Internal Research Search: How Asset Managers Put Proprietary Data to Work

AI search for internal research lets asset managers query years of broker research, meeting notes, internal models and other proprietary data in one place. The firm ingests its documents into a segregated environment or on-premises, then queries them alongside public filings through a research agent that cites every source. Access controls should mirror information barriers.

Key takeaways

●     Most buy-side firms hold years of paid research that no analyst can search, so knowledge leaves with the people who wrote it.

●     Banning public AI tools stops leaks of confidential material. Analysts still need a governed way to search what the firm knows.

●     Usable internal research search needs private ingestion, page-level citations, permissions that mirror information barriers and point-in-time history.

●     Most active managers should buy the ingestion, coverage and citation layer and keep their investment logic in-house through agents they configure.

Why does a firm's internal research end up unsearchable?

Internal research ends up unsearchable because it was never stored in one place. Broker PDFs, call notes, investment committee memos and models sit across shared drives, inboxes and personal folders with no common index.

An internal research archive is a single searchable store of a firm's own documents and proprietary data, indexed for that firm's use only. Most active managers do not have one. When an analyst leaves, the institutional knowledge behind past calls leaves too, and whoever inherits the name starts again from broker consensus.

This matters more when budgets are tight. MiFID II research unbundling is the EU rule that requires managers to pay for research separately from execution. A Substantive Research survey of the 50 largest asset managers found that for managers above $150bn AUM, average annual US research budgets are $8.6m higher than UK and EU equivalents, and European research spend has recovered only 2 basis points. A firm under that pressure should get full value from research it already paid for.

Boston Consulting Group estimates AI could let asset managers redeploy 50% to 65% of junior-heavy analyst capacity from data gathering and first-pass modeling toward management engagement and differentiated insight. That shift depends on the firm's own material being reachable.

An illustrative example

An analyst picks up coverage of a European mid-cap industrial. Before the interim results, she wants to know what the firm thought about management's margin guidance across the last three cycles. The answers exist in two broker notes, a set of meeting notes and an old IC memo. Finding them by hand takes most of a day.

What happens when analysts work around the problem?

Analysts paste documents into public AI tools, and that creates a confidentiality problem. Shadow AI is the use of AI tools the firm has not approved or governed.

The Verizon 2026 Data Breach Investigations Report found that 45% of employees are now regular users of AI on corporate devices, up from 15% a year earlier. It also found that 3.2% of DLP events involved research and technical documentation uploaded to unauthorized AI systems. Data loss prevention (DLP) is the set of controls that detect and block sensitive data leaving a firm.

Firms are clamping down. Deloitte Australia told partners and staff to stop uploading confidential data into ChatGPT and other public AI platforms, according to the Australian Financial Review. CFOtech reported that a regional investment bank blocked hundreds of attempts to send proprietary models, trading strategies, material non-public information (MNPI) and deal data to public AI tools.

Broker research adds a contractual layer, since it arrives under licensing and confidentiality terms. Stephenson Harwood warns that silence in a non-disclosure agreement (NDA) does not mean AI use is permitted, because uploads are typically processed on third-party systems. Legal 500 commentary advises checking a provider's retention terms, data-use terms and controls before confidential information goes in.

Bans stop the leak. They leave the original need in place, and analysts still have to search what the firm knows.

How can asset managers search internal research and broker notes with AI?

Asset managers can search internal research with AI by ingesting it into a governed internal research archive and querying it through a research agent that cites its sources. Five requirements matter most.

●     Private ingestion. Documents are indexed for the firm's use, held in a segregated environment or on-premise, and kept apart from shared data. Read the provider's data-use and retention terms first.

●     Citations to the source page. An AI research assistant with citations links every figure to its document and page, so a PM can check it before acting.

●     Permissions that mirror information barriers. Information barriers are controls that stop sensitive information passing between teams. Skadden advises firms to consider which employees can access data ingested by internal models and to make sure AI systems respect those barriers.

