Private markets don’t have a data problem. They have a truth and accuracy problem.
By Kumar Ujjwal, Founder and CEO
For a decade, every fund-ops team I have met believed the bottleneck was data. So they rebuilt the workflows around it, the cleaning and the aggregating and the endless reshaping into something a person could read. Capital calls, outreach, investor reporting, month-end. All of it redesigned around one assumption: get the data into one place and the rest follows.
That assumption is wrong.
Walk into a fund administrator, a secondaries shop, or an LP back office today and you see the same scene. Folders inside folders. Conference spreadsheets. Half-finished CRM exports. Investor lists passed between four people and edited by all of them. There is more information in the building than ever, and almost none of it can be trusted on sight.
The data is not missing. The data is wrong. Inconsistent, stale, blind to context, and structurally misunderstood. That is the constraint. Not volume. Trust.
Why generative AI breaks the moment it enters private markets
The last two years sold the industry on copilots. Teams plugged them in and waited for clarity. What came back was the oldest problem in a new voice: speed without understanding.
General-purpose models were never built for this. A horizontal model does not know that a sub-fund and an SPV are different animals. It cannot tell you why a CIO outweighs a Partner in one committee and not the next. Backer lineage across three vintages is invisible to it. And it will treat two entities with nearly identical names as the same firm, when they are legal strangers.
You do not fix that with a better prompt. The model was trained to predict language. It was never trained to understand capital flow, ownership structure, or how a decision actually gets made inside a fund. It is a brilliant generalist working in a domain that punishes generalists.
So the output arrives polished and wrong. The wrong title. The wrong backer. A multi-layer fund structure flattened into a label. The prose is clean. The substance collapses underneath it.
Which is why the pattern is always the same. A team tries generative AI once, admires the writing, and is back in spreadsheets by Friday.
The 10-row test
A few weeks ago one of our users ran an experiment. No prompt engineering. No clever workflow. Just a sheet any operations analyst would recognize on sight.
Ten rows. Company names. Partial titles. A few fragments of notes.
He ran it through four of the largest AI tools in the world. Billions of dollars of funding behind them, billions of users in front of them.
All four failed the same way. They invented backers. They misread the CIO. They collapsed multi-entity fund structures into a single "fund." They took a precise operational question and answered it with paragraphs. Worse than wrong, it was confidently wrong, the kind of output a junior analyst pastes into a memo without a second look.
Then he ran the same ten rows through our engine.
The backers were right. CIOs and Partners came back separated. The multi-layer structure rendered the way it actually exists. Geography and strategy landed where they belonged.
He looked at the screen for a second and said something I have now heard more than once, in different words: "Your AI understands how our world works. The others don't."
That sentence is the whole shift.
Truth compounds. Data doesn't.
Once a team sees truth-grade enrichment one time, on ten rows, the argument is over. They stop asking for more data. They start asking for the right picture of reality.
Accuracy compounds in a way volume never has. When the enrichment is right, outreach stops being generic and starts being aimed. A DDQ that used to eat three days gets answered before lunch. Audit prep stops being a fire drill. Reconciliation becomes a check instead of a rebuild, and the capital-call process stops living inside one person's head. Leadership starts trusting the screen in front of them.
Speed without accuracy just manufactures rework. Accuracy is what lets a firm grow without bolting three more people onto every account.
The split nobody is announcing
Every firm is quietly picking one of two roads.
The first road is AI that sounds good and knows nothing. You get fluent summaries and confident sentiment, and you re-check every figure by hand anyway. Progress you can feel and cannot use.
The second road is AI that understands how capital, people, and structures actually fit together. The enrichment survives scrutiny. It feeds strategy. It drops into the workflow without a translation layer.
The industry is not asking for more AI. It is asking for AI that understands the asset class.
Vertical AI is the infrastructure layer
The next decade of private markets will not be won by whoever holds the biggest dataset. It will be won by whoever holds the most accurate picture of reality and can act on it the same hour.
Vertical AI becomes the layer underneath the work itself. Fund administration. Investor relations. Secondaries and co-invest. Compliance, reporting, capital calls, reconciliation, the NAV cycle, the data room the week before an audit. You do not log into it. You notice it as friction that used to be in the work and now isn't.
Private markets have always paid the people who could see clearly before anyone else.
For the first time, the tooling can finally help them do it.