The workflow before
Bregal Sagemount's analysts researched each prospect by hand: website, news, reports, funding history. A first-pass profile took 4 to 6 hours, so the team covered 20 to 30 companies a week and about 60% of its target sectors. Meeting notes, CIM summaries and outreach emails were all written from scratch.
What we built
We built a deal sourcing system on Databricks that monitors company websites, news, industry reports and internal documents, and uses LLMs to profile each company, score it against the firm's investment criteria and route the best fits to the deal team with a research package attached.
Documents are chunked, embedded and stored in Chroma, so answers are grounded in retrieved sources with citations. Models served through MosaicML include ones fine-tuned on the firm's past memos and deal evaluations.
The same system drafts founder outreach, summarizes calls and CIMs, and produces first drafts of IC memos, with results written back to Salesforce. Investment professionals review and decide; the system does the first pass.
Before and after
| Measure | Before | After |
|---|---|---|
| Companies researched per week | 20 to 30 | 200+ |
| First-pass research per company | 4 to 6 hours | Under 30 minutes |
| Target market coverage | About 60% | 90%+ |
What's next
The next step is L4: connecting sourcing, diligence prep and portfolio monitoring agents so a signal on a portfolio company or competitor flows into the deal pipeline without a hand-off.
