Judgment at Scale: How AI Is Changing Due Diligence and Deal Evaluation in Secondaries

A secondaries data room arrives on Tuesday morning. It contains quarterly financial statements for three underlying funds covering twelve quarters each. Capital account histories for every LP class in each fund. Portfolio company lists with GP-reported valuations. Fund-level financial statements. Management fee and carried interest calculations. Audited financials for the two most recent years. The bid is expected by Friday.

This is the standard operational context of secondaries due diligence: substantial data volume, compressed timelines, and a pricing decision that should draw on as much of the available data as possible. In practice, it draws on less than it should — because the time available to process the data is less than the time required to process it manually.

AI is beginning to change this trade-off. Not by making the pricing judgment, but by compressing the time from data room receipt to structured analysis — so more of the available data reaches the judgment before the bid window closes.

ILPA data extraction and GP report processing: from days to hours

Without AI, processing a three-fund secondaries data room takes most of a day for a skilled analyst. Twelve quarters of financial statements for each fund need to be read, the key metrics extracted, and the data organised into a format that allows period-over-period comparison and fund-level attribution. Capital account histories need to be processed to determine unfunded commitments, called capital, distributions, and current NAV for each LP class. Portfolio company lists from three different funds need to be normalised and compared.

With AI document intelligence designed for secondaries data processing, the same processing takes two to three hours. The AI reads each document in the data room, extracts NAV, IRR, MOIC, DPI, TVPI, RVPI, capital account balance, unfunded commitments, and underlying company data for each fund, and organises the results into a structured dataset. The analyst reviews the extracted data for accuracy, challenges any extractions that appear incorrect, and begins the investment analysis from a structured starting point rather than a raw data room.

The time saving is significant in a compressed deal process. More importantly, the quality of the analysis improves because analysts are engaging with structured data rather than extracting it — which is a different and more cognitively demanding skill.

NAV quality assessment: reading the patterns in GP-reported marks

GP-reported NAVs in private equity and private credit funds are Level 3 fair value measurements — management estimates, not observable market prices. The quality of these estimates varies considerably across GPs, and the patterns of variation carry information that is valuable for secondaries pricing.

AI NAV anomaly detection in secondaries portfolio management monitors GP-reported marks over time and across comparable funds to identify statistical patterns: funds whose marks have been unusually stable through periods of sector-wide stress — suggesting marks are not being updated to reflect current conditions; funds whose marks are consistently above or below the median for comparable vintages and strategies; period-over-period mark movements that are inconsistent with the reported operational performance of the underlying portfolio companies. These patterns are not visible to manual analysis at the frequency and across the reference set required to be actionable.

For a secondaries manager pricing an LP stake acquisition, these patterns inform the discount to NAV. A fund with a history of mark stability through volatile periods may carry a GP-reported mark that overstates current fair value. A fund with marks consistently below comparable vintages may carry a mark that understates it. AI surfaces the pattern. The investment professional applies the pricing judgment.

Comparable transaction surfacing for secondaries pricing

The bid price for a secondary LP stake is expressed as a percentage of NAV — a discount or premium that reflects the fund’s quality, vintage, underlying strategy, and GP track record relative to comparable secondary transactions. Pricing it accurately requires a reference set of comparable secondary transactions.

In a high-volume secondaries operation with extensive historical transaction data, AI comparable surfacing identifies historical transactions from the firm’s own deal history that match defined comparability criteria: underlying strategy, vintage year, GP quality tier, portfolio composition stage, entry timing relative to the fund’s expected hold period. The pricing team reviews the AI-surfaced comparables, selects the most relevant, and applies their judgment about what the pattern implies for the current transaction. The reference set is broader, more systematic, and more defensible than what individual deal team memory provides.

The due diligence workflow for GP-led continuation vehicles

GP-led continuation vehicles require a different due diligence workflow from traditional LP stake acquisitions — one that is asset-level rather than fund-level. The data room for a single-asset continuation vehicle contains five or more years of financial statements for the underlying company, management projections, customer contracts, competitive analyses, and operational data that requires analysis equivalent to a direct equity investment.

AI document intelligence for continuation vehicle due diligence extracts the financial metrics from the underlying company’s historical data room — revenue, EBITDA, gross margin, working capital trends, leverage, free cash flow conversion — and organises them into a structured financial model. The investment analyst reviews and builds analysis on the AI-structured foundation rather than spending the first two days of the due diligence process on data extraction. In a secondaries process where the bid window is often shorter for GP-led transactions than for LP stake acquisitions, this compression is operationally significant.

AI does not determine the bid price. It changes what information reaches the pricing decision — how much of the available data in the data room has been processed, how systematic the comparable reference set is, how current the NAV quality assessment is before the bid window closes.