Private market firms run on documents. Stock purchase agreements, limited partnership agreements, side letters, and capital call notices are the legal instruments that define a firm's relationships with its investors, its exposure to portfolio companies, and the terms under which capital moves. The data inside those documents is, in many cases, the ground truth.
And yet for most firms, that data is inaccessible. It lives in PDFs, readable by humans, invisible to systems. Getting answers out of it requires someone to find the right document, open it, read through it, and manually transfer what they found somewhere else. When the same question comes up next quarter, the process starts over.
There are two inadequate alternatives to this. The first is to build point-in-time tools that extract specific fields for specific use cases. Pull the purchase price from an SPA when running a valuation, or pull the carry terms from an LPA when calculating a distribution. The second is to hand documents to AI on a one-off basis and ask it to find what you need. Both approaches are better than pure manual review, but neither actually solves the problem. One doesn't scale consistently across the platform. The other doesn't persist.
We think there's a better way: exhaustive document extraction connected to Carta’s ERP.
The problem with per-use-case extraction
The intuitive approach to document extraction is to start with a specific need. That’s where we started. A team would need a particular data point, so they built a pipeline to extract it. Another team has a different need, so they build a different pipeline. Over time, multiple teams are extracting data from the same underlying documents. Each had slightly different schema definitions, slightly different normalization logic, and no shared representation of what the results mean.
The result is interpretation drift. Different features looking at the same document produce results that don't reconcile with each other. Even when each individual extraction is technically correct, the outputs may not be consistent across the platform. A field that one pipeline calls "purchase price" might be calculated differently from how another pipeline defines the same concept. This is what happens when extraction is treated as an implementation detail of individual features rather than shared infrastructure.
To fix this problem, we landed on schema-first, exhaustive extraction. Instead of extracting the fields each use case needs, we defined a comprehensive schema for each document type—every field has a name, a type, and a place in a shared data model. We run the extraction once against that schema. Every downstream product, skill, or workflow reads from the same structured output.
The analogy is a data warehouse column. You build it once, correctly, and every team that needs that number queries the same column. You don't have each team write their own ETL job against the same source.
Why it has to live inside the ERP
Exhaustive extraction is necessary but not sufficient. Where the data lives matters as much as how it was extracted.
Standalone extraction tools, products whose job is to read a document and return structured data, solve part of the problem. But they're context-free. A standalone tool doesn't know which LP in your extracted side letter corresponds to which investor record in your fund. It doesn't know which portfolio company the SPA is associated with in your portfolio. It extracts in isolation, which means the resulting data can't power the workflows and experiences that depend on understanding relationships between entities. At Carta, we’ve been building the networked map of private capital through the numbers. Now we get to do it with words.
Carta already holds the context that document extraction needs: the firm, the funds, the investors, the portfolio companies. When extraction happens inside Carta, every extracted field can be linked to known entities, validated against existing records, and surfaced with full context. Unlike standalone datasets someone has to manually reconcile against everything else, discrepancies can be automatically surfaced and reconciled. We call this the self-healing network.
The same logic applies to ad-hoc extraction through AI tools. A fund manager who uploads a document to Claude and asks a question gets an answer in the moment. But the data disappears. It's not stored, not linked to the rest of their data, not queryable by other people on their team, and not available for the next question they ask tomorrow.
In Carta’s Doc Intelligence platform, extraction persists. The same SPA data that answers a co-investor question today can power a financial reporting workflow tomorrow. The same LPA terms that are surfaced in response to a natural language question can also feed into a management fee calculation or a distribution waterfall. The data is stored, structured, and connected to be useful.
Carta Document Intelligence is a data layer that makes every other part of Carta smarter.
Trust through verification, scale through skills
Extracting data and making customers confident enough to act on it are two different problems. We've learned trust is the gating factor. Fund managers who aren't sure whether extracted data is accurate are hesitant to share it with their teams or rely on it in workflows.
Our approach to this is bounding boxes. Every extracted field in Document Intelligence is linked back to the exact location in the source document where it was found. Customers can verify any data point by clicking through to the original text. This changes their thought from "do I trust that the AI got this right?" to “I was able to verify in three seconds.”
Once the data is trusted and structured, you can build skills on top of it. Natural language interfaces that let anyone query the data without knowing SQL, without knowing which document to open, without even knowing the underlying document that might house the answer to their question until they ask. We build and ship skills using this data. Our customers can build on these, and even create brand new ones for their particular use cases.
Which of my LPs have side letter terms that affect my next distribution? Who have I syndicated with most frequently across my portfolio? Which SPAs are missing for a given company?
These questions can be answered in seconds. Rather than reading through hundreds of documents in real time, AI queries structured data because the extraction has already taken place.
From point-in-time lookup to continuous intelligence
The near-term value is in automating workflows that are currently blocked by data living only in PDFs. Side letters carry exceptions that today require manual intervention before a cap call can go out. Transfer document reviews that currently require external legal review for what should be routine checks. When the terms are structured and queryable, the workflows can run.
The longer-term value is something closer to continuous portfolio intelligence. The question we hear most often is “can I stay current across my portfolio without pulling a full report every time something changes?” Flag new financing rounds. Surface follow-on opportunities. Pull a year of valuation history on demand.
None of it works without the foundation: exhaustive extraction, standardized schema, data that lives inside the system where everything else already is. The bet we're making is that having all of this data available, before you know exactly how you'll use it, is more valuable than extracting case by case, later, somewhere else. The skills, the automated workflows, and the portfolio intelligence all depend on getting this layer right first.
DISCLOSURE: This communication is on behalf of eShares, Inc. dba Carta, Inc. ("Carta"). This communication is for informational purposes only, and contains general information only. Carta is not, by means of this communication, rendering accounting, business, financial, investment, legal, tax, or other professional advice or services. This publication is not a substitute for such professional advice or services nor should it be used as a basis for any decision or action that may affect your business or interests. Before making any decision or taking any action that may affect your business or interests, you should consult a qualified professional advisor. This communication is not intended as a recommendation, offer or solicitation for the purchase or sale of any security. Carta does not assume any liability for reliance on the information provided herein. © 2026 Carta. All rights reserved. Reproduction prohibited.
