GenAI, Unstructured Data, And The Next Divide In Private Markets

Private markets run on rich but largely unstructured information. The documents that matter most, including LPAs, board materials, loan agreements, ESG reports, operating KPIs, typically live in PDFs, slide decks and email trails rather than clean databases. That is exactly the type of information large language models are designed to work with.

State Street’s 2025 Private Markets Study suggests institutions are starting to treat this as a strategic advantage, not just an operational challenge. A large majority now see GenAI and LLMs as important for turning unstructured content into analyzable data, and many are budgeting accordingly. The interesting gap is between firms that are already implementing these tools in core workflows and those that are still at the planning stage, and this gap is likely to become a real competitive divide over the next few years.

The step change in perceived value

The shift in sentiment in just one year is pretty stark. In the latest survey, 83% of respondents say they are planning use cases for GenAI and large language models to turn unstructured information in their private-markets operations into analyzable data. In the previous survey, only 58% even said they saw value in this kind of application.

This isn’t the vague “AI will change everything” story. It’s a very specific recognition that the bottleneck is all the stuff sitting in file shares and VDRs that nobody can read deeply or consistently.

Budgets are following the rhetoric

Crucially, the money is moving as well. Across LPs and GPs, roughly 68–70% expect to increase technology spend on data management for private markets over the next one to two years, and only about 1% expect to cut it.

So “data plumbing” has effectively become non-discretionary. The open question is not whether to spend, but where that spend lands in internal builds, specialist vendors, or some hybrid.

Adoption reality: lots of planning, limited deployment

Under the hood, the implementation picture is much more modest than the rhetoric. About a third of respondents (33%) say they are using GenAI or LLMs to generate consistent, analyzable data from unstructured information tied to their private-markets investments. Yet only 8% have actually implemented those solutions in production; another 41% are still at the pre-investment planning stage.

LPs and GPs aren’t that far apart. Actual adoption rates are 9% for GPs and 8% for LPs, with GPs slightly more likely to be investing in concrete use cases (36% vs 31%).

So you effectively have three tiers:

• a small group running GenAI against real workflows,
• a larger group piloting and scoping, and
• a laggard fringe still in “interesting, but later” mode.

Scale effects: very large and very small move first

The survey is blunt about the role of size. Assets under management strongly influence how far along institutions are. Among organisations with $100 billion+ in AUM, 16% are already using AI for unstructured data, roughly double the overall respondent base. In the $10–50 billion bracket, 11% are using AI, leaning on their nimbleness. In the US$50–100 billion cohort, only 3% are currently using AI in this way, although 40% of that group is actively investing in AI use cases (versus 29% of the smaller cohort).

So the pattern looks like this:

• very large players can throw money and teams at the problem,
• smaller institutions can move quickly on focused use cases, and
• mid-sized firms risk sitting in the worst of both worlds; too complex to ignore the problem, too slow to gain early-mover advantages.

Where GenAI is actually being pointed

The use cases the survey highlights are, reassuringly, quite practical. At the portfolio level, respondents see GenAI and LLMs as most useful for investment performance reports, NAV statements and aggregate capital-account data, capital-call and distribution documentation, subscriptions and redemptions data, aggregate private-credit spread information, and portfolio-level ESG metrics.

At the holdings level, they point to individual asset performance reports, company financial statements, operating reports on real-asset holdings, ESG data at the asset level, credit and loan agreements, and purchase/sale documentation.

In both views, performance analysis comes out as the single most important area where respondents expect AI to be “most useful”. In plain language: institutions want these systems to read all the documents nobody has time to read properly, reconcile the numbers, and surface patterns or risks.

Data and GenAI as preconditions for democratization

The survey also links this directly to the broader “democratization” narrative. Around 34% of respondents say technology developments that enable more frequent, timely and high-quality data are essential to make private markets accessible to a wider range of individual investors.

That matters more than it sounds. In semi-liquid and retail-style vehicles, if managers cannot produce reliable, near-real-time positions and valuations, liquidity promises become fragile. And if they cannot aggregate exposures across hundreds of underlying holdings quickly, risk management under stress becomes guesswork. The report puts it simply: portfolio liquidity begins with data liquidity.

Where the real edge is likely to sit

There’s a temptation to treat “we’re investing in GenAI for private markets” as a marketing checkbox. The State Street work suggests that’s exactly how you end up stuck in the middle.

The actual edge is more prosaic:

• Owning the data model. Not every institution needs its own foundation model, but if you don’t control how documents, entities and relationships are structured, you’re essentially renting insight from vendors and risking fragmentation.
• Prioritising high-value workflows. Performance reporting, capital-account reconciliations and credit documents are obvious starting points. Pushing out an AI chatbot for LP queries while core portfolio data remains inconsistent is cosmetic, not transformative.
• Building proper validation. GenAI systems can mis-extract or hallucinate. Without clear checks around legal clauses, covenants and instrument terms, automation just scales errors instead of reducing them.

The line in the report that really matters is that today’s leaders are moving from hypothetical to real implementation, and that this will give them an advantage “as portfolio liquidity begins with data liquidity.”

The divide in private markets data over the next few years won’t simply be between firms that “use AI” and those that don’t. It will be between firms that can genuinely see and slice their private-markets exposure in close to real time and firms still managing multi-billion books via slide decks and quarterly PDFs.

Research Analyst

Simran Mehla

Frequently Asked Questions

Q: Why does GenAI matter specifically for private markets?

A: Because most key information sits in documents, not databases, and GenAI can turn that unstructured content into searchable, analyzable data at scale.

Q: Are institutions actually using GenAI yet, or just talking about it?

A: A minority have real deployments, a larger group is actively investing or piloting, and the rest are still at the planning stage — so implementation is early but starting to spread.

Q: Where are the most common use cases today?

A: Performance and portfolio reporting, document and data extraction (LPAs, loan docs, board packs), and ESG and risk reporting are the main focus areas so far.

Q: Does firm size change how GenAI is adopted?

A: Very large firms are furthest ahead, smaller firms are moving on focused use cases, and mid-sized organisations risk being stuck in between: aware of the need, slower on execution.

Q: How could GenAI affect “democratization” of private markets?

A: Semi-liquid and retail-style vehicles depend on timely, accurate data; GenAI and better data infrastructure make it more feasible to support frequent dealing, risk monitoring, and transparent reporting at scale.

Sources

State Street, 2025 Private Markets Study – Driving success in volatile environments.