Should you build or buy Investment Research Management?
Investment research management software gives buy-side analysts and portfolio managers a single system for capturing notes, models, and meeting records against securities and investment theses, with permissions, retention, and an audit trail that satisfies compliance recordkeeping rules. Hedge funds, asset managers, and consultants buy it to replace scattered notes, spreadsheets, and email threads.
Buy Bipsync, FactSet IRN or VerityRMS as the research system of record, then build your own AI retrieval and synthesis layer on top of whatever integration options that vendor supports. Treat SharePoint, Notion or a custom RAG app as a capture surface, not as the compliant repository, unless you deliberately engineer the retention and audit controls.
Build it, buy it, or bridge?
When building makes sense
There's a real build case at the synthesis layer, not the record layer. Balyasny and Bridgewater both run internally built AI research systems in production, described in Balyasny's case with OpenAI as an agentic platform and in Bridgewater's own public description of its Pocket Analyst Tool, and those systems genuinely handle search and synthesis over firm research. What neither replaces is the compliant record itself: durable capture, versioning, retention, entitlements, and the audit trail that a regulated adviser or broker-dealer needs under recordkeeping rules. No fund has documented replacing that layer with a self-built system. If you have engineering capacity and want to build, point it at retrieval and synthesis on top of whatever platform holds the record, the way Balyasny and Bridgewater have, rather than at the record-keeping system itself.
When buying makes sense
Buying the system of record is the right call for nearly every buy-side firm, because Bipsync, VerityRMS, and FactSet IRN already cover the note capture, tagging, permissions, and audit trail that a research process needs daily, and analysts use that core constantly rather than leaving it idle. The taxonomy and workflow are configurable enough to reflect a firm's own process without the firm needing to build the underlying application. Where the buy case gets easier is that vendors are now shipping their own AI layers directly into the record: VerityRMS added AI chat with cited answers and an MCP server, and FactSet IRN ships drafting and topic assistants. That narrows the reason to build a parallel synthesis system from scratch, though a firm with real engineering capacity can still extend either platform's API for its own workflow.
The desk read
Build-versus-buy analysis for Investment Research Management is being written. In the meantime, the framework that drives every B4 call is on the B4 Index page.
Frequently asked
If Bridgewater and Balyasny built their own research AI, why should a smaller fund buy an RMS?
Because what they built is a synthesis layer on top of a governed research record, not a replacement for one. Neither fund has documented running its own compliant capture, retention, and audit system, which is the part that costs the most to get wrong and the part vendors already cover.
Can we just use Notion or SharePoint instead of a dedicated RMS?
Only as a capture surface, not as the compliant repository, unless you deliberately engineer retention, permissions, and audit controls to match what a regulated adviser needs under recordkeeping rules. One documented fund runs investment decisions on Notion, but that's a capture layer choice, not a system-of-record replacement.
What does an RMS actually cost?
Pricing across Bipsync, VerityRMS, and FactSet IRN is quote-based rather than published, and tracks seat count and integration depth rather than a flat per-firm fee. Expect the conversation to start with your headcount and desired integrations, not a list price.
Does buying an RMS lock us out of building our own AI tools?
No. Most RMS platforms, including VerityRMS with its MCP server, are built to let a firm connect its own retrieval or synthesis tools on top of the compliant record, rather than requiring everything to run inside the vendor's own AI features.