all of it. An analyst pulls balances, prices them by hand, checks fee accruals, cross-references the prior month, and only then starts drafting commentary. Most of that time is spent proving the numbers are right, not producing the document.
A live shadow book removes that step by construction: positions are reconstructed from an append-only, double-entry shadow ledger, not from live tables, so a NAV run is reproducible on demand rather than assembled fresh each month.
What does the reporting pipeline actually compute?
A period resolves to a date range — typically calendar month, aligned to how the administrator closes. From there the pipeline assembles the full context in one pass: portfolio breakdown, trailing NAV history, benchmark returns, volatility and drawdown statistics, concentration and liquidity, stablecoin exposure, fee accruals, and net-of-fee returns per investor. Every one of those figures is produced by a deterministic engine before any AI model is involved.
Net-of-fee returns matter specifically because they depend on the same fee logic that runs the ledger: pre-fee NAV, management fee proration, and performance fee crystallisation against a high-water mark. A report that gets fee accrual wrong will misstate every investor's net return, which is why the reporting layer reads fee figures rather than recomputing them independently.
What is the AI actually allowed to do?
The AI drafts prose — performance summary, portfolio composition commentary, manager commentary, market context, and outlook — under a fixed set of section headings. It is given the pre-computed figures as exact strings in its prompt and instructed to narrate around them. It is never given a calculation to perform and never asked to produce a number that isn't already in its context.
This is a narrow, deliberate division of labour: deterministic engines compute, the model narrates. It is the same principle behind a LP report grader — you can score whether a report reads clearly and covers the right sections without that scoring process ever touching the underlying NAV math.
How does a report get from draft to an investor's inbox?
- Resolve the period and assemble context
The reporting period is resolved (usually a calendar month), and every input figure — NAV, returns, risk metrics, fee accruals, concentration — is computed once and stored alongside the report.
- Draft the commentary
An AI pass writes the narrative sections against that fixed context, with no dollar figure or percentage generated outside what was already computed.
- Manager review and edit
The draft lands in a review queue. A manager can edit any section directly before approving — this is the checker step, not a rubber stamp.
- Approve
Approval is a distinct, rate-limited action from drafting, so a report can never move from draft straight to an investor without a human decision in between.
- Distribute
On approval, the report renders to PDF once, publishes to the investor portal, and emails each contact with the PDF attached — with delivery, open, and click tracked per recipient so a manager can see who has actually read it.
Automating the reporting layer does not replace your fund administrator, your auditor, or your fund's legal counsel — it replaces the manual assembly and drafting work that sits on top of a book those parties already touch. A shadow ledger reconciles against the administrator's official NAV; it is not itself the official NAV of record.
Key takeaways
LP report automation only works once the underlying book is reliable — automating a spreadsheet reconciliation just automates the wrong thing faster.
Every figure in the report — NAV, net returns, risk metrics, fee accruals — should come from a deterministic engine computed once, not from a model asked to calculate on the fly.
Net-of-fee returns depend on correct fee accrual logic (pre-fee NAV, proration, crystallisation against a high-water mark), so the reporting layer must read fee figures rather than re-derive them.
AI has a narrow, defensible role: drafting narrative prose around numbers it is handed, never producing a number of its own.
A draft-then-approve workflow, with distribution as a separate step from drafting, keeps a human decision between an AI-drafted paragraph and an investor's inbox.
Automated reporting complements, and never replaces, your fund administrator, auditor, or counsel.
Questions, answered
Can AI generate an LP report without human review?
It should not, and a properly designed pipeline does not allow it. AI-drafted commentary should sit in a review queue where a manager can edit or reject it before an explicit approval step, which is kept separate from the distribution step that actually sends the report to investors.
Does automating LP reports mean the AI calculates NAV or returns?
No. NAV, returns, fee accruals, and every other figure should be computed by deterministic accounting logic and handed to the AI as fixed text in its prompt. The AI's job is to write commentary around numbers it receives, not to calculate or re-derive them.
What data does an automated LP report need to be accurate?
It needs a reliable book of positions and transactions — ideally from a continuously reconciled shadow ledger — plus correct fee accrual logic, since net-of-fee returns are only as accurate as the management and performance fee calculations feeding them.
How is an automated LP report different from a template that auto-fills numbers?
A template still requires someone to source and verify the inputs each period. Automation from a live shadow book removes that step because the inputs are already computed and reconciled continuously, so report generation becomes a query against existing figures rather than a fresh data-gathering exercise.
Can automated LP reporting replace a fund administrator?
No. A shadow ledger and its reporting layer reconcile against the administrator's official NAV; they are not themselves the fund's legal book of record. They complement the administrator by producing the same numbers faster and cross-checking them, not by replacing the administrator's role.
- ILPA Reporting Template — Institutional Limited Partners Association (ILPA)