●     Point-in-time history. Point-in-time data records what was known on a given date, so the analyst sees what was said at the time without hindsight.

●     One query across private and public sources. The firm's view should sit next to what the company actually disclosed.

Check broker licensing terms as well. Some agreements restrict how research may be stored or processed, so not every note will be eligible.

Early results are encouraging. In a vendor case study, StackAI reports that a hedge fund with roughly $8bn to $10bn AUM deployed an assistant over more than a decade of IC memos in under eight weeks, and its analysts reported 65% to 80% less time combing through memos. Adoption is uneven. In a vendor survey by Hebbia, 43% of US finance professionals said their firm regularly uses AI for document search or analysis, while 26% said adoption was limited or inconsistent. Governance gaps help explain the split.

Should an asset manager build or buy AI search for internal research?

Most active managers should buy the ingestion, coverage and citation layer and keep their investment logic in-house. A full build makes sense for a small group of firms.

Judge the options on these criteria:

●     Time to value

●     Parsing quality on complex PDFs with tables and footnotes

●     Citation accuracy at page level

●     Permissioning that maps to desks and barriers

●     Public filings coverage to query against

●     Ongoing engineering cost and audit trail

A-Team Insight argues the in-house case holds only for a narrow cohort with proprietary investment logic, a sustained AI engineering bench and the data architecture to support it. Most firms would spend their budget on parsing, pipelines and filing coverage before an analyst runs a single useful query.

Regulators expect adoption with governance. The FCA says existing rules apply to AI use and cites good practice such as documented senior manager oversight of each use case. Law360 reported that the SEC Division of Investment Management's director told advisers not to wait, saying guidance may be "too limited and too late".

Our position: buy the plumbing and keep the edge by configuring your own agents on top. We set out the wider argument in our view on build versus buy for AI research.

How does Orbit handle internal research and proprietary data?

Orbit handles it through Internal data, which ingests a firm's own documents and treats them like any other source on the platform. Orbit is an award-winning AI investment research platform for institutional buy-side firms, named Best Research Provider/Vendor at the Waters Technology IMD/IRD Awards 2026.

Inside Orbit Insight, Internal data ingests broker research, meeting notes and internal documents privately. It runs in Orbit's environment, or on-premise if the firm prefers. Access is fully customizable, so documents can be restricted by team or desk to match information barriers. Those documents are queried alongside Orbit's Knowledge Base of filings and transcripts: 75,000+ companies and 70M+ documents across 80+ exchanges, with point-in-time history.

Back to the analyst. She asks the Orbit Research Agent what the firm thought about margin guidance across the last three cycles. The answer cites a past broker note, the firm's own meeting notes and the latest interim report, each linked to its source page.

The Orbit Research Agent is the quick entry point. Running the same question across a whole portfolio is a job for an agent built in Agent Builder without code, and that is where the firm's own investment logic lives.

Ingested documents are treated the same as Orbit's own, so they are available through the platform, the API and MCP. MCP (Model Context Protocol) is an open standard that lets AI assistants call external data sources. Teams that prefer their own assistant can connect through Orbit MCP to Claude, ChatGPT, Microsoft Copilot or Gemini and get grounded, cited answers.

Internal usage is unrestricted with no metering, so analysts can query as often as the work needs. For holdings in China, the firm's notes sit next to Orbit's exclusive China A-share coverage of 5,500 names, including earnings calls and research visit records translated into English.

Where should a firm start?

Start with one team and one archive. A sector desk's broker research and call notes from the last five years make a good first set.

Agree permissions and retention with compliance before ingestion, and check each broker's licensing terms. Assign a senior owner for the use case, in line with FCA good practice. Then measure time saved on a real task, such as pre-earnings prep on current holdings.

These are conversations that matter deeply to us at Orbit. Explore Orbit for asset managers, or book a demo to see Internal data working on your own documents.