How Data Ontology Powers Digital and Intelligent Penetration Management in Private Funds

The private‑equity market manages 14.32 trillion CNY across 12,083 managers, with only about 110,000 professionals, meaning each firm handles roughly 10 billion CNY of assets with a team of ten people. Daily operations still rely on WeChat, DingTalk, and Excel spreadsheets, creating a "small‑but‑precise" environment under intense regulatory pressure.

Regulators are tightening disclosure rules: the 2026 CSRC regulation elevates disclosure oversight from self‑regulatory bodies to the commission itself, making full‑project penetration mandatory rather than optional.

To move beyond traditional report‑driven digitization, the team adopted ontology as the foundational semantic layer, integrating knowledge graphs and large‑model technology to build an explainable, inferable disclosure system. Core entities defined include parent funds, sub‑funds, GPs, LPs, and portfolio companies, with attributes such as fund size, duration, industry track, and valuation, and relationships like "invests in", "manages", and "holds position".

Stage 1 – Automating High‑Frequency, Heavy‑Filling Reports : The team targeted the most complex tasks (e.g., insurance‑company regulatory reports, quarterly fund reports). Using a workflow‑orchestrated Agent, data is extracted from multiple source tables, processed by rule engines, and written into target templates. The process remains white‑box and includes manual review nodes for compliance. Result: a single reporting cycle was reduced from two days to two hours.

Stage 2 – Generalized Intelligence for Ad‑Hoc Disclosures : Fixed‑path workflows cannot handle variable input formats and ambiguous semantics of ad‑hoc tasks that require external knowledge (industry classification catalogs, Qichacha API). The new architecture, built on Knora AI 3.0 Trusted Agent Engine, adds two layers: precise context engineering that translates enterprise data, business logic, and program instructions into semantic inputs for large models; and autonomous dynamic planning where agents infer optimal decisions, invoke generic tools (e.g., Excel parsing) and domain‑specific tools (e.g., fund agreement lookup), and continuously learn from decision data. Crucially, each automated field is traced back to its data source, preserving auditability.

To date, about ten agents have been deployed, covering the full data‑acquisition, analysis, and output chain. Co‑locating IT functions within the operations department compressed the demand‑to‑delivery cycle to under a week, eliminating cross‑departmental friction.

The initiative demonstrates that AI can reshape niche‑industry software markets: while the private‑fund digitalization market is only tens of millions, large‑model‑driven agents enable scalable, personalized solutions with dramatically lower marginal costs.

Key takeaways: (1) Data governance must be re‑examined in the LLM era, with ontology bridging data assets to knowledge assets; (2) In highly regulated sectors, the primary metric is "explainable automation rate"—every autonomous decision must be traceable to explicit business rules and data origins.

Q&A : The team initially tried traditional knowledge bases (stacked Markdown) and predefined field queries, both of which failed to capture nuanced financial terms like IRR or Net IRR. Switching to ontology allowed explicit encoding of entities, attributes, and relationships, enabling machine‑readable reasoning and flexible, domain‑specific knowledge reuse.