Private markets compliance professionals know the drill: an investor asks whether they have co-investment rights in Fund III, and the answer is locked inside a 200-page side letter compendium. For years, someone had to open a spreadsheet, run Ctrl+F, and hope the search surfaced the right clause. In recent months, firms have begun trying to find information with the help of generic LLMs. The results may be faster, but accuracy and completeness are issues more often than not.
Ontra continues to develop new ways for firms to improve their obligation management processes within the Insight for Funds solution. In fall 2024, Ontra initially introduced AI Search. Since then, advances in AI have enabled a significant upgrade — from a standalone search feature to a comprehensive, conversational AI assistant that works across a firm’s entire fund and obligation documentation.
Search scope expands beyond obligations
When AI Search first launched, it queried the structured obligations that Ontra’s AI had already extracted and categorized from side letters and LPAs. That alone was a major step forward: users could ask a question in natural language and receive a direct, sourced answer, rather than scrolling through a spreadsheet.
Now, AI Search queries the full text of every document in a firm’s Insight for Funds repository. Users can surface critical information that may not have been specifically categorized as an obligation: reporting timelines buried in LPA language, notice provisions, operational preferences, or fee calculation methodologies that sit outside the standard obligation taxonomy.
The practical impact is significant. Instead of knowing exactly which category to look in, anyone can simply ask a question and trust that AI Search will review all of the relevant documents.
Documents digitized for the future
AI Search’s expanded capabilities are possible because of the rigorous document preparation that happens during implementation — long before a user ever types a question.
When a document is uploaded to Insight, it’s broken into small, overlapping chunks of text. Think of it as creating a detailed index where every paragraph has its own entry. Each chunk is then processed in two ways. First, it is converted into a mathematical representation — called an embedding — that captures the meaning of the text, not just the words it contains.
Second, the system extracts structured attributes — fund, document type, and category — and uses them to automatically narrow the search scope before retrieval begins. Where available, additional context, like counterparty and effective date, helps return the most relevant answers.
The result is a foundation that supports fast, accurate, context-aware retrieval at scale, whether the firm manages one fund or five.
A redesigned, conversational approach
AI Search is now a conversational AI assistant embedded across Insight for Funds, designed to fit the way legal and compliance teams actually work.
Users can:
- Ask follow-up questions: Users can refine their inquiry within the same thread — asking “Which investors have this right?” and then immediately following up with “Are any of those investors in Fund IV as well?” The experience mirrors working with a research partner who retains context from one question to the next.
- Saved chat history: AI Search saves conversation threads so users can return to a prior line of inquiry days or weeks later. Team members can also reference shared research, building institutional knowledge that doesn’t disappear when someone closes a browser tab.
- Share and collaborate: Answers can be formatted and copied directly into emails, memos, or board-ready reports. Users can also create task workflows from AI Search findings — turning a research result into an assigned, trackable compliance action without leaving the platform.
Semantic search that understands intent
Under the hood, AI Search runs two types of search simultaneously and merges the results to ensure users quickly surface the information they need.
- Keyword search matches exact words and phrases — the traditional approach that works well when the user knows the precise language used in a document.
- Semantic search compares the meaning of the user’s question to the meaning of each document chunk, surfacing relevant results even when the exact terminology differs.
For example, a query about “what I can’t invest in” surfaces clauses about “investment restrictions” — even though the words don’t overlap. A search for “key man” returns key person provisions. The system handles typos gracefully, too: “Most favred nations” still returns the right MFN provisions.
The two result sets are mathematically merged, with the ranking algorithm prioritizing results that score highly in both keyword and semantic search — a strong signal of genuine relevance. The effect is a search experience that is both precise and forgiving, meeting users where they are, regardless of whether they use the exact contractual language.
What to expect: reliability and verification
In obligation management, almost right is wrong. A partially accurate answer about an investor’s reporting rights can lead to a missed deliverable or an audit finding. AI Search is built around the principle that every answer must be verifiable.
Quality scoring
Every answer AI Search generates is evaluated by a separate AI judge before reaching users’ dashboards. The evaluation covers four dimensions: correctness, completeness, groundedness (whether the answer is supported by the source text, with no fabricated information), and formatting. This scoring runs automatically on every query, creating a continuous quality signal.
Grounded answers
The system has undergone significant improvements in accuracy through rigorous internal evaluations. Each iteration is measured against real-world queries from compliance professionals, ensuring the model improves on the questions that matter most.
Inline citations
Answers are strictly grounded in retrieved document chunks. Users can click inline citations to jump directly to the source text in the original document, whether a side letter, LPA, or MFN form. There is no need to take the AI’s word for it; verification is one click away.
Smarter, faster obligation management
AI Search in Insight for Funds represents a meaningful upgrade for how private markets firms manage investor obligations. An expanded document scope means answers aren’t limited to pre-categorized obligations. A conversational interface lets teams research iteratively and share findings. Dual keyword and semantic searches ensure that the right information surfaces regardless of how a question is phrased. And automated quality scoring with inline citations gives legal and compliance teams the confidence to act on the AI’s output.
The result: faster, more reliable answers to the questions compliance teams face every day — without the spreadsheet archaeology.
See AI Search in action — walk through the interactive demo.
