AI Can't Fix a Broken Book of Record — It Just Makes You Wrong Faster

The promise of AI in net asset value (NAV) oversight is genuine for private markets firms. Faster closes at whatever cadence a firm runs — quarterly for a traditional closed-end fund, daily for an interval fund or business development company (BDC). Programmatic valuation that takes hours instead of days. Exception-based governance that lets teams review what broke instead of re-checking every capital call, distribution, and position.

The industry is investing accordingly. According to KPMG's Quarterly AI Pulse Survey, asset management and private equity leaders plan to invest an average of $101 million in AI over the next 12 months, with 68% already piloting AI agents and 24% deploying them in production.ⁱ The spending backs it up: McKinsey reports that the asset management industry's total cost base rose to $167 billion in 2024, with technology spending growing 9% — the largest increase of any category.ⁱⁱ

But AI is an accelerant, not a foundation. It doesn't discriminate between correct and broken data. It processes whatever it's given, faster. And in a NAV process, what it's given — the book of record — is either complete, reconciled, and defensible, or it isn't.

AI doesn't change which one it is. It just changes how quickly the consequences arrive.

What happens when AI meets a broken book of record

An agent that mis-maps a data field, silently drops an exception, or hallucinates a source doesn't just produce a bad chart. In a NAV process, it produces a valuation number that a board, an auditor, or an LP may rely on. Errors propagate before anyone can catch them.

The problem isn't theoretical. In Deloitte's 2024 Banking & Capital Markets Data and Analytics Market Survey, 81% of respondents cited data quality as a top challenge, and more than 90% reported that the data they need is often unavailable or takes too long to retrieve.ⁱⁱⁱ For private markets firms, the challenge is compounded by structural complexity: Most firms coordinate NAV inputs across multiple fund administrators, each communicating in different formats and on different timelines. When firms layer AI on top of data that's incomplete, inconsistent, or siloed across administrators, the model inherits every gap and every inconsistency.

Consider what happens at a firm running daily NAV for an interval fund with an incomplete book of record. A capital call is recorded against the wrong fund vehicle. A private credit position is missing from the administrator's feed. A real estate valuation is stale. Without AI, a human reviewer catches the anomaly during the close, corrects it, and moves on. With AI, the model processes the broken data, produces a NAV that looks plausible, and pushes it downstream — to LP capital account statements, to regulatory filings, to subscription and redemption pricing — before anyone has time to notice the gap. The error is amplified and distributed.

Why “the AI did it” isn't a defense

Under the SEC's Rule 2a-5, which took effect in 2021 with a compliance date of September 2022, fund boards or their valuation designees must make good faith determinations of fair value, backed by documented methodologies, testing, and escalation procedures.^iv The rule applies to any registered fund calculating NAV, at any frequency. The disclosure and audit obligations — and the cost of making it right — land on the firm regardless of who, or what, generated the number.

The model doesn't get to hide behind “the AI did it.” If the resulting NAV is wrong, the firm still must restate, notify LPs, and explain the break to the board. If a hallucinated source inflates a private credit valuation and an investor transacts on it, the liability doesn't shift to the model provider. The regulatory framework doesn't distinguish between errors made by a spreadsheet, a junior analyst, or a language model. The obligation is the same: The number has to be right, and the firm has to be able to prove how it got there.

The sequence that works

The firms getting AI right in NAV oversight are the ones that sequenced it correctly: book of record first, governance second, acceleration third. Each step depends on the one before it. You can't govern what you can't trust, and you can't accelerate what you can't govern.

Firms that try to reverse the sequence — starting with AI and working backward to fix the data — end up generating clean-up projects alongside their AI initiatives rather than results. McKinsey reports that, “Many asset managers continue to operate on aging infrastructure that is expensive to maintain, and the absence of well-integrated systems has made supporting core operations costlier and stymied innovation with newer technologies like generative AI.” The firms that skipped the foundational work on data quality and governance are the ones now playing catch-up, spending on AI without the data foundation to make it productive.

The problem is growing more acute as crossover strategies blur the lines between asset classes. Firms launching private credit strategies or allocating a percentage of their equities to pre-IPO investments are taking on private markets positions that don't flow through the same data feeds. There's no Bloomberg ticker for a private credit position or a pre-IPO holding. These positions sit outside the firm's existing book of record, invisible to the systems built for liquid strategies. Layering AI on top of that gap teaches the model to operate on incomplete data and present the result as complete.

The correct sequence is nonnegotiable because each layer serves a specific purpose. The book of record provides the complete, reconciled, defensible data: every capital call, distribution, position, and valuation across private equity, private credit, real estate, and infrastructure, reconciled to custodian and fund administrator.

Governance provides the exception-based oversight that confirms the data is trustworthy before anything runs on top of it. AI provides the acceleration — faster closes, programmatic valuation, real-time LP reporting — but only because the layers beneath it are solid. Remove any layer, and the whole stack collapses.

What AI unlocks when the foundation is right

Get the sequence right, and AI becomes a genuine multiplier. A complete, reconciled book of record lets a firm move faster and makes AI-assisted acceleration possible without the AI itself becoming the risk. Agents can summarize valuation drivers for a board memo, flag pricing or market anomalies before they become audit findings, and answer an LP's question about their capital account balance in real time because the data is trustworthy.

The firms achieving this did the same work every firm has to do — complete the book of record, reconcile it across every dimension that matters, and build the governance to prove it's right — they just did it before layering AI on top. AI made the speed they already earned more valuable, whether they're running daily NAV for retail vehicles or quarterly closes for institutional funds.

AI can't fix a broken book of record. It just makes you wrong faster. But when the book of record is complete, reconciled, and defensible, AI makes you right faster — and that's the outcome worth investing in. The firms that recognize the difference, and sequence accordingly, are the ones that will turn AI spending into operational results. The ones that don't will keep compounding the same errors at a faster clip, wondering why the technology that was supposed to fix everything is making it worse.